Electricity-hydrogen-heat system integrated operation management and control method

Abstract

The invention discloses an electricity-hydrogen-heat system integrated operation management and control method. The method comprises the following steps: (1) constructing an electricity-hydrogen-heat system integrated optimal operation model; (2) constructing an electricity-hydrogen-heat neural network model for the electricity-hydrogen conversion device; (3) equivalently embedding the neural network model into the optimization problem; and (4) solving an optimization problem. Based on the constraint learning method, the scheduling decision quality of the integrated operation of the electricity-hydrogen-heat system is improved, so that the safety and economy of the operation are improved.

Description

Technical Field

[0001] The present invention relates to the field of hydrogen storage fuel cells, and particularly to a method for integrated operation control of an electric-hydrogen-thermal system. Background Art

[0002] With the increasing demand for the consumption of clean fuels and renewable electric energy in recent years, electric-hydrogen conversion devices such as hydrogen electrolyzers and hydrogen fuel cells have achieved rapid technological development and wide application. Such devices can achieve flexible "bidirectional electric-hydrogen conversion". At the same time, a large amount of heat energy is generated during the process. On the one hand, if the temperature of the device is not managed, inappropriate operating temperatures will seriously reduce the operating efficiency of the device and even pose safety risks; on the other hand, the generated heat energy can meet certain heat demand through waste heat utilization, further improving the economic efficiency of system operation. In summary, the integrated operation control of the electric-hydrogen-thermal energy system is crucial for the safety and economic efficiency of system operation.

[0003] Currently, the existing technical solutions for integrated electric-hydrogen-thermal control can be mainly divided into two categories:

[0004] One is the model-based optimization method, which aims to minimize the system operation cost, constructs an optimization problem that satisfies the operation constraints of each energy component, and uses an optimization solver to solve for the scheduling instructions. For the electric-hydrogen-thermal coupling problem, the existing solutions have constructed a "Power-Temperature-Hydrogen" (P-T-H) mechanism model around the relationships among electric power, electrolyzer / fuel cell temperature, and hydrogen production / consumption rate. Considering that the significant nonlinearity of the P-T-H model is difficult to directly handle by the solver, the mainstream solutions use (piecewise) linearization methods to approximate it. For example, the P-T-H model surface is approximated by several triangular planes, and an iterative verification method is used to control the approximation error. Based on the above methods, scheduling methods for electric-hydrogen-thermal coupling energy systems of different scales, such as at the site and distribution network levels, have been proposed respectively. However, the piecewise linearization method for the P-T-H model introduces a large number of binary decision variables into the optimization problem, resulting in a relatively low computational efficiency for the obtained mixed-integer optimization problem.

[0005] In addition, limited by the representation ability of the mechanism model, the existing control technologies based on optimization methods have insufficient fineness in depicting the heat transfer process, and do not fully consider the influence of factors such as environmental temperature and the power of heat dissipation equipment on the temperature of the electro-hydrogen conversion device. Especially considering the significant diurnal temperature difference in inland areas with rich renewable resources, ignoring the influence of environmental temperature will bring serious modeling deviations. This process is also significantly non-linear. Moreover, the actual environmental temperature is difficult to accurately estimate at the day-ahead decision-making stage, and the uncertainty of the prediction deviation threatens the safety and economy of the operation of the energy system. Although uncertainty optimization methods such as stochastic programming and robust optimization provide reliable and economic energy control methods for uncertainty, these methods often rely on specific mathematical problem forms or require a large amount of calculations. The significant non-linearity and computational cost of the P-T-H model and the heat transfer process limit the integrated application of these methods.

