DESC:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
SYSTEMS AND METHODS TO PROVIDE AN INTEGRATED SUPPLY CHAIN NETWORK FOR OPTIMIZATION OF OXYGEN DISTRIBUTION

Applicant:
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th Floor,
Nariman Point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the invention and the manner in which it is to be performed.
CROSS-REFERENCE TO RELATED APPLICATIONS AND PRIORITY
[001] The present application claims priority from Indian provisional patent application no. 202121027069 filed on June 17, 2021. The entire contents of the aforementioned application are incorporated herein by reference.

TECHNICAL FIELD
The disclosure herein generally relates to an integrated supply chain network, and, more particularly, to systems and methods to provide an integrated supply chain network for optimization of oxygen distribution.

BACKGROUND
In a pandemic scenario, with abrupt increase in the number of patients, demand for medical oxygen increases to a high level. However, daily production capacity and availability of only 10-15 percent oxygen for medical use leads to scarcity of supply in hospitals, thereby meeting the demand of the medical oxygen becomes challenging. Insufficient quota allocation for worst hit places without proper monitoring, visibility and control on local supplier leads to black marketing, higher pricing and increased fatalities. Further, lack of infrastructure such as limited cryogenic tankers, low capacity to store liquid oxygen limited number of high-capacity gas cylinders, difficulties in On-time delivery without required number of vehicles lead to longer delivery times. Thus, with variation in demand at different levels, such as centre level, state level, hospital facility level, prioritization of the supply to meet demand at a critical level becomes necessary. Conventional systems and methods fail in providing efficient, effective and optimum distribution of the medical oxygen.

SUMMARY
Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a processor implemented method is provided. The method comprising: receiving, via one or more hardware processors, a plurality of input datasets pertaining to a demand based oxygen supply chain network from one or more data sources; applying, via the one or more hardware processors, a similarity based weighted distance technique on the plurality of input datasets to obtain a plurality of weighted similarity scores; imputing, via the one or more hardware processors, one or more missing values in the plurality of input datasets based on a maximum value of weighted similarity score from the plurality of weighted similarity scores to obtain a plurality of balanced input datasets; predicting, via the one or more hardware processors, location wise oxygen demand using one or more machine learning models on the plurality of balanced input datasets, wherein the oxygen demand at any location is calculated by considering oxygen demand of new and existing users; estimating, via the one or more hardware processors, an overall oxygen demand based on prediction of the location wise oxygen demand of the new and the existing patient at each of a plurality of demand nodes to be visualized on a Graphic interface system (GIS) platform; generating, via the one or more hardware processors, a plurality of priority scores using an analytical hierarchy process (AHP) model to prioritize the plurality of demand nodes visualized on the Graphic interface system (GIS) platform for oxygen supply; allocating, via the one or more hardware processors, an optimal oxygen supply to the plurality of demand nodes using a mixed integer programming (MIP) based optimization model for a critical demand and a normal demand determined based on the plurality of priority scores; and
executing, via the one or more hardware processors, an optimized dynamic routing and re-routing of a plurality of vehicles based on an adaptive parallel vehicle based genetic algorithm for optimized distribution of the optimal oxygen supply to the plurality of demand nodes, wherein the adaptive parallel vehicle based genetic algorithm optimizes the dynamic routing of the plurality of vehicles based on minimum delivery time and minimum congestion.
In another aspect, a system to provide an integrated supply chain network for optimization of oxygen distribution is provided. The system comprising a memory storing instructions; one or more communication interfaces; and one or more hardware processors coupled to the memory via the one or more communication interfaces, wherein the one or more hardware processors are configured by the instructions to receive, a plurality of input datasets pertaining to a demand based oxygen supply chain network from one or more data sources; apply, a similarity based weighted distance technique on the plurality of input datasets to obtain a plurality of weighted similarity scores; impute, one or more missing values in the plurality of input datasets based on a maximum value of weighted similarity score from the plurality of weighted similarity scores to obtain a plurality of balanced input datasets; predict, location wise oxygen demand using one or more machine learning models on the plurality of balanced input datasets, wherein the oxygen demand at any location is calculated by considering oxygen demand of new and existing users; estimate, an overall oxygen demand based on prediction of the location wise oxygen demand of the new and the existing patient at each of a plurality of demand nodes to be visualized on a Graphic interface system (GIS) platform; generate, a plurality of priority scores using an analytical hierarchy process (AHP) model to prioritize the plurality of demand nodes visualized on the Graphic interface system (GIS) platform for oxygen supply; allocate, an optimal oxygen supply to the plurality of demand nodes using a mixed integer programming (MIP) based optimization model for a critical demand and a normal demand determined based on the plurality of priority scores; and execute, an optimized dynamic routing and re-routing of a plurality of vehicles based on an adaptive parallel vehicle based genetic algorithm for optimized distribution of the optimal oxygen supply to the plurality of demand nodes, wherein the adaptive parallel vehicle based genetic algorithm optimizes the dynamic routing of the plurality of vehicles based on minimum delivery time and minimum congestion.
