# US 2022/0036166 A1 · System and method for operation optimization of an equipment > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2022/0036166 A1 - Document ID: US20220036166A1 - Status: Published application - Kind: A1 - Application number: 17/309,432 - Filed: 2019-11-28 - Published: 2022-02-03 - Earliest claimed: 2018-11-28 - Assignees: Tata Consultancy Services Limited - Applicants: Tata Consultancy Services Limited - Inventors: Vishnu TANKASALA VEPARALA; Solomon Pushparaj MANUELRAJ; Ankit BANSAL; Pankaj MALHOTRA; Lovekesh VIG; Gautam SHROFF; Venkataramana RUNKANA; Sivakumar SUBRAMANIAN; Aditya PAREEK; Vishnu Swaroopji MASAMPALLY; Nishit RAJ - Web page: https://patentgrep.com/patent/US20220036166A1 ## CPC - G06N 3/006: based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO] - G06N 3/063: using electronic means - G06N 3/042: Knowledge-based neural networks; Logical representations of neural networks - G06N 3/044: Recurrent networks, e.g. Hopfield networks - G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] - G06N 3/045: Combinations of networks - G06N 3/08: Learning methods - G06N 3/092: Reinforcement learning - G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn - G06N 3/10: Interfaces, programming languages or software development kits, e.g. for simulating neural networks - G06N 5/04: Inference or reasoning models - G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract This disclosure relates to optimizing an operation of an equipment by a neural network based optimizer is provided. The method include receiving, information associated with at least one equipment instance (j) as an input at a predefined sequence of timestamps; training, a plurality of simulation models for each equipment instance to obtain a function (fj); processing, the external input parameters (et) to obtain a fixed-dimensional vector and passed as an input to obtain an vector (it); generating, a modified (it) from the output vector (it) based on a domain constraint value; computing, a reward (rt) based on (i) the function (fj), (ii) the modified (it), (iii) the external input parameters (et), and (iv) a reward function (Rj); and iteratively performing the steps of processing, generating, and computing reward (rt) for a series of subsequent equipment instances after expiry of the predefined sequence of timestamps associated with a first equipment instance. ## Claims ### Claim 1 A processor implemented method of optimizing an operation of an equipment by a neural network based optimizer, comprising: receiving, information associated with at least one equipment instance (j) as an input at a predefined sequence of timestamps, wherein the information associated with the at least one equipment instance (j) corresponds to at least one of (i) a plurality of control parameters (it), (ii) external input parameters (et), (iii) output parameters (ot), and (iv) historical operational data; training, by a simulation model trainer, a plurality of simulation models for each equipment instance using the information associated with the at least one equipment instance (j) to obtain a function (fj), wherein the function (fj) corresponds to a set of simulation models of the equipment instance (j); generating, by a sequence generator, at least one sequence of the external input parameters (et); processing, by a multi-head input handler, the external input parameters (et) to obtain a fixed-dimensional vector that is passed as an input to a neural network to obtain an vector (it); generating, by a domain constraints handler, a modified (it) from the output vector (it) based on a domain constraint value; computing, by a multi-head output handler, a reward (rt) based on (i) the function (fj), (ii) the modified (it), (iii) the external input parameters (et), and (iv) a reward function (Rj); and iteratively performing the steps of processing, generating, and computing reward (rt) for a series of subsequent equipment instances after expiry of the predefined sequence of timestamps associated with a first equipment instance. ### Claim 2 (depends on 1) The method as claimed in claim 1, wherein the multi-head input handler comprises a processing head for each equipment instance, wherein number of external input parameters differs across different instances of equipments. ### Claim 3 (depends on 1) The method as claimed in claim 1, wherein the function fj is generated by at least one of (i) a Gaussian processes, or (ii) Gaussian mixture models constrained on dimensionality of the output vector it and the vector of external input parameters et, and combination thereof. ### Claim 4 (depends on 1) The method as claimed in claim 1, further comprising, inputting, the output vector (it) and reward (rt) into the Neural Network with the vector of external input parameters at time (‘t+1’), et+1 for iteratively performing the steps of processing, generating, and computing reward to obtain the set of control input parameters at time (t+1), it+1, wherein dimensionality of it and et is determined by a target domain. ### Claim 5 (depends on 1) The method as claimed in claim 1, wherein the step of training the neural network with a loss function is performed based on a maximum attainable value of the reward function (Rj). ### Claim 6 (depends on 1) The method as claimed in claim 1, wherein the neural network is trained by including an additional loss function using a plurality of penalty values for violating domain constraints. ### Claim 7 A system for operation optimization of an equipment by a neural network based optimizer, wherein 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, information associated with at least one equipment instance (j) as an input at a predefined sequence of timestamps, wherein the information associated with the at least one equipment instance (j) corresponds to at least one of (i) a plurality of