A comprehensive energy system modeling method based on meta-learning and physical information neural networks

By combining meta-learning and physical information neural networks, a comprehensive energy system model is constructed, which solves the problem of rapid modeling and real-time optimization of complex systems and achieves efficient system optimization with limited data.

CN117709198BActive Publication Date: 2026-11-17ZHEJIANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311765467.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-21
Publication Date
2026-11-17
Estimated Expiration
2043-12-21

AI Technical Summary

Technical Problem

In integrated energy systems, existing technologies struggle to quickly solve physical models and perform real-time optimizations for complex systems. Traditional numerical simulation models and neural network methods have high training costs and cannot meet the real-time modeling requirements for system state changes.

Method used

A comprehensive energy system model is constructed by combining meta-learning and physical information neural networks. By collecting operational data to construct training and testing task sets, and combining equipment models and pipeline physical formulas, a meta-learning model is built to achieve rapid learning and optimization.

Benefits of technology

Under specific system conditions, the model can be quickly learned and optimized using a small amount of data to meet the overall system's optimal modeling requirements. It takes into account network topology and device mechanism models, thereby improving the efficiency and accuracy of modeling.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117709198B_ABST
    Figure CN117709198B_ABST
Patent Text Reader

Abstract

This invention discloses a comprehensive energy system modeling method based on meta-learning and physical information neural networks. The comprehensive energy system model constructed based on this invention can simultaneously consider network topology, equipment mechanism models, and pipeline physical models. The meta-learning model constructed in this invention can learn all the parameter variables in the physical information neural network models of different pipelines that require human setting and definition, under a limited amount of monitoring data in a specific system state. This allows for the rapid learning of the spatial and temporal distribution of operating data (specific operating data can be determined based on the key parameters to be monitored in the system, such as temperature, flow rate, and pressure) in each pipe segment and equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of integrated energy system design, planning and modeling, and specifically relates to an integrated energy system modeling method based on meta-learning and physical information neural networks. Background Technology

[0002] With the deepening of the carbon neutrality concept, integrated energy systems are undergoing transformation and upgrading towards efficient operation and low-carbon energy consumption. In integrated energy systems, sensors deployed on pipelines monitor the state of working fluids of different energy forms for equipment scheduling optimization, economic control, and fault monitoring. However, in reality, integrated energy systems are quite complex. Physical models for certain energy flow patterns are difficult to solve analytically, and while simulation software can approximate the system, its solution speed is too slow and cannot be used for real-time system optimization. Furthermore, if neural network methods are used for black-box fitting, retraining or fine-tuning based on historical data is required when the system state changes, resulting in high training costs. Therefore, under these circumstances, traditional numerical simulation models and neural network methods may not be able to predict satisfactory results, making modeling integrated energy systems challenging. Summary of the Invention

[0003] The purpose of this invention is to provide a comprehensive energy system modeling method based on meta-learning and physical information neural networks. The comprehensive energy system model constructed based on this invention can simultaneously consider network topology, equipment mechanism models, and pipeline physical models. The meta-learning model constructed by this invention can learn all the parameter variables in the physical information neural network models of different pipelines that require human setting and definition, under a limited amount of monitoring data in a specific system state. This allows for the rapid learning of the spatial and temporal distribution of operating data (specific operating data can be determined based on the key parameters to be monitored in the system, such as temperature, flow rate, and pressure) in each pipe segment and equipment.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A comprehensive energy system modeling method based on meta-learning and physical information neural networks includes the following steps:

[0006] S1, for a certain integrated energy system, collect the operation data (such as temperature) of the pipeline at various locations and times under different system states, and construct the training task set and the test task set;

[0007] S2. Based on the network topology of the integrated energy system and the characteristics of the equipment on the pipeline, construct a black box model (heat exchanger, etc.) or a white box model (pump, valve, etc.) for the equipment.

[0008] S3, based on the physical formula of heat transfer in the pipeline, constructs a neural network model of the physical information of each pipeline in the integrated energy system;

[0009] S4. Based on all the equipment models constructed in step S2 and the physical information neural network models of all the pipelines constructed in step S3, construct a meta-learning model of the integrated energy system pipeline network based on the training task set, and verify the meta-learning model based on the test task set.

[0010] S5. Input a small amount of operational data under a specific system state into the meta-learning model constructed in step S4 to obtain a modeling result that satisfies the overall system optimality.