In yet another aspect, a non-transitory computer readable medium is provided. The non-transitory computer readable medium comprising receiving, via one or more hardware processors, a plurality of input datasets pertaining to a demand based oxygen supply chain network from one or more data sources; applying, via the one or more hardware processors, a similarity based weighted distance technique on the plurality of input datasets to obtain a plurality of weighted similarity scores; imputing, via the one or more hardware processors, one or more missing values in the plurality of input datasets based on a maximum value of weighted similarity score from the plurality of weighted similarity scores to obtain a plurality of balanced input datasets; predicting, via the one or more hardware processors, location wise oxygen demand using one or more machine learning models on the plurality of balanced input datasets, wherein the oxygen demand at any location is calculated by considering oxygen demand of new and existing users; estimating, via the one or more hardware processors, an overall oxygen demand based on prediction of the location wise oxygen demand of the new and the existing patient at each of a plurality of demand nodes to be visualized on a Graphic interface system (GIS) platform; generating, via the one or more hardware processors, a plurality of priority scores using an analytical hierarchy process (AHP) model to prioritize the plurality of demand nodes visualized on the Graphic interface system (GIS) platform for oxygen supply; allocating, via the one or more hardware processors, an optimal oxygen supply to the plurality of demand nodes using a mixed integer programming (MIP) based optimization model for a critical demand and a normal demand determined based on the plurality of priority scores; and executing, via the one or more hardware processors, an optimized dynamic routing and re-routing of a plurality of vehicles based on an adaptive parallel vehicle based genetic algorithm for optimized distribution of the optimal oxygen supply to the plurality of demand nodes, wherein the adaptive parallel vehicle based genetic algorithm optimizes the dynamic routing of the plurality of vehicles based on minimum delivery time and minimum congestion.
In an embodiment, the similarity based weighted distance technique utilizes a Levenshtein edit distance for string type data and Euclidean distance for numerical data in the plurality of input datasets.
In an embodiment, the adaptive parallel vehicle based genetic algorithm to perform the dynamic route planning and re-routing of the plurality of vehicles further comprising: executing, in a simultaneous manner, a genetic algorithm (GA) for each of the plurality of vehicles; and adaptively updating, in real time, an information pertaining to one or more vehicle constraints and availability of one or more resources after convergence of the genetic algorithm of each of the plurality of vehicles till a final route for each of the plurality of vehicles is obtained.
In an embodiment, a status of the plurality of demand nodes and the plurality of vehicles is monitored in real time using the GIS platform and a global positioning system (GPS).
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
FIG. 1 illustrates an exemplary system to provide an integrated supply chain network for optimization of oxygen distribution according to some embodiments of the present disclosure.
FIG. 2 illustrates an exemplary flow diagram of a method to provide an integrated supply chain network for optimization of oxygen distribution using the system of FIG. 1, in accordance with an embodiment of the present disclosure.
FIG. 3 provides an overview of the integrated supply chain network for optimization of oxygen distribution, in accordance with some embodiments of the present disclosure.
FIG. 4 illustrates a block diagram for performing missing data imputation on a plurality of datasets to provide the integrated supply chain network for optimization of oxygen distribution, according to some embodiments of the present disclosure.
FIGS. 5A and 5B are block diagrams illustrating process of predicting location wise oxygen demand of a new and an existing patient respectively, in accordance with some embodiments of the present disclosure.
FIG. 6 illustrates the process of demand node prioritization, in accordance with some embodiments of the present disclosure.
FIG. 7 is a block diagram illustrating process of planning oxygen supply allocation to the plurality of demand nodes, in accordance with some embodiments of the present disclosure.
FIG. 8 is an interconnected graph showing an example of oxygen supply allocation to the plurality of demand nodes, in accordance with some embodiments of the present disclosure.
FIG. 9 is a flow diagram illustrating working of adaptive parallel vehicle based genetic algorithm (APVGA), in accordance with some embodiments of the present disclosure.
FIG. 10 is a flow diagram illustrating working of genetic algorithm for one vehicle routing, in accordance with some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments. It is intended that the following detailed description be considered as exemplary only, with the true scope being indicated by the following embodiments described herein.
The embodiments herein provide a system and method to provide an integrated supply chain network for optimization of oxygen distribution. In Covid pandemic scenario, considering average Covid-19 positivity of patients, a daily production capacity of 7127 MT and an approximate demand of ~6600 MT is observed. Further, total number of available cryogenic Tanks is 1171 with capacity ranging from 2 MT to 30 MT. Further, predicting patient positivity and hence likely demand for oxygen cylinder is complex. Therefore, matching supply to demand in a state having PSA plants and in-house production and supporting outside the state such as center to state allocation from liquid oxygen manufacturing plants and import is an intricate problem. The proposed system and/or method provides prediction and optimization algorithms to help with efficient distribution of oxygen cylinders from manufacturing plant to facilities within state and across a country. The method of the present disclosure utilizes a plurality of data sources as input into the system for training a plurality of models. First, a plurality of datasets comprising patient data, test data, location data are received and analyzed to identify missing values. The missing values are imputed based on similarity scores using machine learning models. After successful imputation, oxygen demand at lowest level is predicted using testing data and patient profile. Further, the demand up to state level is aggregated and available oxygen in stock at each level such as hospital, region, district, state, and/or the like is calculated. A plurality of demand locations including both hospitals and local supplier are further prioritized based on a plurality of criteria and a plurality of sub-criteria which further aggregates priority scores up to state level. Supply of oxygen is allocated to a plurality of demand nodes (e.g., center to state) based on (i) critical and normal demand, (ii) priority scores of the plurality of demand locations and (iii) available oxygen supply. Finally, a route for each vehicle from a plurality of vehicles is planned in optimal manner using travel distance, time to delivery, cylinder availability and congestion data for hassle free distribution of oxygen cylinders to the plurality of demand nodes. The method and system of the present disclosure also provides an option of re-routing in case of any sudden surge of oxygen demand. For re-routing, a secondary data (e.g., containment zone, positivity rate, and the like) along with pandemic (e.g., Covid-19) test positivity data is utilized to predict the demand with efficacy. The method and system of the present disclosure provides an overall picture of demand and supply at national level and simultaneously, have a view of demand, supply and allocation at the lowest level also. The method and system of the present disclosure give a real-time monitoring of demand nodes, oxygen inventory and vehicle positioning. Therefore, it ensures the effective and efficient oxygen distribution planning across the country
Referring now to the drawings, and more particularly to FIG. 1 through FIG. 10, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
FIG. 1 illustrates an exemplary system 100 (hereinafter referred as system 100 throughout the description) to provide an integrated supply chain network for optimization of oxygen distribution, in accordance with an embodiment of the present disclosure.
