A method for dynamic modeling and solving of multi-energy flow systems
By using dynamic modeling and neural network solution methods for multi-energy flow systems, the problem of inconsistent energy balance in multi-energy flow systems in industrial parks was solved, achieving efficient and safe operation of the system and providing a theoretical basis.
Patent Information
- Application Number
- CN202310379687.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-11
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-04-11
AI Technical Summary
In existing multi-energy flow systems in industrial parks, the mass, energy, and momentum balance characteristics of various energy carriers are inconsistent with the supply and demand balance characteristics, resulting in uncoordinated system operation and making it difficult to achieve efficient and safe energy efficiency and safety in synergy.
A dynamic modeling method for multi-energy flow systems is adopted. The transmission and distribution process of various energy carriers is described by a unified modeling equation set, and the solution is performed by combining a neural network model. A physical information neural network is established, and the loss term of the neural network is optimized by using the residual error in the time and frequency domain, so as to realize the dynamic process solution of multi-energy flow systems.
It provides a technical foundation for the efficient and safe design and operation of multi-energy flow systems, realizes the dynamic process solution of multi-energy flow systems, and improves the system's ability to coordinate energy efficiency and safety.
Smart Images

Figure CN116451394B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of integrated energy systems, specifically relating to a dynamic modeling and solution method for multi-energy flow systems. Background Technology
[0002] The multi-energy flow system in the industrial park includes various energy carriers, such as heat (cold / hot water), steam (high pressure / low pressure), natural gas, compressed air and steam.
[0003] In actual industrial parks with integrated production and daily life, energy consumption patterns are diverse. Daily life-related energy users generally rely on carriers to deliver energy to the end-user, while production-related energy users not only rely on energy carriers to deliver energy to the end-user, but also directly deliver and consume energy carriers (such as steam and compressed air) in production. The mass, energy, and momentum balance characteristics of the multiple energy sources involved in this process are inconsistent with the supply and demand balance characteristics of the existing system.
[0004] This invention proposes a dynamic modeling and solution method for energy flow systems, which promotes the collaborative development and safety analysis theory of multi-energy flow energy efficiency and safety. It helps to explore the dynamic evolution characteristics and synergistic laws of energy efficiency and safety of multi-energy flow systems, and provides a solid theoretical and technical foundation for establishing efficient and safe operation technologies and design methods for multi-energy flow systems in industrial parks. Summary of the Invention
[0005] To address the problems existing in the prior art, this invention provides a method for dynamic modeling and solving of multi-energy flow systems, comprising the following steps:
[0006] Step S1 involves dynamic modeling of the pipeline distribution process in a multi-energy flow system. The four types of carriers—heat (cold / hot water), steam (high / low pressure), natural gas, and compressed air—generally follow similar distribution equations and network topologies, exhibiting many commonalities in their distribution mechanism models. During multi-energy flow distribution, a unified set of modeling equations can be derived to express the spatiotemporal transport and distribution patterns of common physical quantities in the distribution processes of these carriers.
[0007]
[0008]
[0009]
[0010] In the formula, t and x represent time and distance, respectively. These are vectors composed of physical quantities whose partial derivatives with respect to time and space are required. Let ρ, v, u, h, P, and T be a vector composed of physical quantities that do not require partial differentials, where ρ, v, u, h, P, and T are the density, velocity, internal energy, enthalpy, pressure, and temperature of the energy flux; g is the local gravitational acceleration; θ, λ, and D are the horizontal inclination angle, friction coefficient, and inner diameter of the pipe; K is the heat transfer coefficient; and T is the pressure. r Let P be the ambient temperature, and ρ(P,T) and u(P,T) represent the density ρ and internal energy u, respectively, which are functions of pressure P and temperature T.
[0011] Step S2 involves constructing the topology of the multi-energy flow system and combining it with the modeling results from step S1 to obtain the dynamic model of the multi-energy flow system. The multi-energy flow network system can be viewed as a mesh system composed of several pipe segments interconnected according to a specific topology. Nodes for flow entry, exit, aggregation, and distribution are set up within the system. The connecting pipes between nodes are called pipe segments, and the basic loops in the system are called loops. Finally, the topology of the multi-energy flow system is analyzed and described using matrix representation.