[0006] Another type of technology is based on model-free decision-making methods, represented by reinforcement learning. By simulating or physically constructing the operation environment of the electro-hydrogen-thermal energy system, through the interaction between the agent's decision-making and the environment, the rewards of operation strategies under various working conditions are recorded and used to train the artificial neural network for generating electro-hydrogen-thermal operation control decisions. Thanks to the advantages of being model-free, this solution can consider multiple factors, such as environmental temperature and energy source-load uncertainty. However, on the one hand, the sample efficiency of the model-free decision-making method is low, and the agent needs to interact with the environment a large number of times to consider various factors, resulting in high training costs; on the other hand, the model-free method has poor interpretability, and the reliability and optimality lack theoretical guarantees. Especially considering the flammable and explosive characteristics of hydrogen, its practical application is worrying.

[0007] In summary, among the existing technical solutions for the integrated operation control of electro-hydrogen-thermal energy systems, the optimization methods of the models have insufficient adaptability and are difficult to comprehensively consider complex heat processes, environmental uncertainty factors, etc. The approximated and simplified models reduce the safety and economy of the integrated operation of electro-hydrogen-thermal energy systems. At the same time, although the model-free decision-making methods can consider the above factors, their high training costs and lack of reliability limit their applications. Therefore, it is urgent to construct a new electro-hydrogen-thermal energy control paradigm to break through the deficiencies of the existing technical framework. Summary of the Invention

[0008] In view of the above-mentioned defects of the existing technology, the technical problem to be solved by the present invention is the low safety and low economy of the integrated operation of the traditional electro-hydrogen-thermal energy system.

[0009] To achieve the above object, the present invention provides a method for integrated operation control of an electro-hydrogen-thermal system, including:

[0010] (1) Construction of an integrated optimal operation model for the electro-hydrogen-thermal system;

[0011] (2) Construction of an electro-hydrogen-thermal neural network model for the electro-hydrogen conversion device;

[0012] (3) Equivalent embedding of the neural network model into the optimization problem;

[0013] (4) Solving the optimization problem.

[0014] Existing integrated operation and control technology methods for electro-hydrogen-thermal energy systems are difficult to balance modeling flexibility and decision reliability. Specifically, model-based optimization methods have high requirements for the mathematical form of the model and are difficult to achieve fine modeling considerations for complex non-linear factors such as the electro-hydrogen-thermal coupling relationship and heat transfer process; while model-free operation and control methods face problems such as high training costs and poor decision reliability.

[0015] The present invention first proposes an integrated operation and control scheme for electro-hydrogen-thermal systems based on constraint learning technology; by equivalently embedding the trained neural network into the optimization problem, the scheme realizes the decoupling of complex non-linear mechanism description and optimization problem modeling, breaks through the mathematical form of existing mechanism modeling, and improves the problem of complex non-linear mechanism in the electrolytic hydrogen production process on decision accuracy.

[0016] For the integrated operation and control task of the electro-hydrogen-thermal system provided by the present invention, compared with model-based optimization methods, on the one hand, by finely constructing the influence of factors such as ambient temperature and radiator power on temperature, the modeling accuracy of the temperature of the electro-hydrogen conversion device is improved, avoiding the self-safety risk of equipment temperature exceeding the limit caused by modeling errors. On the other hand, the present invention realizes the estimation of composite factors from operation decisions to states such as electro-hydrogen conversion efficiency, which can not only keep the electro-hydrogen conversion device operating under high conversion efficiency conditions, but also achieve good coordination among all links of the energy system, significantly improving the system operation efficiency.

[0017] In addition, compared with model-free operation and control methods, the scheme proposed by the present invention obtains decisions by means of an optimization solver, avoiding high training and development costs. At the same time, the decision optimality and constraint reliability of the method proposed by the present invention are guaranteed.

[0018] Further, step (1) specifically includes the following steps:

[0019] S1. Determine the scale and operation mode of the electro-hydrogen-thermal system;

[0020] S2. Determine the operation control objective function;

[0021] S3. Determine the operation constraints, where the operation constraints include power-side operation constraints, heat-side operation constraints, and operation constraints of the electrolytic hydrogen production device and the hydrogen energy side.