In an embodiment, the system 100 includes processor (s) 104, communication interface device(s), alternatively referred as or input/output (I/O) interface(s) 106, and one or more data storage devices or memory 102 operatively coupled to the processor (s) 104. The processor(s) alternatively referred as one or more processors 104 may be one or more software processing modules and/or hardware processors. In an embodiment, the hardware processors can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals/inputs based on operational instructions. Among other capabilities, the processor(s) is configured to fetch and execute computer-readable instructions stored in the memory. In an embodiment, the system 100 can be implemented in a variety of computing systems, such as laptop computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud and the like.
The I/O interface(s) 106 can include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks N/W and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the I/O interface device(s) can include one or more ports for connecting a number of devices to one another or to another server. The I/O interface 106, through the ports can be configured to receive inputs stored external to the system 100.
The memory 102 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. Further, the memory 102 can include a plurality of datasets, a plurality of models that can be implemented by the processor 104 to perform actions according to the embodiments of the present disclosure. In an embodiment, the memory 102 includes a data repository 108 for storing data processed, received, and generated as output(s) by the system 100. The plurality of models stored in the memory 102 may include routines, programs, objects, components, data structures, and so on, which perform particular tasks or implement particular (abstract) data types.
The data repository 108, amongst other things, includes a system database and other data. In an embodiment, the data repository 108 may be external to the system 100 and accessed through the I/O interfaces 106. The memory 102 may further comprise information pertaining to input(s)/output(s) of each step performed by the processor 104 of the system 100 and methods of the present disclosure. In an embodiment, the system database may store information not limited to, the plurality of received datasets and the like. Further, the system database stores information pertaining to inputs fed to the system 100 and/or outputs generated by the system (e.g., at each stage), specific to the methodology described herein. More specifically, the system database stores information being processed at each step of the proposed methodology. The other data may include, data generated as a result of the execution of the plurality of models stored in the memory 102. The generated data may be further learnt to provide improved learning in the next iterations to output desired results with improved accuracy.
In an embodiment, the one or more hardware processors 104 can be configured to provide an integrated supply chain network for optimization of oxygen distribution, which can be carried out by using methodology, described in conjunction with FIG. 2, and use case examples.
FIG. 2 illustrate an exemplary flow diagram of a method to provide an integrated supply chain network for optimization of oxygen distribution using the system of FIG. 1, in accordance with an embodiment of the present disclosure. In an embodiment, the system 100 comprises one or more data storage devices or the memory 102 operatively coupled to the one or more processors 104 and is configured to store instructions for execution of steps of the method 200 by the one or more processors 104. The steps of the method 200 of the present disclosure will now be explained with reference to the components or blocks of the system 100 as depicted in FIG. 1 and the steps of flow diagram as depicted in FIG. 2. Although process steps, method steps, techniques or the like may be described in a sequential order, such processes, methods and techniques may be configured to work in alternate orders. In other words, any sequence or order of steps that may be described does not necessarily indicate a requirement that the steps to be performed in that order. The steps of processes described herein may be performed in any order practical. Further, some steps may be performed simultaneously.
Referring to the steps of the method 200 depicted in FIG. 2, in an embodiment of the present disclosure, at step 202, the one or more hardware processors 104 are configured to receive, a plurality of input datasets pertaining to a demand based oxygen supply chain network from one or more data sources. In an embodiment, the oxygen demand is applicable but not restricted to a pandemic scenario. For example, in light of the present scenario, the pandemic assumed in the present disclosure could be but not limited to Covid-19. Further, the plurality of input datasets may comprise but not limited to (i) Demand Supply Geospatial Data including oxygen demand data, Daily/ weekly oxygen demand for states/ districts/ distributors/ hospitals/local suppliers, oxygen manufacturing data such as manufacturing site locations, Capacity/planned production Manufacturer (e.g., Liquid oxygen manufacturer, augmented manufacturer and PSA), locational Data of manufacturer, demand and distribution Centre, oxygen import data, oxygen inventory such as on hand (e.g., hospital/ district/ distributor/ local suppliers) and in-transit, oxygen inventory held by state, oxygen predefined allocation at State/ district/ distributor/ hospital level, and geospatial database, (ii) pandemic related data including positivity rate, per day testing data, patient specific test result data (iii) facility related data in real-time such as available beds, positivity rate, type of facility used by patient, Daily/ weekly oxygen demand, capacity of ICU/HDU/ ICU with ventilator bed and /or the like (iv) secondary research data including stage of pandemic, infection prediction, and number of patients for ICU/HDU/ ICU with ventilator bed, and (v) logistics data including Distributors/ local suppliers/ hospitals’ location, logistic lead-time oxygen logistics data, road network, road attributes, route data, geo-coded data from address data, and geo-tagging of associates non-spatial data.