[0012] For energy flow i, the topology of the energy flow network is represented by the node-branch incidence matrix as follows:
[0013]
[0014] Where, matrix A i The rows in the multi-energy flow system represent nodes of energy flow i, and the columns represent pipe segments of energy flow i in the multi-energy flow system, and:
[0015]
[0016] For energy flow i, the topology of the energy flow network can be represented by the basic loop matrix as follows:
[0017]
[0018] Wherein, matrix B i In the multi-energy flow system, the rows represent the basic loops of energy flow i, and the columns represent the pipe segments of energy flow i in the multi-energy flow system, and:
[0019]
[0020] In a multi-energy flow system, the net mass flow rate entering and exiting each node, the mass flow rate of each pipe section, the total resistance pressure drop, and the closure error of each basic loop are all represented in vector form to facilitate the calculation of the multi-energy flow system.
[0021] The net mass flow rate entering and exiting each node in the multi-energy flow system is:
[0022] q = [q1, q2, ..., q n ] T
[0023] Among them, element qn Let n represent the net mass flow rate of node n, and:
[0024]
[0025] The mass flow rate of each pipe section in the multi-energy flow system is:
[0026] Q = [Q1, Q2, ..., Q m ] T
[0027] Among them, element Q m This represents the mass flow rate of pipe segment m.
[0028] The total resistance pressure drop across all pipe sections in a multi-energy flow system is:
[0029] ΔP=[ΔP1,ΔP2,…,ΔP m ] T
[0030] Among them, element ΔP m This represents the total resistance loss of pipe section m, which mainly consists of flow resistance, gravity pressure drop, and dynamic pressure drop.
[0031] The closure error of each basic loop in a multi-energy flow system is:
[0032] ΔH=[ΔH1,ΔH2,…,ΔH s ] T
[0033] Among them, element ΔH s This represents the resistance closure difference on the basic loop s.
[0034] Step S3: Establish a general equivalent physical information neural network model, determine the neural network loss term by combining the time-frequency domain residual error, and optimize the solution. Based on the optimized neural network model, solve the dynamic model of the multi-energy flow system. This invention will construct a physical information neural network corresponding to the transport equations using the following method. This method includes three main aspects: setting the neural network input and output, establishing the physical information neural network structure, and determining and optimizing the neural network loss term.
[0035] Configure the neural network input and output. Choose vector X as the input, where X is a vector of length 2 containing distance x and time t. The output is vector U containing the operating conditions and differential information.
[0036] A feedforward fully connected deep neural network structure is established, where the input vector X has a length of 2, the network contains n1 hidden layers, each hidden layer contains n2 neurons, and the output vector U has a length equal to the number of operating states included. The neural network mapping of output U with respect to input X is obtained as follows:
[0037] U p =w L σ L (w L-1 σ L-1 (…w 2 σ 2 (w 1 σ 1 (w 0 X+b 0 )+b 1 )...)+b L-1 )+b L
[0038] In the formula, U p The neural network predicts the vector; i1 = 0, 1, ..., L, where i1 is the weight coefficient of the neural network; j1 = 0, 1, ..., L, is the neural network bias vector; k1 = 1, ..., L, are the activation functions of the neural network, including but not limited to the sigmoid function, tanh function and relu function.
[0039] Based on the dynamic model of the multi-energy flow system, the physical property equations and operating parameters are selected according to the actual energy flow. The algebraic terms and differential terms are decomposed, and the physical information neural network structure is formulated according to the dynamic model equation set structure. A two-layer network structure (i.e., feedforward fully connected deep neural network structure) can be established, which includes mapping from time and space to operating parameters and from operating parameters to residual errors. Finally, the total output including the operating state can be obtained.
[0040] The loss function is optimized using a stochastic gradient descent algorithm with automatic differentiation and backpropagation, and the automatic differentiation of the loss function is achieved through the chain rule:
[0041]
[0042] In the formula, y and z are functions of the independent variable x.
[0043] Using the backpropagation method with automatic differentiation, the first derivative of the loss function with respect to the parameter vector is obtained:
[0044]
[0045] By combining the stochastic gradient descent algorithm, we obtain the values of the parameter vector when the loss function is close to its minimum:
[0046]
[0047] In the formula, ω is the learning rate, and n is the number of gradient descent operations.
[0048] Ultimately, this minimizes the loss function value, thus completing the training process of the neural network.