[0022] Furthermore, the operating constraints on the power side include:

[0023]

[0024] E0≤E T

[0025]

[0026] where respectively represent the electricity quantities purchased from the power grid, generated by renewable energy, released by the electrical energy storage, charged into the electrical energy storage, consumed by the electrical load, and consumed by electrolytic hydrogen production at time t in the system; respectively represent the electricity quantity stored in the electrical energy storage, and the energy efficiency of the electrical energy storage for storing and releasing electrical energy at time t.

[0027] Furthermore, the operating constraints on the heat side include:

[0028]

[0029] where β boil respectively represent the heat generated, fuel consumed, and conversion efficiency of the boiler at time t; and respectively represent the heat load and the heat utilized from the waste heat of electrolytic hydrogen production in the system at time t.

[0030] Furthermore, the electrolytic hydrogen production device and the operating constraints on the hydrogen energy side include:

[0031]

[0032]

[0033] where respectively represent the electrolyzer temperature, ambient temperature, total power consumption of the electrolyzer, heat generation of the electrolyzer, heat dissipation of the electrolyzer, and heat utilized from the waste heat of the electrolyzer at time t; and α Hy respectively represent the hydrogen production amount and the hydrogen energy unit conversion coefficient; respectively represent the hydrogen production amount of electrolytic hydrogen production, hydrogen load amount, and absorption / release amount of the hydrogen storage device at time t; η HS，in 、η HS，out respectively represent the hydrogen storage amount and the energy efficiency of hydrogen storage charging / discharging at time t.

[0034] Step (1) of the present invention constructs an electric-hydrogen-heat combined operation model, which can achieve an optimal operation decision considering multiple energy media of electricity, hydrogen, and heat, and provides a scheduling model basis for the waste heat utilization of electrolytic hydrogen production and the temperature control of the electrolyzer.

[0035] Furthermore, the specific steps of step (2) include:

[0036] S1. Obtain the electrolytic hydrogen production temperature and efficiency model samples;

[0037] S2. Construct a data set;

[0038] S3. Train a neural network.

[0039] In the hourly operation and control of traditional electric-hydrogen-thermal systems, the thermal dynamic process is insufficiently considered. Due to the slow heat transfer and temperature change processes and large time constants, the system operating conditions will not reach the adjustment target instantaneously, resulting in large scheduling deviations and reducing the operating efficiency. Although in other optimization fields, there are works that integrate dynamic equations into optimization problems based on the finite element difference method, the piecewise linearization of the electric-hydrogen-thermal model introduces a large number of binary variables, and the combination of the two will bring a serious computational burden.

[0040] In the present invention, the integration of the dynamic process of temperature change and the non-linear relationship of electric-hydrogen-thermal is realized through a neural network. The computational efficiency problem introduced by solving dynamic equations is avoided. At the same time, the present invention supports the further acceleration by special structures such as neural network pruning, sparsification, and convex input neural networks. At the same time, the trade-off between computational efficiency and modeling accuracy can be flexibly adjusted by the scale of the neural network.

[0041] The neural network modeling and corresponding constraint learning method adopted in the present invention, in addition to improving the model accuracy, significantly improves the computational efficiency and reduces the computational overhead compared with the existing mechanism methods. The method can be deployed on a smaller computing power platform, such as an edge computing device or a private host, without relying on a large computing power platform.

[0042] Furthermore, the specific way to obtain the electrolytic hydrogen production temperature and efficiency model samples in S1 is as follows:

[0043] S1.1. Electrolyzer temperature dynamic model;

[0044] Assume that the device operates at a constant power within the scheduling period, and for the electrolyzer heat dissipation power and the heat power utilized from the electrolyzer waste heat there is

[0045]

[0046]

[0047] where τ represents the time of a scheduling time slot;

[0048] The dynamic temperature change of the electrolyzer follows the following differential equation:

[0049]

[0050] Among them, C t is the lumped thermal resistance;

[0051] S1.2, Calculation of hydrogen production heat in electrolytic hydrogen production based on the dynamic model;

[0052] For the electrolyzer, the electrolytic hydrogen production power and the electrolytic hydrogen production heat have the following relationship:

[0053]

[0054] Among them, is the electrolytic cell voltage, is the electrolytic hydrogen production power;

[0055] is the total heat production within the statistical period. Considering the dynamic process of temperature change, based on the dynamic finite element difference solution of temperature, calculate the heat production power at each moment and sum to calculate the total power value;

[0056]

[0057] Among them, τ represents the total time, and Δt represents the time step.