The plurality of received input datasets may comprise many instances where a particular element is absent because of various reasons. Using such incorrect or incomplete data for prediction may affect quality and accuracy of prediction. Hence the input data needs to be corrected for better prediction. By doing missing data analysis, important missing data elements are identified which are important for prediction and imputed from previous patient data. Conventional prediction algorithms cannot be directly applied to incomplete data (i.e. data with missing values). State of the art methods deal with missing data by data reduction which deletes instances with missing values. However, this leads to great loss of information, and lead to wrong predictions. To improve prediction and analytics accuracy, data imputation is needed to replace missing values. Thus, as depicted in step 204 of FIG. 2, the one or more hardware processors are configured to apply a similarity based weighted distance technique on the plurality of input datasets to obtain a plurality of weighted similarity scores. In an embodiment, the similarity based weighted distance technique utilizes a Levenshtein edit distance for string type data and Euclidean distance for numerical data in the plurality of input datasets. Further, as depicted in FIG. 2, at step 206, the one or more hardware processors are configured to impute one or more missing values in the plurality of input datasets based on a maximum value of weighted similarity score from the plurality of weighted similarity scores to obtain a plurality of balanced input datasets. As depicted in FIG. 4, for missing data imputation, random forest model is executed on a plurality of cleaned dataset without missing values to identify a plurality of important variables, wherein each of the plurality of important variables is used as weightage for scoring and depicted as w_i weightage for variable i. The importance of variables is normalized so that sum of importance is 1. This describes how much single variable is important for prediction. Further, data with missing values for the important variables is identified. It is expected that all data sets should have values for important variables. If any data variable is missing with value, then this need to be replaced by value of most similar patient record. Furthermore, distances are calculated for each patient data for which values of important variables are available using Levenshtein edit distance method if the important variables include string data type. Euclidean distance method is used for numerical variables. Here, the calculated distance for variable i and data j is depicted as d_ij. Upon calculating distance, a weighted score for each patient data is calculated using equation (1) provided below as:
?ws?_j= w_i* (1/d_ij ) (1)

Here, ?ws?_j denotes weighted score for data j. Further, the patient data is ranked based on weighting score and variable value with highest ranking score is selected for imputation.
As depicted in FIG. 3, there are three variations in the supply chain network proposed in the present disclosure including large scale oxygen manufacturing, medium scale oxygen manufacturing, and captive plant. In the large scale oxygen manufacturing, liquid oxygen storage tank is transported from cryogenic oxygen plant that moves in cryogenic tanker to the distributor, then local supplier and finally to the hospital. In the medium scale oxygen manufacturing, zeolite tubes that works as absorber of PSA oxygen generator is supplied to the PSA oxygen Plant and manufactured medical oxygen is transported to local suppliers as Gaseous O2 and finally supplied to hospitals. In the captive plant, zeolite tubes that works as absorber is supplied in the PSA oxygen generator, which is then built as PSA O2 plant that finally gets transported to the hospitals.
Referring back to FIG. 2, at step 206, the one or more hardware processors 104 are configured to predict location wise oxygen demand using one or more machine learning models on the plurality of balanced input datasets, wherein the oxygen demand at any location is calculated by considering oxygen demand of new and existing users. For example, the new user may refer to new patients, data of whom is not saved in the system database. However, the existing user may refer to existing patients, data of whom is saved in the system database. In an embodiment, FIGS. 5A is a block diagram illustrating process of predicting location wise oxygen demand of a new a patient, in accordance with some embodiments of the present disclosure. As depicted in FIG. 5A, major data requirements for such prediction include recent history (1 month) of testing data (e.g., CT value, positive/ negative report) and person profile like age, gender, CRP value, CT value etc., location data, published secondary data like positivity rate, containment zone etc., and action required like shift to HDU/ ICU/ ICU with ventilator. Based on the above data, the one or more machine learning models are developed with selected features using train data (e.g., 80% of whole data set) and tested using test data (e.g., 20% of data). The validated machine learning models are used for scoring of new patient profile (for example, patients who comes for Covid-19 testing), and predict further action required (for example, shift to home isolation/HDU/ ICU/ ICU with ventilator). On the basis of action required, the requirement of oxygen for new patient is calculated. In other words, for a new patient profile, multilevel prediction is required which predicts what type of bed will be chosen for a given profile. For this prediction, target variable is type of bed chosen whereas independent variables are testing data, various test reports and their value, age, gender, marital status, location like containment zone and other data like positivity rate, no. of active cases etc. Data gets prepared for prediction modelling. All rows without availability of values for important variables are removed and some feature are selected using a random forest algorithm. After data preparation, data is separated into training and testing data (80: 20 ratio). The one or more machine learning models are trained using 80% data and tested using 20% data. Further, some hyperparameters including but not limited to ‘n_estimators’, ‘max_features’, and ‘min_sample_leaf’ are tuned to improve speed of the one or more machine learning models. Here, use of the random forest algorithm for model development is described through pseudo-code below:
Random Forest – Model Development
Select randomly “k” features out of total “m” features
Calculate a node “n” using greatest break point for “k” features
Again, node split into progeny nodes using greatest break point
Repeat above steps
While n<= X (maximum number of nodes)
Repeat above steps r times
If required no. of trees = r
Further, use of the random forest algorithm for testing and prediction is described through pseudo-code below:
Random Forest – Testing and Prediction
Predict the outcome (using the test features and the rules of each decision tree)
Estimate the votes for each predicted target
final prediction = high voted prediction
Convert it to labels (binary or multiple labels)
From the above prediction, the new patients and type of bed allocated (to-be) can be predicted at the time of testing. This can be included in the demand of oxygen for next day. After patient being admitted to the hospital, the oxygen demand is calculated in terms of length of stay.