[0049] Time-frequency domain residual error control is combined with the solution. The neural network loss term is established and optimized; the total loss function of energy flow i can be expressed as:
[0050]
[0051] In the formula, Let be the loss function for the i-th type of energy flow; This is the residual loss term for the multi-energy flow dynamic model; The loss term generated by the boundary conditions; The loss term generated by the initial conditions; ξ represents the error loss term of the neural network itself, α, β, γ, and ξ are weighting coefficients, and k is the k-th equation among the three equations of mass, momentum, and energy.
[0052] The boundary and initial conditions of each energy flow node in the overall system are diverse, and the changes of some parameters may be periodic or impulsive. To ensure that the neural network model is equivalent to the actual multi-energy flow system, this invention introduces the boundary and initial errors of the solution in addition to the residuals of the solution equations when training the neural network.
[0053] To ensure the model's universality under different conditions, this invention calculates the time-by-time function error in the time domain and the amplitude error in the frequency domain simultaneously when calculating the residual loss, boundary condition loss, initial condition loss, and the neural network's own error loss.
[0054] The various loss terms in the total loss function of energy flow i can be expressed as:
[0055]
[0056] In the formula, These can represent residual losses respectively. Boundary condition loss Initial condition loss Or the error loss of the neural network itself Any one of them; N0 is the length of the time-domain sequence; This represents the predicted function value of the k-th equation in the time domain; This represents the actual function value of the k-th equation in the time domain; f is the length of the frequency domain sequence; f is the frequency domain component. This represents the prediction function value of the k-th equation in the frequency domain; This represents the actual function value of the k-th equation in the frequency domain.
[0057] Finally, by using historical data to train a neural network model and minimizing the loss term, the training of the neural network model can be completed, resulting in a neural network model suitable for the dynamic model of multi-energy flow systems. This model can solve the dynamic process of multi-energy flow in a certain pipe section, and by combining the topology of the multi-energy flow system, it can solve the entire multi-energy flow system.
[0058] The beneficial effects of this invention are as follows:
[0059] This invention establishes a physical information neural network model based on the dynamic model equations of a multi-energy flow system. By training the neural network model with the time-frequency domain characteristics of the integrated residuals and boundary conditions, the dynamic process of the entire multi-energy flow system can be solved, providing a theoretical basis for efficient and safe design methods and operation technologies for multi-energy flow systems. Attached Figure Description
[0060] Figure 1 This is a flowchart illustrating the steps of a dynamic modeling and solution method for a multi-energy flow system according to the present invention.
[0061] Figure 2 This is a technical roadmap for the research on dynamic model construction, topological structure matrix, and physical information neural network of this invention.
[0062] Figure 3 This is a diagram of the physical information neural network model of the present invention. Detailed Implementation
[0063] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0064] like Figure 1 As shown, this invention discloses a method for dynamic modeling and solving of multi-energy flow systems, including the following steps:
[0065] Step S1: Dynamic modeling of the pipeline transmission and distribution process in a multi-energy flow system;
[0066] Step S2: Construct the topology of the multi-energy flow system and combine it with the modeling results from step S1 to obtain the dynamic model of the multi-energy flow system.
[0067] Step S3: Establish a general equivalent physical information neural network model, determine the neural network loss term by combining the time-frequency domain residual error and optimize the solution, and solve the dynamic model of the multi-energy flow system based on the optimized neural network model.
[0068] Figure 2 This invention demonstrates the technical approach to dynamic model construction, topological structure matrix, and physical information neural network research.
[0069] In step S1,
[0070] In multi-energy flow transmission and distribution processes, the four types of carriers—heat (cold / hot water), steam (high / low pressure), natural gas, and compressed air—generally follow similar transmission and distribution equations and network topologies, and their transmission and distribution mechanism models share many similarities. A unified set of modeling equations can be derived for multi-energy flow transmission and distribution processes to express the spatiotemporal transport and distribution patterns of common physical quantities in these carriers.
[0071]
[0072]
[0073]
[0074] In the formula, t and x represent time and distance, respectively. These are vectors composed of physical quantities whose partial derivatives with respect to time and space are required. Let ρ, v, u, h, P, and T be a vector composed of physical quantities that do not require partial differentials, where ρ, v, u, h, P, and T are the density, velocity, internal energy, enthalpy, pressure, and temperature of the energy flux; g is the local gravitational acceleration; θ, λ, and D are the horizontal inclination angle, friction coefficient, and inner diameter of the pipe; K is the heat transfer coefficient; and T is the pressure. r Let P be the ambient temperature, and ρ(P,T) and u(P,T) represent the density ρ and internal energy u, respectively, which are functions of pressure P and temperature T.