[0058] Furthermore, the construction of the dataset described in S2 specifically includes:

[0059] Randomly sample within the operating state range to be fitted to obtain different previous moment temperatures Ambient temperature Total power of the electrolyzer Heat dissipation of the electrolyzer Heat utilization of the waste heat of the electrolyzer Values;

[0060] Obtain the current moment temperature of the electrolyzer based on the S1 method Electrolytic hydrogen production heat As the sample label;

[0061] Repeat the operation to obtain data samples, thereby realizing the construction of the dataset.

[0062] Furthermore, the training of the neural network described in S3 includes: constructing an artificial neural network and realizing the fitting of the non-linear function through the dataset constructed in S2.

[0063] Furthermore, the equivalent embedding of the neural network model into the optimization problem in step (3) includes using the neural network model trained in step (2) to equivalently replace the non-linear function in the operation control optimization problem in step (1), and ensuring that the results are equivalent.

[0064] Furthermore, the non - linear function of the operation control optimization problem in step (1) of equivalent substitution is specifically the operation constraints of the electrolytic hydrogen production device and the hydrogen energy side, that is:

[0065]

[0066] Among them, respectively represent the electrolyzer temperature, ambient temperature, total power consumption of the electrolyzer, heat generation of the electrolyzer, heat dissipation of the electrolyzer, and heat utilization of the electrolyzer waste heat at time t.

[0067] Furthermore, in step (3), the process of equivalent embedding optimization for the multi - layer perceptron neural network based on ReLU includes:

[0068] The multi - layer perceptron is composed of multiple fully - connected layers. The operation result of a single neuron is as follows, and the neuron calculation is equivalently represented as a set of constraints:

[0069]

[0070] y ≤ x - M L (1 - z),

[0071] y ≤ M U z,

[0072] Where [L] and [N l respectively represent the number of layers of the neural network and the number of neurons in the l - th layer, and represent the output and input of the activation function of the i - th neuron in the l - th layer; and respectively represent the weight and bias from the i - th neuron in the (l - 1) - th layer to the current neuron; where M L < and M U >0 are respectively the lower and upper bounds of all possible x; z is a newly added auxiliary binary decision variable;

[0073] It also includes the following relationship:

[0074]

[0075]

[0076] Furthermore, in step (3), the equivalent embedding optimization problem of the integrated operation control method for the power - hydrogen - heat system is integrally represented as:

[0077]

[0078] The temperature measurement method provided by the present invention has the following technical effects:

[0079] The present invention improves the scheduling decision-making quality of the integrated operation of the electricity-hydrogen-heat system, thereby enhancing the operation safety and economy. This is because the solution provided by the present invention breaks through the limitations of traditional mechanism models, refines the characterization and consideration of significant non-linear factors such as the electricity-hydrogen conversion heat process in the decision-making stage, and thus obtains reliable and high-performance decisions for the integrated operation and control of the electricity-hydrogen-heat system. On this basis, the present invention takes into account the volatility of the ambient temperature and constructs a neural network mechanism characterization model based on uncertainty measurement, ensuring the operation reliability of the electricity-hydrogen-heat system under the prediction deviation of the ambient temperature.

[0080] Compared with the prior art, the practicability of the present invention is as follows:

[0081] The technical solution of the present invention realizes low-cost batch deployment and migration. Thanks to the advantages of the optimal constraint learning framework, the neural network model is continuously updated and real-time embedded into the optimization problem to be integrated with the operation and control problem. With the continuous enrichment of operation data during the research and deployment process of the electricity-hydrogen conversion equipment, the system model can be adaptively updated, thereby continuously improving the decision-making performance of the integrated operation and control of the electricity-hydrogen-heat system.