FIG. 5B is a block diagrams illustrating process of predicting location wise oxygen demand of an existing patient, in accordance with some embodiments of the present disclosure. In a similar way, as depicted in FIG. 5B, the oxygen demand prediction for existing patients is dependent on their Length of Stay (LoS) in hospital/ home isolation. Major data requirement for such predictions are include recent history (1 month) of patients’ profile like age, gender, CRP value, CT value etc., previous medical history/ comorbidities like diabetic, allergic etc., type of facility used like Home Isolation with oxygen/ HDU/ ICU/ ICU with ventilator along with required oxygen flow for each type of bed like the requirement is 10 Ltr/hr in HDU and 30 Ltr/hr on ventilator. The length of stay predicts number of days patient will be in hospital and type of bed he/she might require during stay, where the type of bed required may or may not change from ICU to HDU to oxygen bed. The Length of stay prediction helps to understand oxygen requirement during hospital stay. The goal is to create a model that predicts the length-of-stay for each patient at time of admission and subsequently for each day. To predict hospital LoS, data needs to be separated into terms of a dependent target variable (LoS in this case) and independent variables (features) to be used as inputs to the machine learning model. The LoS is a continuous variable (measured in days), hence a regression model is used for prediction. Each patient’s condition is used to estimate the oxygen requirement. It is expected to predict LoS accurately than normal regression models. Commonly used evaluation metric includes average, or mean LoS. To measure the performance of machine learning model, prediction model is compared against median and average LoS using root-mean-square error (RMSE) provided in equation (2) below as.
RMSE= v((?_(i=1)^n¦?((y_i ) ^-y_i)?^2 )/N) (2)

Further, LoS with lower RMSE is predicted. In an embodiment, after reviewing major hospital data and patient data, major datasets are identified as hospital admission data, patient details data, testing data, diagnosis data, patient bed occupancy data. Admission data gives information about patient admission and discharge date. Patient details as age, sex, weight, insurance, blood pressure, diabetic condition. Patient bed occupancy days for each bed type is calculated separately. In an embodiment, many of the variables under study are categorical and required to be classified into broad categories. This includes vital parameters as blood pressure, temperature, pulse, respiration, oximetry, BMI etc. The hospital bed data is classified into ICU, HDU, ventilator with oxygen bed. Based on the above data, the machine learning model is developed with selected features using train data (80% of whole data set) and tested using test data (20% of data). Using training dataset, different regression models are executed and compared on testing dataset to select final prediction model. In an embodiment, Random forest regressor, Gradient boosting regressor are used for predicting LoS. The validated model is used for the scoring of new patient profile and predict the length of stay in ICU/ home isolation like next 5 day in ICU after that 5 days in HDU. On the basis of length of stay and oxygen flow requirement, the daily requirement of oxygen for existing patient is calculated by following formula provided below as:
Oxygen Demand=Length of Stay (in days)×Required Oxygen Flow (ltr/day)
The total daily demand of oxygen is the sum of oxygen demand for new patient and daily requirement of existing patient. The exceptional demand is not considered here as it is not regular event and it is handled during distribution by fair share allocation. Finally, the oxygen demand for the existing patients is calculated at the time of admission. Further, as depicted in step 210 of FIG. 2, the one or more hardware processors 104 are configured to estimate an overall oxygen demand based on prediction of the location wise oxygen demand of the new and the existing patient at each of a plurality of demand nodes to be visualized on a Graphic interface system (GIS) platform. For estimating total demand, sum of oxygen demand for new patients and oxygen demand for existing patients is computed. Here, the plurality of demand node may comprise but not limited to hospitals, isolation centers and the like.