[0075] In step S2,
[0076] Step S21: The multi-energy flow network system can be regarded as a mesh system composed of several pipe segments connected to each other according to a specific topology. Nodes for flow entry, exit, collection and distribution are set in the system, the connecting pipes between nodes are pipe segments, and the basic loops in the system are called loops. Finally, the topology of the multi-energy flow system is analyzed and described by matrix representation method.
[0077] Step S22, for energy flow i, the topology of the energy flow network is represented by the node-branch correlation matrix as follows:
[0078]
[0079] Where, matrix A i The rows in the multi-energy flow system represent nodes of energy flow i, and the columns represent pipe segments of energy flow i in the multi-energy flow system, and:
[0080]
[0081] Step S23, for energy flow i, the topology of the energy flow network is represented by the basic loop matrix as follows:
[0082]
[0083] Wherein, matrix B i In the multi-energy flow system, the rows represent the basic loops of energy flow i, and the columns represent the pipe segments of energy flow i in the multi-energy flow system, and:
[0084]
[0085] Step S24, the net mass flow rate entering and exiting each node in the multi-energy flow system is:
[0086] q = [q1, q2, ..., q n ] T
[0087] Among them, element q n Let n represent the net mass flow rate of node n, and:
[0088]
[0089] The mass flow rate of each pipe section in the multi-energy flow system is:
[0090] Q = [Q1, Q2, ..., Q m ] T
[0091] Among them, element Q m This represents the mass flow rate of pipe segment m.
[0092] The total resistance pressure drop across all pipe sections in a multi-energy flow system is:
[0093] ΔP=[ΔP1,ΔP2,…,ΔP m ] T
[0094] Among them, element ΔP m This represents the total resistance loss of pipe section m, which mainly consists of flow resistance, gravity pressure drop, and dynamic pressure drop.
[0095] The closure error of each basic loop in a multi-energy flow system is:
[0096] ΔH=[ΔH1,ΔH2,…,ΔH s ] T
[0097] Among them, element ΔH s This represents the resistance closure difference on the basic loop s.
[0098] like Figure 3 The diagram shows the physical information neural network model of the present invention. In step S3,
[0099] Step S31: The present invention will construct a physical information neural network corresponding to the transport equations using the following method. This method includes three main aspects: setting the neural network input and output, establishing the physical information neural network structure, and establishing and optimizing the neural network loss term.
[0100] Configure the neural network input and output. Choose vector X as the input, where X is a vector of length 2 containing distance x and time t. The output is vector U containing the operating conditions and differential information.
[0101] A feedforward fully connected deep neural network structure is established, where the input vector X has a length of 2, the network contains n1 hidden layers, each hidden layer contains n2 neurons, and the output vector U has a length equal to the number of operating states included. The neural network mapping of output U with respect to input X is obtained as follows:
[0102] U p =w L σ L (w L-1 σ L-1 (…w 2 σ 2 (w 1 σ 1 (w 0 X+b 0 )+b 1 )...)+b L-1 )+b L
[0103] In the formula, U p The neural network predicts the vector; i1 = 0, 1, ..., L, where i1 is the weight coefficient of the neural network; j1 = 0, 1, ..., L, is the neural network bias vector; k1 = 1, ..., L, are the activation functions of the neural network, including but not limited to the sigmoid function, tanh function and relu function.
[0104] Based on the dynamic model of the multi-energy flow system, the physical property equations and operating parameters are selected according to the actual energy flow. The algebraic terms and differential terms are decomposed, and the physical information neural network structure is formulated according to the dynamic model equation set structure. A two-layer network structure (i.e., feedforward fully connected deep neural network structure) can be established, which includes mapping from time and space to operating parameters and from operating parameters to residual errors. Finally, the total output including the operating state can be obtained.
[0105] Step S32 involves optimizing the loss function using a stochastic gradient descent algorithm with automatic differentiation and backpropagation to minimize its value, thereby completing the training process of the neural network. The automatic differentiation of the loss function is achieved using the chain rule.
[0106]
[0107] Using the backpropagation method with automatic differentiation, the first derivative of the loss function with respect to the parameter vector is obtained:
[0108]
[0109] By combining the stochastic gradient descent algorithm, we obtain the values of the parameter vector when the loss function is close to its minimum:
[0110]
[0111] In the formula, ω is the learning rate, and n is the number of gradient descent operations.