control parameters (it), (ii) external input parameters (et), (iii) output parameters (ot), and (iv) historical operational data; train, a plurality of simulation models for each equipment instance using the information associated with the at least one equipment instance (j) to obtain a function (fj), wherein the function (fj) corresponds to a set of simulation models of the equipment instance (j); generate, at least one sequence of the external input parameters (et); process, the external input parameters (et) to obtain a fixed-dimensional vector that is passed as an input to a neural network to obtain an output vector (it); generate, a modified (it) from the output vector (it) based on a domain constraint value; compute, a reward (rt) based on (i) the function (fj), (ii) the modified (it), (iii) the external input parameters (et), and (iv) a reward function (Rj); and iteratively perform the steps of process, generate, and compute reward (rt) for a series of subsequent equipment instances after expiry of the predefined sequence of timestamps associated with a first equipment instance. ### Claim 8 (depends on 7) The system as claimed in claim 7, wherein a multi-head input handler comprises a processing head for each equipment instance, wherein number of external input parameters differs across different instances of equipments. ### Claim 9 (depends on 7) The system as claimed in claim 7, wherein the function fj is generated by at least one of (i) a Gaussian processes, or (ii) Gaussian mixture models constrained on dimensionality of the output vector it and the vector of external input parameters et, and combination thereof. ### Claim 10 (depends on 7) The system as claimed in claim 7, wherein the one or more hardware processors are further configured by the instructions to input, the output vector (it) and reward (rt) into the Neural Network with the vector of external input parameters at time (‘t+1’), et+1 to iteratively perform the steps of process, generate, and compute reward (rt) to obtain the set of control input parameters at time (t+1), it+1, wherein dimensionality of it and et is determined by a target domain. ### Claim 11 (depends on 7) The system as claimed in claim 7, wherein the step of training the neural network with a loss function is performed based on a maximum attainable value of the reward function (Rj). ### Claim 12 (depends on 7) The system as claimed in claim 7, wherein the neural network (208A) is trained by including an additional loss function using a plurality of penalty values for violating domain constraints. ### Claim 13 One or more non-transitory machine readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors causes: receiving, information associated with at least one equipment instance (j) as an input at a predefined sequence of timestamps, wherein the information associated with the at least one equipment instance (j) corresponds to at least one of (i) a plurality of control parameters (it), (ii) external input parameters (et), (iii) output parameters (ot), and (iv) historical operational data; training, a plurality of simulation models for each equipment instance using the information associated with the at least one equipment instance (j) to obtain a function (fj), wherein the function (fj) corresponds to a set of simulation models of the equipment instance (j); generating, at least one sequence of the external input parameters (et); processing, the external input parameters (et) to obtain a fixed-dimensional vector that is passed as an input to a neural network to obtain an vector (it); generating, a modified (it) from the output vector (it) based on a domain constraint value; computing, a reward (rt) based on (i) the function (fj), (ii) the modified (it), (iii) the external input parameters (et), and (iv) a reward function (Rj); and iteratively performing the steps of processing, generating, and computing reward (rt) for a series of subsequent equipment instances after expiry of the predefined sequence of timestamps associated with a first equipment instance. ### Claim 14 (depends on 13) The one or more non-transitory machine readable information storage mediums as claimed in claim 13, wherein the multi-head input handler comprises a processing head for each equipment instance, wherein number of external input parameters differs across different instances of equipments. ### Claim 15 (depends on 13) The one or more non-transitory machine readable information storage mediums as claimed in claim 13, wherein the function fj is generated by at least one of (i) a Gaussian processes, or (ii) Gaussian mixture models constrained on dimensionality of the output vector it and the vector of external input parameters et, and combination thereof. ### Claim 16 (depends on 13) The one or more non-transitory machine readable information storage mediums as claimed in claim 13, wherein the one or more hardware processors are further configured by the instructions for inputting, the output vector (it) and reward (rt) into the Neural Network with the vector of external input parameters at time (‘t+1’), et+1 for iteratively performing the steps of processing, generating, and computing reward to obtain the set of control input parameters at time (t+1), it+1, wherein dimensionality of it and et is determined by a target domain. ### Claim 17 (depends on 13) The one or more non-transitory machine readable information storage mediums as claimed in claim 13, wherein the step of training the neural network with a loss function is performed based on a maximum attainable value of the reward function (Rj). ### Claim 18 (depends on 13) The one or more non-transitory machine readable information storage mediums as claimed in claim 13, wherein the neural network is trained by including an additional loss function using a plurality of penalty values for violating domain constraints.