Furthermore, as depicted in step 212, after getting demand from the hospitals, isolation centers and local suppliers, a plurality of priority scores using an analytical hierarchy process (AHP) model are generated to prioritize the plurality of demand nodes visualized on the Graphic interface system (GIS) platform for oxygen supply. FIG. 6 illustrates the process of demand node prioritization, in accordance with some embodiments of the present disclosure. In an embodiment, the plurality of demand nodes indicative of demand locations is prioritized using a multi criteria decision making (MCDM) approach, wherein the MCDM approach provides a ranking to all of the plurality of demand nodes. As shown in FIG. 6, the ranking is done based on three criteria namely supply, facility and geographical region. In an embodiment, defined criteria are further divided into nine sub-criteria including days of supply, storage capacity, in-house production, type of facility, occupied oxygen beds, available oxygen beds, location type, predicted demand and positivity rate. Further, weightages are provided according to each sub-criterion for each of the plurality of demand node. For the plurality of demand nodes, an analytical hierarchy process (AHP) model is developed for a given hierarchy and a final priority score is determined.
AHP is a multi-criteria decision-making technique which works on hierarchical relationship. In an embodiment, demand node prioritization problem is set as hierarchical relationship. Here, first level is goal of the problem which is ranking of demand nodes and second level is criteria and sub-criteria (i.e. 3 criteria and 9 sub-criteria) and third level is alternatives namely demand nodes such as hospitals and local supplier. At each level, a comparison matrix is calculated using one ranking scale from 1-9, as shown in Table 1 below
Importance Value Definition Description
1 Equal Strong Two factors are equally Contributing to objective
3 Moderate Strong One factor is marginally superior over other
5 Fairly Strong One factor is strongly superior over other
7 Very Strong One factor is very strongly superior over other
9 Absolute Strong The highest level of superiority of one factor over other
2,4,6,8 Intermediate Values According to the negotiation required
Table 1
Table 2 below provides a generic pairwise comparison matrix, where IV denotes importance value.
Criteria C1 C2 C3
C1 1 IV12 IV13
C2 1/ IV12 1 IV23
C3 1/ IV13 1/ IV23 1
Table 2
Table 3 below provides an example of pairwise comparison matrix, where criteria C1, C2 and C3 are oxygen supply, facility and geographical region respectively.
Criteria Oxygen Supply Facility Geographical Region
Oxygen Supply 1 5 7
Facility 1/ 5 1 3
Geographical Region 1/7 1/ 3 1
Table 3
It is observed from Table 3 that pairwise comparison of same criteria (i.e. oxygen supply to oxygen supply) gives a value 1. Whereas pairwise comparison of oxygen supply to facility has high importance value in comparison to pairwise comparison of facility to oxygen supply. Upon calculation of comparison matrix, one or more eigen values and eigen vectors (?) are computed as shown in equation (4). Further, a plurality of weights indicative of priority scores is determined using equation 3, 4 and 5 provided below as:
W_ij=A_ij/(?¦A_kj ) ? i,j,k (3)
?W'?_i=?=(?¦W_ij )/n (4)
W^'=[?W'?_1,?W'?_2,…….?W'?_n ]^T (5)
Here, i,j,k represents criteria, W_ij represents a weight matrix based on the comparison matrix, A_ij represents comparison value of i^th criteria to j^th criteria, A_kj represents comparison value of k^th criteria to j^th criteria in each column, W_i^' represents eigen vector for i^th criteria which is also representative of normalized weights for i^th criteria, and W^' represents transposed weight vector. Further, a step of consistency checking is performed by determining consistency of a matrix, consistency index and consistency ratio using equation (6), (7), (8) and (9) respectively.
C_i=C_1,C_2,C_3,……..C_n ? i (6)
?_max= (?¦C_i )/n (7)
C.I.= (?_max-n)/(n-1) (8)
C.R.=(C.I.)/(R.I.) (9)

Here, C_i represents consistency of i^th criteria where i=1 to n, ?_max represents maximum value of eigen vector, C.I. denotes consistency index and R.I. is random consistency index, whereas C.R. denotes consistency ratio.
The steps of calculating comparison matrix and determining the plurality of weights are iteratively performed for all the nodes to compute and a final weight of the whole hierarchy. The plurality of demand nodes is ranked in accordance with their respective weight from the plurality of weights. A pseudocode depicting the process of demand node prioritization is provided below as:
Loop i = 1 to n
Read decision variables a_1,a_2,…,? a?_n
Produce Comparison Matrix
Loop k = 1 to n-1
If imp(n) > imp(n-1) then
Comparison Matrix (n to n-1) =IV (IV=2,…,9)
else
if imp(n) = imp(n-1) then
Comparison Matrix (n to n-1) = 1
else
Comparison Matrix (n to n-1) = 1
End loop
end loop

Compute eigenvalue and eigenvector
Loop i=1 to n
Loop j = 1 to n
Sum(j) = sum(a_1j,a_2j,a_3j,…,a_nj)
E_((ij)) = a_ij/(sum(j),where (E_((ij)) = eigenvalue))
End loop
?_max = avg(a_i1,a_i2,a_i3,…,a_in ), where (?_max= eigenvector)
End loop
loop d = 1 to a (a is the number of alternative aspirants)
Read decision variable rating, r_i,where (i = 1,2,3,4,5)
compute? w?_i=?¦?r_i R(i) ?