[0082] The following will further illustrate the concept, specific structure and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 is a flowchart of a method for integrated operation and control of an electricity-hydrogen-heat system according to a preferred embodiment of the present invention;

[0084] Figure 2 is a schematic diagram of the composition of an electricity-hydrogen-heat system according to a preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0085] The following introduces multiple preferred embodiments of the present invention with reference to the accompanying drawings of the specification, making its technical content clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0086] In the accompanying drawings, components with the same structure are denoted by the same numerical reference signs, and components with similar structures or functions everywhere are denoted by similar numerical reference signs. The size and thickness of each component shown in the drawings are arbitrarily shown, and the present invention does not limit the size and thickness of each component. In order to make the drawings clearer, the thickness of some parts in the drawings is appropriately exaggerated.

[0087] As Figure 1 shown, the present invention provides a method for integrated operation and control of an electricity-hydrogen-heat system, including:

[0088] (1) Construction of an integrated optimal operation model for the electricity - hydrogen - heat system;

[0089] (2) Construction of an electricity - hydrogen - heat neural network model for the electro - hydrogen conversion device;

[0090] (3) Equivalent embedding of the neural network model into the optimization problem;

[0091] (4) Solving the optimization problem.

[0092] As Figure 2 shown, it is a schematic diagram of the system composition for the 24 - hour operation scheduling of the site - level electricity - hydrogen - heat system. On the power side, renewable energy and grid - purchased electricity jointly supply the electricity for electrolytic hydrogen production and electrical loads, and an electrical energy storage device is equipped. The system operation control must maintain the hourly power supply - demand balance. A part of the electricity used for hydrogen production is converted into hydrogen energy, and a part is converted into heat energy. In addition, the electrolytic hydrogen production device is equipped with an actively controllable heat dissipation device to ensure equipment safety and improve operation efficiency. On the hydrogen energy side, a hydrogen storage tank is equipped and finally supplies hydrogen loads. On the heat energy side, fuel generates heat through a boiler, and together with the heat energy generated by electrolytic hydrogen production, it supplies heat loads.

[0093] Furthermore, in step (1) of constructing the integrated optimal operation model for the electricity - hydrogen - heat system, a basic decision - making framework for the integrated operation control of the electricity - hydrogen - heat system is constructed, and the system operation objectives, operation control objects, and operation restrictions are clarified, that is, the optimization objective, decision variables, and constraints of the optimization problem. It mainly includes the following steps:

[0094] S1. Determine the scale and operation mode of the electricity - hydrogen - heat system;

[0095] S2. Determine the operation control objective function; the operation control objective is to minimize the economic cost, that is:

[0096]

[0097] where the first term and the second term represent the electricity purchase cost and the fuel purchase cost respectively. c t respectively represent the electricity purchase quantity, fuel purchase quantity, and corresponding prices at time t.

[0098] S3. Determine the operation constraints. Specifically:

[0099] The operation constraints on the power side include:

[0100]

[0101] E0 ≤ E T (4)

[0102]

[0103] Constraint (2) restricts the balance between the supply and demand of system power, where respectively represent the electricity purchased from the power grid, generated by renewable energy, released from the electrical energy storage, charged into the electrical energy storage, consumed by the power load, and consumed for electrolytic hydrogen production at time t of the system. Constraint (3) represents the change in the electrical energy storage over time, where respectively represent the electricity stored in the electrical energy storage at time t, the energy efficiency of the electrical energy storage for storing and releasing electrical energy. Constraint (4) ensures that the electrical energy storage is not less than that at the start of the day after a day's operation, so as to ensure that the system can continue to operate in the next few days. Constraints (5)-(9) respectively define the upper and lower bounds of the electrical energy storage capacity, the electricity purchased, the electricity consumed for electrolytic hydrogen production, the electricity charged into the energy storage, and the electricity released from the energy storage.