[0112] After establishing the neural network loss term and performing optimization, the total loss function of energy flow i can be expressed as:
[0113]
[0114] In the formula, Let be the loss function for the i-th type of energy flow; This is the residual loss term for the multi-energy flow dynamic model; The loss term generated by the boundary conditions; The loss term generated by the initial conditions; ξ represents the error loss term of the neural network itself, α, β, γ, and ξ are weighting coefficients, and k is the k-th equation among the three equations of mass, momentum, and energy.
[0115] The boundary and initial conditions of each energy flow node in the overall system are diverse, and the changes of some parameters may be periodic or impulsive. To ensure that the neural network model is equivalent to the actual multi-energy flow system, this invention introduces the boundary and initial errors of the solution in addition to the residuals of the solution equations when training the neural network.
[0116] To ensure the model's universality under different conditions, this invention calculates the time-by-time function error in the time domain and the amplitude error in the frequency domain simultaneously when calculating the residual loss, boundary condition loss, initial condition loss, and the neural network's own error loss.
[0117] The various loss terms in the total loss function of energy flow i are expressed as follows:
[0118]
[0119] In the formula, These can be expressed as residual losses respectively. Boundary condition loss Initial condition loss and the error loss of the neural network itself N0 is the length of the time-domain sequence; This represents the predicted function value of the k-th equation in the time domain; This represents the actual function value of the k-th equation in the time domain; f is the length of the frequency domain sequence; f is the frequency domain component. This represents the prediction function value of the k-th equation in the frequency domain; This represents the actual function value of the k-th equation in the frequency domain.
[0120] Finally, by using historical data to train a neural network model and minimizing the loss term, the training of the neural network model can be completed, resulting in a neural network model suitable for the dynamic model of multi-energy flow systems. This model can solve the dynamic process of multi-energy flow in a certain pipe section. By combining the topology of the multi-energy flow system, the model can solve the entire multi-energy flow system. It can obtain the implicit neural network solution of the dynamic process of heat (cold water / hot water), steam (high pressure / low pressure), natural gas and compressed air coupling in the multi-energy flow system of the industrial park, and realize the output of dynamic operating condition information.
[0121] It should be noted that the above description is only a part of the embodiments of the present invention, and all equivalent changes made to the methods described in this invention are included within the protection scope of this invention. Those skilled in the art can make similar substitutions to the specific examples described, as long as they do not deviate from the structure of the invention or exceed the scope defined in these claims, all of which fall within the protection scope of this invention.
Claims
1. A method for dynamic modeling and solving of multi-energy flow systems, characterized in that, Includes the following steps: Step S1: Dynamic modeling of the pipeline transmission and distribution process in a multi-energy flow system; Step S2: Construct the topology of the multi-energy flow system and combine it with the modeling results from step S1 to obtain the dynamic model of the multi-energy flow system. Step S3: Establish a general equivalent physical information neural network model, determine the neural network loss term by combining the time-frequency domain residual error and optimize the solution, and solve the dynamic model of the multi-energy flow system based on the optimized neural network model. The method of establishing the neural network loss term by combining the time-frequency domain residual error is as follows: The total loss function of energy flow i is expressed as: In the formula, Let be the loss function for the i-th type of energy flow; This is the residual loss term for the multi-energy flow dynamic model; The loss term generated by the boundary conditions; The loss term generated by the initial conditions; This is the error loss term of the neural network itself. , , , is the weighting coefficient; k is the k-th equation among the three equations of mass, momentum, and energy; When calculating the residual loss, boundary condition loss, initial condition loss, and neural network error loss, the time-by-time function error in the time domain and the magnitude error in the frequency domain are calculated simultaneously. Therefore, the various loss terms in the total loss function of energy flow i can be expressed as: In the formula, Residual loss Boundary condition loss Initial condition loss and the error loss of the neural network itself Any one of them; N0 is the length of the time-domain sequence; This represents the predicted function value of the k-th equation in the time domain; This represents the actual function value of the k-th equation in the time domain; The length of the frequency domain sequence; f is the frequency domain component; This represents the prediction function value of the k-th equation in the frequency domain. This represents the actual function value of the k-th equation in the frequency domain. By training a neural network model using historical data and minimizing the loss term, the training of the neural network model can be completed, resulting in a neural network model suitable for dynamic models of multi-energy flow systems. This model can solve the dynamic processes of multi-energy flow in a certain pipe section, and by combining the topology of the multi-energy flow system, it can solve the entire multi-energy flow system.