[0011] In the above technical solution, step S1 further includes the following steps:

[0012] Step S11: Based on the energy supply and load sides of the integrated energy system, as well as the topology of the pipeline network, the state of the integrated energy system can be obtained. In the j-th state of the integrated energy system, the change of operating data in the i-th pipeline with time t and location x is represented as OP(x,t). i | j Then, in the j-th state of the integrated energy system, the operating dataset containing n pipelines is represented as {OP(x,t)1| j ,OP(x,t)2| j ,…,OP(x,t) i | j ,…,OP(x,t) n | j}, 1≤o≤n, simplified as {OP(x,t)} n,j Taking the temperature of the working fluid in the i-th pipe under the j-th state of the integrated energy system as an example, it is denoted as T(x,t). i | j A dataset containing n pipes can be specifically represented as {T(x,t)1| j ,T(x,t)2| j ,…,Txt) n | j}, simplified as {T(x,t)} n,j .

[0013] Step S12: Based on the operational dataset of each pipeline in the integrated energy system under m different states, construct a task set, denoted as {{OP(x,t)}. n,1 ,{OP(x,t)} n,2 ,...,{OP(x,t)} n,j ,...,{OP(x,t)} n,m}, 1≤j≤m, the task set is divided into a training task set and a test task set for subsequent training and validation of the meta-learning model. Taking temperature as an example, based on the temperature dataset of each pipe in the integrated energy system under m different states, the constructed task set is {{T(x,t)}. n,1 ,{Txt)} n,2 ,...,{Txt)} n,m The task set is divided into a training task set and a test task set by using a specific ratio or random number.

[0014] Furthermore, step S2 includes the following steps:

[0015] Based on historical operating data from the pipelines connected to the equipment in the integrated energy system, white-box models are constructed for pumps and valves, and black-box models are constructed for heat exchangers and turbine units. The white-box or black-box models are represented by the following formula:

[0016] OP out =f(OP) in ,j)

[0017] Among them, OP out The output is the model, where f is the constructed white-box or black-box model, and OP is the model output. in The input to the model is j, which represents the j-th state of the system.

[0018] The models built based on the above two types of equipment can be applied to different system states of integrated energy systems. Taking single-variable temperature as an example, the white-box model of pumps, valves, etc. on a single pipeline can be expressed by the following formula:

[0019] T o =f w (T i ,m)

[0020] Among them, T o The model outputs temperature, f w For the constructed white-box model, T i Input the temperature to the model, where m represents the system state. For black-box models of equipment such as heat exchangers connected by multiple pipes, the following formula can be used:

[0021] To = f b (Ti,m)

[0022] Among them, T o Output the temperature matrix for the model, f b For the constructed white-box model, T i Input the temperature matrix into the model.

[0023] Furthermore, step S3 includes the following steps:

[0024] Step S31, the physical formula for heat transfer of fluid within the integrated energy system pipeline is expressed as:

[0025] F(OP,x,t)=0

[0026] Based on the above formula and the boundary conditions BC of each pipeline in the j-th state of the integrated energy system. j (x,t), constructing a physical information neural network model of each pipeline in the integrated energy system using location x and time t as features; the loss function of the physical information neural network model of the i-th pipeline is:

[0027]

[0028] Among them, Loss pinn | i Let be the training error of the neural network for the physical information of the i-th pipe, where a, b, and c are the weights for different errors, and MSE() is the mean squared error loss. OP represents the predicted values ​​of the running data, and BC represents the actual values ​​of the running data. j (x,t) represents the boundary conditions of the pipeline.

[0029] Taking the steady-state thermodynamics of a heating system as an example, the steady-state heat dissipation formula for a single pipe, derived from the energy conservation equation, is expressed as follows:

[0030]

[0031] Where c is the specific heat capacity of water, ρ is the density of water, A is the cross-sectional area of ​​the pipe, G is the mass flow rate of the pipe, and μ is the heat loss coefficient of the pipe to the outside.

[0032] Based on the above formula and the boundary conditions BC of each pipe in the j-th state of the system. j (x,t), constructing a physical information neural network model of each pipeline in the integrated energy system using location x and time t as features; the loss function of the physical information neural network model of the i-th pipeline is:

[0033]

[0034] Among them, Loss pinn | i Let be the training error of the neural network for the physical information of the i-th pipe, where a, b, and c are the weights for different errors, and MSE() is the mean squared error loss. Here, T represents the predicted temperature, and BC represents the actual temperature. j (x,t) represents the boundary conditions of the pipeline.