[0104] The operating constraints on the thermal energy side include:

[0105]

[0106]

[0107] Constraint (10) describes the fuel and thermal energy conversion efficiency of the boiler, where respectively represent the heat generated, the fuel consumed, and its conversion efficiency by the boiler at time t. Constraint (11) ensures the balance between the supply and demand of the system's thermal energy, where and respectively represent the heat load and the heat utilized from the waste heat of electrolytic hydrogen production at time t of the system. Constraints (12) and (13) respectively define the upper and lower bounds of the heat generated by the boiler and the heat utilized from the waste heat of electrolytic hydrogen production.

[0108] The electrolytic hydrogen production device and the operating constraints on the hydrogen energy side include:

[0109]

[0110]

[0111]

[0112] Constraint (14) describes the change in the temperature of the electrolytic hydrogen production cell and the heat generated by the electrolytic cell with respect to the temperature of the cell at the previous moment, the current ambient temperature, and the operating operation. Among them, respectively represent the temperature of the electrolytic cell, the ambient temperature, the total power consumption of the electrolytic cell, the heat generated by the electrolytic cell, the heat dissipated by the electrolytic cell, and the heat utilized from the waste heat of the electrolytic cell at time t. The constraint relationship is significantly non-linear and is represented by which will be replaced by a neural network embedding in the subsequent steps. Constraint (15) limits the upper and lower bounds of the electrolytic cell temperature. Constraint (16) defines the energy conversion relationship of the electrolytic cell, that is, the conversion of electrical energy into thermal energy and hydrogen energy, where and αHy respectively represent the conversion coefficient of hydrogen production and hydrogen energy unit. Constraint (17) restricts the hydrogen flow balance of the system, where respectively represent the hydrogen production amount by electrolysis, hydrogen load amount, and absorption / release amount of the hydrogen storage device at time t. Constraint (18) clarifies the variation relationship of the hydrogen storage amount over time, where η HS，in and η HS，out respectively represent the hydrogen storage energy at time t and the energy efficiency of hydrogen storage charging / discharging. Constraints (19)-(21) clarify the upper and lower bounds of the power consumption of electrolytic hydrogen production and the charging / discharging of hydrogen storage. Constraint (22) ensures that the hydrogen storage energy is not less than that at the start of the day after a day's operation, so as to ensure that the system can continue to operate in the next few days.

[0113] Furthermore, for the construction of the electro-hydrogen-thermal neural network model of the electro-hydrogen conversion device in step (2), in the integrated optimal operation model of the electro-hydrogen-thermal system constructed in step one, constraint (14) is significantly non-linear and difficult to directly solve based on an optimization solver.

[0114] The present invention constructs a model of the electrolytic hydrogen production temperature and efficiency based on neural network representation, providing a model basis for subsequent embedding of the neural network into the optimization problem.

[0115] Furthermore, in step (2), to fit relationship (14) through a neural network, a corresponding data set needs to be constructed, that is, the temperature at the previous moment ambient temperature power consumption of the electrolyzer heat dissipation of the electrolyzer heat utilization of the waste heat of the electrolyzer as the input, the temperature of the electrolyzer at time t heat production of electrolytic hydrogen production

[0116] S1. Method for obtaining samples of the electrolytic hydrogen production temperature and efficiency model:

[0117] The above data can be obtained through simulation or physical operation. The acquisition principle of physical operation is relatively direct. The following details the specific operations for obtaining the data set using the simulation method:

[0118] S1.1. Dynamic model of electrolyzer temperature

[0119] For the dispatching and control tasks, it is generally assumed that each device operates at a constant power within the dispatching cycle. Therefore, for the heat dissipation power of the electrolyzer and the heat power of the waste heat utilization of the electrolyzer

[0120]

[0121] Among them, τ represents the time of a scheduling time slot.