2. The method for dynamic modeling and solving of a multi-energy flow system according to claim 1, characterized in that, The specific steps of S1 are as follows: For the four types of carriers in a multi-energy flow system—heat, steam, natural gas, and compressed air—a unified set of modeling equations is derived based on mass, momentum, and energy equations to express the spatiotemporal transport and distribution patterns of common physical quantities during the transport and distribution of these four types of carriers. In the formula, t and x represent time and distance, respectively. , These are vectors composed of physical quantities whose partial derivatives with respect to time and space are required. Let ρ, v, u, h, P, and T be a vector composed of physical quantities that do not require partial differentials, where ρ, v, u, h, P, and T are the density, velocity, internal energy, enthalpy, pressure, and temperature of the energy flux; g is the local gravitational acceleration; θ, λ, and D are the horizontal inclination angle, friction coefficient, and inner diameter of the pipe; K is the heat transfer coefficient; and T is the pressure. r For ambient temperature, and Let ρ and u represent density ρ and internal energy u as functions of pressure P and temperature T, respectively.
3. The method for dynamic modeling and solving of a multi-energy flow system according to claim 2, characterized in that, In step S2: A multi-energy flow network system is viewed as a mesh system consisting of several pipe segments interconnected according to a specific topology. Nodes for flow entry, exit, aggregation, and distribution are set up in the system, the connecting pipes between nodes are pipe segments, and the basic loops in the system are called loops. The topology of the multi-energy flow system is represented by a matrix. For energy flow i, the topology of the energy flow network is represented by the node-branch incidence matrix as follows: Among them, matrix The rows in the multi-energy flow system represent the nodes of energy flow i, and the columns represent the pipe segments of energy flow i in the multi-energy flow system, and: For energy flow i, the topology of the energy flow network can be represented by the basic loop matrix as follows: Among them, matrix In the multi-energy flow system, the rows represent the basic loops of energy flow i, and the columns represent the pipe segments of energy flow i in the multi-energy flow system, and: The net mass flow rate entering and exiting each node in a multi-energy flow system, the mass flow rate of each pipe section, the total resistance pressure drop, and the closure error of each basic loop are all represented in vector form.
4. The method for dynamic modeling and solving of a multi-energy flow system according to claim 3, characterized in that, The specific steps of step S3 are as follows: A general equivalent physical information neural network model is established. The model includes a feedforward fully connected deep neural network structure that maps from time and space to working parameters and from working parameters to residual errors, thereby obtaining the total output containing the working state. A physical information neural network model corresponding to the transport equations of the dynamic model of the multi-energy flow system is constructed. The specific method is as follows: set the input and output of the neural network model, establish the physical information neural network structure, and determine the neural network loss term by combining the time-frequency domain residual error and optimize the solution.
5. The method for dynamic modeling and solving of a multi-energy flow system according to claim 4, characterized in that, The specific method for setting the input and output of the neural network model and establishing the physical information neural network structure is as follows: Set the input and output of the neural network model. Select vector X as the input of the neural network model. X is a vector containing distance x and time t. The output is vector U containing operating conditions and differential information. U includes the density, flow rate, internal energy, enthalpy, pressure and temperature operating conditions of the multi-energy flow system carrier. Based on the dynamic model of a multi-energy flow system, the physical property equations and operating parameters are selected according to the actual energy flow. Algebraic and differential terms are decomposed, and a physical information neural network structure is formulated according to the dynamic model equation set structure. The physical information neural network structure is a feedforward fully connected deep neural network structure, where the input vector X has a length of 2, the network contains n1 hidden layers, each hidden layer contains n2 neurons, and the output vector U has a length equal to the number of operating states it contains. The neural network mapping of output U with respect to input X is obtained as follows: In the formula, The neural network predicts the vector; i1=0,1,…L, are the weight coefficients of the neural network; j1=0,1,…L, is the neural network bias vector; k1=1,…L, are the activation functions of the neural network, including but not limited to the sigmoid function, tanh function and relu function.
Citation Information
Patent Citations
Dynamic modeling simulation method and system for steam heating system
CN114912262A
A Dynamic Modeling Method for Natural Gas Systems
CN114936522A