[0035] Step S32, generate the operating dataset {OP(x,t)} of the integrated energy system in the j-th state.n,j The dataset is divided into a training set and a test set, denoted as {OP(x,t)}. n,j train ,{OP(x,t))} n,j test The training set {OP(x,t)} is used for training. n,j train By calculating Loss pinn And the backpropagation mechanism is used to update the network weights of the neural network model of the physical information of each pipeline in the integrated energy system; while the test set {OP(x,t)} n,j test This is used to verify the training effect of the model.

[0036] Furthermore, step S4 includes the following steps:

[0037] Step S41: Based on the training task set constructed in step S1, train a meta-learning model of the integrated energy system pipeline network under different system states. The loss function of this meta-learning neural network model is:

[0038]

[0039] Loss eq,i =MSE(OP) in At that time, OP in )+MSE(OP out ',OP out )

[0040] Loss topo,o =MSE(OP) node,i ',OP node,i )

[0041] Among them, Loss meta Let α, β, and γ be the training error of the meta-learning neural network, and let Loss be the weights for different errors. pinn | i Loss represents the physical information neural network error of the i-th pipe. eq,i Loss topo,i Let be the errors in the operating data of the i-th device and the i-th node, respectively, and OP be the metric. in ′ and OP in These are the predicted and actual values ​​of the operating data for the fluid flowing into the equipment from the i-th pipeline, respectively. out ′ and OP out These are the predicted and actual operating data values ​​for the i-th pipeline fluid outflow device side, respectively. node,i ′ and OP node,i These are the predicted and actual values ​​of the operating data at the i-th pipeline terminal node, respectively.

[0042] Taking the pipe network of a district heating system as an example, a meta-learning model of the pipe network under different system states is trained. The loss function of this meta-learning neural network model is:

[0043]

[0044] Loss eq,i =MSE(T) in ′,T in )+MSE(T out ′,T out )+MSE(G in ′,G in )+MSE(G out ′,G out )

[0045] Loss topo,i =MSE(T) node,i ′,T node,i )+MSE(G node,i ′,G node,i )

[0046] Among them, T in ′ and T in T represents the predicted and actual temperature values ​​of the fluid flowing into the equipment from the i-th pipe. out ′ and T out G represents the predicted and actual temperature values ​​of the fluid flowing out of the equipment in the i-th pipe, respectively. in ′ and G in G represents the predicted and actual mass flow rates of the fluid flowing into the equipment from the i-th pipe, respectively. out ′ and G out T represents the predicted and actual mass flow rates of the fluid exiting the equipment from the i-th pipe, respectively. node,i ′ and T node,i G represents the predicted and actual temperature values ​​at the i-th pipe end node, respectively. node,i ′ and G node,i These are the predicted and actual mass flow rates at the i-th pipeline terminal node, respectively.

[0047] Step S42: Based on the constructed test task set, verify the effect of the trained meta-learning neural network on the physical information neural network of each pipeline under different system states.

[0048] Furthermore, step S5 includes the following steps:

[0049] When a trained and validated meta-learning network is used in a real integrated energy system, the current state of the system is input into the meta-learning neural network. The network can provide the physical information corresponding to each pipeline. All the parameter variables in the neural network that need to be set and defined by humans can be obtained, thereby achieving the modeling result with the least overall system error.

[0050] The beneficial effects of this invention are:

[0051] The purpose of this invention is to provide a comprehensive energy system modeling method based on meta-learning and physical information neural networks. First, operational data of pipelines at various locations and times under different system states are collected to construct training and testing task sets. Based on the network topology of the comprehensive energy system and the characteristics of equipment on the pipeline network, black-box models of equipment such as heat exchangers and white-box models of equipment such as pumps and valves are constructed. Based on the physical formulas of fluid heat transfer in pipelines, physical information neural network models of each pipeline in the comprehensive energy system are constructed. A meta-learning model of the comprehensive energy system pipeline network is constructed based on the training task set, and the testing task set is used for model verification. A small amount of data under specific system states is input into the meta-learning model to obtain the optimal modeling result for the overall system. The comprehensive energy system model constructed by this invention can simultaneously consider network topology, equipment mechanism models, and pipeline physical models. The constructed meta-learning model can learn all the parameter variables that need to be set and defined manually in the physical information neural network models of different pipelines under a small amount of operational data in a specific system state, thereby quickly learning the spatial and temporal distribution of operational data in each pipe segment and equipment. Attached Figure Description

[0052] Figure 1 This is a flowchart of the integrated energy system modeling method based on meta-learning and physical information neural networks as described in this invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings.