[0122] The dynamic temperature change of the electrolyzer follows the following differential equation:

[0123]

[0124] Among them, C t is the lumped thermal resistance. is the heat accumulation power in the electrolyzer and can be expressed as:

[0125]

[0126] Among them and are the heat production power of electrolytic hydrogen production and the heat power of the natural heat dissipation loss of the electrolyzer respectively. It can be expressed as:

[0127]

[0128] Among them, n c , R t are parameters related to the electrolyzer, representing the number of electrolytic cells and the lumped thermal resistance respectively. and represent the electrolytic cell voltage and the thermal neutral voltage of the electrolytic cell respectively. The specific expressions will be derived in detail in S1.2.

[0129] The differential equation of formula (25) is discretized by the finite element difference method:

[0130]

[0131] Among them represents the temperature at the i-th moment, and Δt is the finite element difference time step.

[0132] S1.2, Calculation of the Hydrogen Production Heat of Electrolytic Hydrogen Production Based on the Dynamic Model

[0133] ① For the electrolyzer, the electrolytic hydrogen production power and the electrolytic hydrogen production heat have the following relationship:

[0134]

[0135] Among them, the thermal neutral voltage and the electrolytic cell voltage can be characterized by the following formula:

[0136]

[0137]

[0138]

[0139]

[0140] Equation (32) is the thermoneutral voltage as a function of temperature and is an empirical formula, where a0, a1, and a2 are device-related parameters. Equation (33) shows that the electrolytic cell voltage is composed of three overvoltages, where is the reversible potential, also known as the open-circuit voltage, represents the ohmic overvoltage, which is the voltage loss due to the internal resistance and contact impedance of the electrolytic cell, is the activation overvoltage, which is the voltage loss due to the reaction kinetics limitations inside the electrolytic cell, and i is the current density. Equation (34) is the empirical formula for the open-circuit voltage as a function of temperature and b0, b1, b2, and b3 are device-related parameters. Equation (35) shows the relationship between the current density i and the electrolytic hydrogen production power . Equations (36) and (37) are based on Ohm's law and define the value of the ohmic overvoltage , where is the equivalent resistance, R is the universal gas constant, t m represents the thickness of the membrane, T ref1 represents the reference temperature, σ ref represents the conductivity value of the conductor at T ref1 , and E pro is a parameter independent of temperature and represents the activation energy for proton transport in the membrane. Equations (38) and (39) define the calculation method for the activation overvoltage , where z is the number of moles of electrons transferred in the reaction (for hydrogen, in our model, z = 2), F is the Faraday constant, α a / c is the charge transfer coefficient in the reaction, i0 is the exchange current density, T ref2 represents the reference temperature, E exc can be defined as the activation energy of the electrode reaction, i.e., the activation energy of the anode electrocatalyst, and i 0，ref represents the exchange current density at the reference temperature T ref2 .

[0141] ② is the total heat generation during the statistical period. Considering the dynamic process of temperature change, the heat generation power at each moment is calculated based on the finite element difference solution of temperature dynamics and summed to calculate the total power value.

[0142]

[0143] Among them, τ represents the total time, and Δt represents the time step.

[0144] S2, Dataset construction:

[0145] Random sampling is carried out within the range of operating states to be fitted to obtain different previous temperatures Ambient temperature Total power of the electrolyzer Heat dissipation of the electrolyzer Heat utilization of the waste heat of the electrolyzer Values are obtained. Based on the S1 method, the current temperature of the electrolyzer is obtained Heat generation of electrolytic hydrogen production As a sample label. Repeat the operation to obtain data samples, thereby realizing the construction of the dataset.

[0146] S3, Neural network training:

[0147] An artificial neural network is constructed to fit the non-linear function (14) through the dataset constructed in S2. In particular, to realize the subsequent embedding of the neural network into the optimization problem, the neural network needs to use non-linear activation functions such as ReLU and LeakyReLU. In terms of the structure of the artificial neural network, any network structure that performs linear operations between neurons except for the activation function can be added, such as multi-layer perceptrons, graph neural networks, convex input neural networks, etc., and operations such as pruning are allowed.