end loop
rank r = relative size of w_i
return
Upon finding the ranking for each of the plurality of demand node, a thematic map is built by developing geocode for each demand node based on address (without latitude-longitude). The priority scores are mapped to geocodes and provide a color like red for high priority nodes, blue for low priority nodes. Referring back to FIG. 2, at step 214, the one or more processors 104 are configured to allocate, an optimal oxygen supply to the plurality of demand nodes using a mixed integer programming (MIP) based optimization model for a critical demand and a normal demand based on the plurality of priority scores. In other words, after prioritization of the plurality of demand nodes, the oxygen supply should be allocated for the plurality of demand nodes using the priority score, critical demand and normal demand. In an embodiment, supply allocation planning minimizes time required to deliver oxygen to various demand nodes spread across geographies by allocating supply nodes to demand nodes. Total supply available is fixed and limited to fulfill the demand and it requires prioritizing the demand based on criticality of each demand node. FIG. 7 is a block diagram illustrating process of planning oxygen supply allocation to the plurality of demand nodes, in accordance with some embodiments of the present disclosure. In an embodiment, the supply allocation starts with first goal to fulfill critical demand. As depicted in FIG. 7, if total available supply is more than total critical demand from all nodes, then it is straight forward to fulfill complete critical demand of all the nodes and it will be critical demand fulfillment (A). However, if total available supply is less than total critical demand, then priority scoring of the plurality of demand nodes with critical demand is required to decide which demand nodes to be fulfilled to maximize priority scoring. For critical demand fulfilment, mix integer linear programming with objective of maximizing priority score is used with satisfying minimum fulfilment constraint. In an embodiment, mathematical formulation of mixed integer programming model with an objective to maximize priority scoring to fulfill critical demand with available supply requires decision variables including x_ij which indicates x units of supply from i^th supply node to j^th demand node, s_i indicating total supply from node i, cd_j indicating total critical demand at node j, m_j indicating minimum allocation to node j, and ?SC?_j referring to priority score of demand at node j. The Objective function is to Maximize ?¦?x_ij* ?SC?_j ?, wherein the objective function is constrained to oxygen supply, critical demand and transportation feasibility. The constraints on the objective function are as follows:
Constraint 1: Allocation should not exceed the available supply from any supply node ?¦x_ij = s_i for all i
Constraint 2: Allocation should not exceed the critical demand at any demand node, ?¦x_ij = ?cd?_j for all j
Constraint 3: Minimum allocation constraint ?¦x_ij = m_j for all j
Constraint 4: Transportation feasibility constraint
t_ij={¦(0 No transportation link and x_ij=0@1 Any established link and x_ij=0 )¦
After solving above mixed integer programming problems, critical demand fulfillment (A) can be derived where total supply is less than total critical demand. Once critical demand fulfillment is achieved, supply data is updated for leftover supplies to start fulfilling normal demand which is second goal. The approach for fulling normal demand is same as fulfilling critical demand. If total updated available supply is more than total normal demand from all nodes, then complete normal demand is fulfilled. If updated supply is less than total normal demand, then again, mixed integer programming is used for normal demand fulfillment (B). In an embodiment, total fulfillment is sum of critical demand fulfillment (A) and normal demand fulfillment (B).
In an embodiment, after identifying demand fulfillment, a mixed integer programming model with objective of minimizing delivery lead time provides final allocation of supply nodes to demand nodes to fulfill maximum demand with minimal delivery time. Here, the mixed integer programming model is formulated by minimizing an objective function as Minimize ?¦?x_ij* ?LT?_ij ?of More specifically, the objective function is to minimize total time required to fulfill both critical demand fulfillment (A) and normal demand fulfillment (B). Here, decision variables x_ij indicates x units of supply from i^th supply node to j^th demand node. s_i indicates total supply from node i, a_j indicates total demand fulfilment for demand node j, ?LT?_ij indicates time required from supply node i to demand node j (alternatively referred as lead time). In an embodiment, the objective function is constrained to:
Constraint 1: oxygen supply, critical demand, transportation feasibility. Allocation should not exceed the available supply from any supply node, ?¦x_ij = s_i for all i.
Constraint 2: Allocation should not be less than fulfillment quantity ?¦x_ij = a_j for all j
Constraint 3: Transportation feasibility constraint
t_ij={¦(0 No transportation link and x_ij=0@1 Any established link and x_ij=0 )¦
After solving above mixed integer programming problem values of x_ij are the final supply allocation from i^th supply node to j^th demand node. FIG. 8 is an interconnected graph showing an example of oxygen supply allocation to the plurality of demand nodes, in accordance with some embodiments of the present disclosure. In FIG. 8, s_i are supply nodes and d_i are demand nodes. It can be seen from FIG. 8 that all nodes are distributed at different place and connected to each other with transportation link.
Referring to FIG. 2, at step 216, the one or more hardware processors 104 are configured to execute, an optimized dynamic routing and re-routing of a plurality of vehicles based on an adaptive parallel vehicle based genetic algorithm for optimized distribution of the optimal oxygen supply to the plurality of demand nodes, wherein the adaptive parallel vehicle based genetic algorithm optimizes the dynamic routing of the plurality of vehicles based on minimum delivery time and minimum congestion. In an embodiment, the allocated supplies are used as capacity constraint for dynamic route planning for hassle free distribution. In an embodiment, vehicle planning, and route planning is based on demand/supply nodes’ location, storage capacity, oxygen inventory, vehicle capacity, cylinder availability (Filled and empty), and route information like distance to travel, time to travel, average speed, and congestion along with operational constraints or government regulations, if any. The objective is to minimize travel time along with congestion whereas maximum and minimum constraint (e.g. maximum 3 days and minimum 1 day) for demand fulfillment is provided. For smooth distribution, it is assumed that the hospitals return empty cylinder on next supply (e.g. within 3 days) and retailers return all the cylinders within a given time limit (e.g. 2 weeks). If they are failed to do so, the next supply becomes equal to returned empty cylinders only.