[0054] A comprehensive energy system modeling method based on meta-learning and physical information neural networks includes the following steps:

[0055] S1, for a certain integrated energy system, collect the operation data (such as temperature) of the pipeline at various locations and times under different system states, and construct the training task set and the test task set;

[0056] S2. Based on the network topology of the integrated energy system and the characteristics of the equipment on the pipeline, construct a black box model (heat exchanger, etc.) or a white box model (pump, valve, etc.) for the equipment.

[0057] S3, based on the physical formula of heat transfer in the pipeline, constructs a neural network model of the physical information of each pipeline in the integrated energy system;

[0058] S4. Based on all the equipment models constructed in step S2 and the physical information neural network models of all the pipelines constructed in step S3, construct a meta-learning model of the integrated energy system pipeline network based on the training task set, and verify the meta-learning model based on the test task set.

[0059] S5. Input a small amount of operational data under a specific system state into the meta-learning model constructed in step S4 to obtain a modeling result that satisfies the overall system optimality.

[0060] Step S1 includes the following steps:

[0061] Step S11: Based on the energy supply and load sides of the integrated energy system, as well as the topology of the pipeline network, the state of the integrated energy system can be obtained. In the j-th state of the integrated energy system, the change of operating data in the i-th pipeline with time t and location x is represented as OP(x,t). i | j Then, in the j-th state of the integrated energy system, the operating dataset containing n pipelines is represented as {OP(x,t)1| j ,OP(x,t)2| j ,…,OP(x,t) i | j ,…,OP(x,t) n | j}, 1≤i≤n, simplified as {OP(x,t)} n,j Taking the temperature of the working fluid in the i-th pipe under the j-th state of the integrated energy system as an example, it is denoted as T(x,t). i | j A dataset containing n pipes can be specifically represented as {T(x,t)1| j ,T(x,t)2| j ,…,Txt) n | j}, simplified as {T(x,t)} n,j .

[0062] Step S12: Based on the operational dataset of each pipeline in the integrated energy system under m different states, construct a task set, denoted as {{OP(x,t)}. n,1 ,{OP(x,t)} n,2 ,...,{OP(x,t)} n,j ,...,{OP(x,t)} n,m}, 1≤j≤m, the task set is divided into a training task set and a test task set for subsequent training and validation of the meta-learning model. Taking temperature as an example, based on the temperature dataset of each pipe in the integrated energy system under m different states, the constructed task set is {{t(x,t)}. n,1 ,{Txt)} n,2 ,...,{Txt)} n,m The task set is divided into a training task set and a test task set by using a specific ratio or random number.

[0063] Step S2 includes the following steps:

[0064] Based on historical operating data from the pipelines connected to the equipment in the integrated energy system, white-box models are constructed for pumps and valves, and black-box models are constructed for heat exchangers and turbine units. The white-box or black-box models are represented by the following formula:

[0065] OP out =f(OP) in ,j)

[0066] Among them, OP out The output is the model, where f is the constructed white-box or black-box model, and OP is the model output. in The input to the model is j, which represents the j-th state of the system.

[0067] The models built based on the above two types of equipment can be applied to different system states of integrated energy systems. Taking single-variable temperature as an example, the white-box model of pumps, valves, etc. on a single pipeline can be expressed by the following formula:

[0068] T o =f w (T i ,m)

[0069] Among them, T o The model outputs temperature, f w For the constructed white-box model, T i Input the temperature to the model, where m represents the system state. For black-box models of equipment such as heat exchangers connected by multiple pipes, the following formula can be used:

[0070] To = f b (Ti,m)

[0071] Among them, T o Output the temperature matrix for the model, f b For the constructed white-box model, T i Input the temperature matrix into the model.

[0072] Step S3 includes the following steps:

[0073] Step S31, the physical formula for heat transfer of fluid within the integrated energy system pipeline is expressed as:

[0074] F(OP,x,t)=0

[0075] Based on the above formula and the boundary conditions BC of each pipeline in the j-th state of the integrated energy system. j (x,t), constructing a physical information neural network model of each pipeline in the integrated energy system using location x and time t as features; the loss function of the physical information neural network model of the i-th pipeline is:

[0076]

[0077] Among them, Loss pinn | i Let be the training error of the neural network for the physical information of the i-th pipe, where a, b, and c are the weights for different errors, and MSE() is the mean squared error loss. OP represents the predicted values ​​of the running data, and BC represents the actual values ​​of the running data. j (x,t) represents the boundary conditions of the pipeline.