In an embodiment, the adaptive parallel vehicle based genetic algorithm to perform the dynamic route planning and re-routing of the plurality of vehicles further comprising: executing, in a simultaneous manner, a genetic algorithm (GA) for each of the plurality of vehicles and adaptively updating, in real time, an information pertaining to one or more vehicle constraints and availability of one or more resources after convergence of the genetic algorithm of each of the plurality of vehicles till a final route for each of the plurality of vehicles is obtained. In an embodiment, the objective of dynamic route planning is to minimize the overall delivery time along with congestion. Thus, for dynamic route planning, an improved genetic algorithm is deployed. The improvement has been done to minimize the total time to converge and to provide the feasible solution. The method of the proposed disclosure utilizes parallel genetic algorithm in an adaptive manner. However, a classic parallel genetic algorithm is used to get multiple local solutions and best of best solution is selected by genetic algorithm master, but in the method of the present disclosure, multiple genetic algorithms (GA)s are running parallelly and each GA is assigned to one vehicle. Therefore, each GA provides one best (near-optimal) route for each vehicle. To make it adaptive, result of each GA (after conversion, the best route for 1 vehicle) is updated with genetic algorithm master and remaining capacities and demands are also updated simultaneously. The objective for dynamic routing is formulated in such a way that it provides flexibility to the user to assign the weights to each objective. For example, if user wants to focus majorly on minimizing delivery time, higher weightage is assigned to delivery time. The objective is defined as follows:
Minimize (w_1*normalized delivery time+ w_2*normalized total congestion)
Where, w_1+w_2=1
FIG. 9 is a flow diagram illustrating working of adaptive parallel vehicle based genetic algorithm (APVGA), in accordance with some embodiments of the present disclosure. As can be seen in FIG. 9, In adaptive parallel vehicle based genetic algorithm (APVGA), the genetic algorithm master receives inputs from various data sources and share information with all the parallel GAs. Here, the number of parallel GA depends on number of vehicles and each GA provides a solution of one vehicle route. FIG. 10 is a flow diagram illustrating working of genetic algorithm for one vehicle routing, in accordance with some embodiments of the present disclosure. As depicted in FIG. 10, process of each of the Genetic Algorithms (GA), which are running parallelly for optimal routing includes an initial population, selection, crossover, and mutation. In an embodiment, genetic algorithms maintain population diversity speeding up convergence and avoid premature convergence. In an embodiment, the initial population aims to randomly generate possible feasible solution by constructing random route, along with satisfying constraints such as vehicle capacity, cylinder availability, oxygen supply. Further, the objective function is defined as a fitness function which gives fitness score to each individual solution. The probability of individual selected for reproduction is based on fitness score. The selection step includes selection of parent based on fitness function, wherein the individual with high fitness score has chances to be selected. Further, crossover probability is applied on selected parents to produce offspring. Upon crossover, a uniform mutation operator is applied to produce offspring with mutation probability to mutant offspring. The new offspring are then placed in new population. Finally, the genetic algorithm terminates when generation reaches to maximum defined number. In an embodiment, the genetic algorithm dynamically changes search process through the probabilities of crossover and mutation to reach optimal solution. Further, for adaptivity, the updated route and remaining capacity is shared continuously through GA master to all the parallel GA. For example, if GA for vehicle1 is converged which mean that the route is final for vehicle1, then the remaining availability of oxygen and cylinders and the plurality of demand nodes are updated to remaining GAs. Further, genetic algorithm approach also helps to re-route in real time such as in case of any exception demand due to sudden surge and the rerouting is performed with remaining capacities including oxygen, cylinder, vehicle, and/or the like.
In an embodiment, pseudocode for parallel vehicle based genetic algorithm (APVGA) is provided below as:
Objective: Optimal routing in delivery of oxygen cylinder for demand fulfilment using APVGA
Input: instance graph denoted by G(N,A), number of vehicles denoted by v, demand at node i denoted by d_i, average travel time denoted by t_ij, Congestion at each segment denoted by c_ij, weightages of delivery time and congestion denoted by w_1 and w_2 respectively.
Initialize parameters of GA for i^th vehicle (?GA?_i ) as: P_i (population size parameter for i^th GA) ?maxGEN?_i;?pc?_i_ini; ?pc?_i_min;?pm?_i_ini; ?pm?_i_mi
vp=0 vehicles planned
routePlans=list of route plan
max?? 1/(?w_1 t?_ij+w_2 c_ij )? fitness function
While vp= v
Start parallel threads for all vehicles if not scheduled earlier
Begin
Generate first feasible plan for vehicle i using
?gen?_i=1
generate initial population
Constraints ?¦d_i =vehicle capacity
oxygen delivery =oxygen availability
Number of cylenders =cylender availability
evaluate the population
for i:=1 to P do
evaluate (p[i])
while ?gen?_i