[0078] Taking the steady-state thermodynamics of a heating system as an example, the steady-state heat dissipation formula for a single pipe, derived from the energy conservation equation, is expressed as follows:

[0079]

[0080] Where c is the specific heat capacity of water, ρ is the density of water, A is the cross-sectional area of ​​the pipe, G is the mass flow rate of the pipe, and μ is the heat loss coefficient of the pipe to the outside.

[0081] Based on the above formula and the boundary conditions BC of each pipe in the j-th state of the system. j (x,t), constructing a physical information neural network model of each pipeline in the integrated energy system using location x and time t as features; the loss function of the physical information neural network model of the i-th pipeline is:

[0082]

[0083] Among them, Loss pinn | i Let be the training error of the neural network for the physical information of the i-th pipe, where a, b, and c are the weights for different errors, and MSE() is the mean squared error loss. Here, T represents the predicted temperature, and BC represents the actual temperature. j (x,t) represents the boundary conditions of the pipeline.

[0084] Step S32, generate the operating dataset {OP(x,t)} of the integrated energy system in the j-th state. n,jThe dataset is divided into a training set and a test set, denoted as {OP(x,t)}. n,j train ,{OP(x,t))} n,j test The training set {OP(x,t)} is used for training. n,j train By calculating Loss pinn And the backpropagation mechanism is used to update the network weights of the neural network model of the physical information of each pipeline in the integrated energy system; while the test set {OP(x,t)} n,j test This is used to verify the training effect of the model.

[0085] Step S4 includes the following steps:

[0086] Step S41: Based on the training task set constructed in step S1, train a meta-learning model of the integrated energy system pipeline network under different system states. The loss function of this meta-learning neural network model is:

[0087]

[0088] Loss eq,i =MSE(OP) in ',OP in )+MSE(OP out ',OP out )

[0089] Loss topo,i =MSE(OP) node,i ',OP node,i )

[0090] Among them, Loss meta Let α, β, and γ be the training error of the meta-learning neural network, and let Loss be the weights for different errors. pinn | i Loss represents the physical information neural network error of the i-th pipe. eq,i Loss topo,i Let be the errors in the operating data of the i-th device and the i-th node, respectively, and OP be the metric. in ′ and OP in These are the predicted and actual values ​​of the operating data for the fluid flowing into the equipment from the i-th pipeline, respectively. out ′ and OP out These are the predicted and actual operating data values ​​for the i-th pipeline fluid outflow device side, respectively. node,i ′ and OP node,i These are the predicted and actual values ​​of the operating data at the i-th pipeline terminal node, respectively.

[0091] Taking the pipe network of a district heating system as an example, a meta-learning model of the pipe network under different system states is trained. The loss function of this meta-learning neural network model is:

[0092]

[0093] Loss eq,i =MSE(T) in ′,T in )+MSE(T out ′,T out )+MSE(G in ′,G in )+MSE(G out ′,G out )

[0094] Loss topo,i =MSE(T) node,i ′,T node,i )+MSE(G node,i ′,G node,i )

[0095] Among them, T in ′ and T in T represents the predicted and actual temperature values ​​of the fluid flowing into the equipment from the i-th pipe. out ′ and T out G represents the predicted and actual temperature values ​​of the fluid flowing out of the equipment in the i-th pipe, respectively. in ′ and G in G represents the predicted and actual mass flow rates of the fluid flowing into the equipment from the i-th pipe, respectively. out ′ and G out T represents the predicted and actual mass flow rates of the fluid exiting the equipment from the i-th pipe, respectively. node,i ′ and T node,i G represents the predicted and actual temperature values ​​at the i-th pipe end node, respectively. node,i ′ and G node,i These are the predicted and actual mass flow rates at the i-th pipeline terminal node, respectively.

[0096] Step S42: Based on the constructed test task set, verify the effect of the trained meta-learning neural network on the physical information neural network of each pipeline under different system states.

[0097] Step S5 includes the following steps:

[0098] When a trained and validated meta-learning network is used in a real integrated energy system, the current state of the system is input into the meta-learning neural network. The network can provide the physical information corresponding to each pipeline. All the parameter variables in the neural network that need to be set and defined by humans can be obtained, thereby achieving the modeling result with the least overall system error.

Claims

1. A method for integrated energy system modeling based on meta-learning and physics-informed neural networks, characterized in that, Includes the following steps: S1, for a specific integrated energy system, collect operational data of pipelines at various locations and times under different system states, and construct training task sets and test task sets; S2. Based on the network topology of the integrated energy system and the characteristics of the equipment on the pipeline, construct a black-box model or white-box model of the equipment. S3, based on the physical formula of heat transfer in the pipeline, constructs a physical information neural network model of each pipeline in the integrated energy system; S4. Based on all the equipment models constructed in step S2 and the physical information neural network models of all the pipelines constructed in step S3, construct a meta-learning model of the integrated energy system pipeline network based on the training task set, and verify the meta-learning model based on the test task set. S5. Input the system's operating data into the meta-learning model constructed in step S4 to obtain the modeling result that satisfies the overall system's optimality. Step S3 includes the following steps: Step S31, the physical formula for heat transfer of fluid within the integrated energy system pipeline is expressed as: Based on the above formula and the boundary conditions of each pipeline in the jth state of the integrated energy system A physical information neural network model of each pipeline in the integrated energy system is constructed with position x and time t as characteristics; and a loss function of the physical information neural network model of the ith pipeline is: in, Let be the training error of the neural network for the physical information of the i-th pipe, where a, b, and c are the weights for different errors, and MSE() is the mean squared error loss. The predicted values ​​for the running data, For the actual values ​​of the running data, These are the boundary conditions for the pipeline; Step S32: Obtain the operating dataset of the integrated energy system in the j-th state. Divided into training and test sets, denoted as follows: , Among them, the training set By calculating the value under each pipe And the backpropagation mechanism is used to update the network weights of the neural network model of the physical information of each pipeline in the integrated energy system; while the test set This is used to verify the model training effect; Step S4 includes the following steps: Step S41: Based on the training task set constructed in step S1, train a meta-learning model of the integrated energy system pipeline network under different system states; the loss function of the meta-learning neural network model is: in, The training error of the meta-learning neural network is... , , Weights for different errors, Let the physical information of the i-th pipe be the neural network error. These represent the errors in the operating data of the equipment on the i-th pipeline and the terminal node of the i-th pipeline, respectively. and These are the predicted and actual values ​​of the operating data for the fluid flowing into the equipment from the i-th pipeline, respectively. and These are the predicted and actual values ​​of the operating data for the i-th pipeline fluid exiting the equipment side, respectively. and These are the predicted and actual values ​​of the operating data at the i-th pipeline terminal node, respectively. Step S42: Based on the constructed test task set, verify the effect of the trained meta-learning neural network under different system states.

2. The integrated energy system modeling method based on meta-learning and physical information neural networks according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Based on the conditions of the energy supply side and load side of the integrated energy system, as well as the topology of the pipeline network, the state of the integrated energy system is obtained; in the j-th state of the integrated energy system, the operating data of the i-th pipeline over time... and location The changes are represented as Then, in the j-th state of the integrated energy system, the operating dataset containing n pipelines is represented as follows: Simplified representation as ; Step S12: Based on the operational dataset of each pipeline in m different states of the integrated energy system, construct a task set, represented as follows: The task set is divided into a training task set and a test task set.

3. The integrated energy system modeling method based on meta-learning and physical information neural networks according to claim 1, characterized in that, In step S2: Based on the historical operating data of the pipelines connected to the equipment in the integrated energy system, a white-box model is constructed for pumps and valves, and a black-box model is constructed for heat exchangers and units; the white-box model or black-box model is represented by the following formula: in, For model output, For the constructed white-box model or black-box model, The input to the model is j, which represents the j-th state of the system.

4. The integrated energy system modeling method based on meta-learning and physical information neural networks according to claim 1, characterized in that, In step S5: When a trained and validated meta-learning network is used in a real integrated energy system, the current state of the system is input into the meta-learning neural network. The network can provide the physical information corresponding to each pipeline. All the parameter variables in the neural network that need to be set and defined by humans can be obtained, thereby achieving the modeling result with the least overall system error.

Citation Information

Patent Citations

  • Neural network initialization design method of embedded physical knowledge based on meta-learning

    CN114417497A

  • Pipeline leakage monitoring system based on deep learning and physical information fusion

    CN115264406A