[0148] Furthermore, step (3) equivalently embeds the trained neural network into the optimization problem, which is the core step of the present invention, realizing the integration of data-driven machine learning modeling and operational optimization decision-making. Compared with the energy management method based on operational optimization decision-making, the present invention improves the non-linear representation ability of the model through the embedding of the neural network; compared with the end-to-end decision-making method based on the neural network, the proposed neural network embedding optimization scheme can ensure the optimality and reliability of the decision with the help of an optimization solver.

[0149] Specifically, the way to realize the equivalent embedding of the optimization problem in step (3) is: using the neural network model trained in step (2) to equivalently replace the non-linear function (14) in the operation control optimization problem in step (1), and ensuring the equivalence of the results.

[0150] The method provided by the present invention can be applied to graph input neural networks and graph convolutional neural networks, etc. (requiring piecewise linear activation functions such as ReLU and leakyReLU). The above-mentioned neural networks can all convert the activation function into a set of piecewise linear constraints based on the same method.

[0151] In the preferred embodiment of the present invention, for the process of equivalent embedding based on the multi-layer perceptron neural network of ReLU, the specific process is as follows:

[0152] The multi-layer perceptron is composed of multiple fully connected layers. The calculation result of the output of a single neuron can be expressed in the following explanatory form

[0153]

[0154] where [L] and [N l represent the number of layers of the neural network and the number of neurons in the l-th layer respectively, and represent the output and input of the activation function of the i-th neuron in the l-th layer. and represent the weight and bias from the i-th neuron in the (l - 1)-th layer to the current neuron respectively. In addition, there is also the relationship:

[0155]

[0156] The neuron operation (41) can be equivalently expressed as the following set of constraints:

[0157] y ≥ x, (38)

[0158] y ≤ x - M L (1 - z), (39)

[0159] y ≤ M U z, (40)

[0160] y ≥ 0, z ∈ {0, 1} (41)

[0161] where M L <0 and M U >0 are the lower and upper bounds of all possible x respectively. z is a newly added auxiliary binary decision variable. By performing the above transformation on each neuron of the multi-layer perceptron, the neural network can be completely equivalently embedded into the optimization problem.

[0162] In summary, the integrated operation and control optimization problem of the electricity-hydrogen-thermal energy system can be integrally expressed as:

[0163]

[0164] In a preferred embodiment of the present invention, in the process of equivalent embedding based on the LeakyReLU graph convolutional neural network, it is specifically as follows:

[0165] The single-layer inference process of the graph convolutional neural network can be expressed as

[0166]

[0167] where [L] represents the number of layers of neurons, H l and W lThey respectively represent the neuron value output value and the neural network weight of the graph convolutional neural network at the l-th layer. is the normalized adjacency matrix.

[0168] In addition, there is also the relationship:

[0169]

[0170] Denote the expression of the LeakyReLU function as:

[0171]

[0172] The activation function operation of each neuron in formula (46) can be equivalently represented as the following set of linear constraints:

[0173] x ≥ -M(1 - z), (45)

[0174] x ≤ Mz, (46)

[0175] y ≤ x + M(1 - z), (47)

[0176] y ≤ x + M(1 - z), (48)

[0177] y ≤ αx + Mz, (49)

[0178] y ≥ αx - Mz, (50)

[0179] z ∈ {0, 1}. (51)

[0180] where M L <0 and M U >0 are respectively the lower bound and the upper bound of all possible x. z is the newly added auxiliary binary decision variable. By performing the above transformation on each neuron of the multi-layer perceptron, the neural network can be completely equivalently embedded into the optimization problem.

[0181] In summary, the integrated operation control and optimization problem of the electric-hydrogen-thermal energy system can be integrally represented as:

[0182]

[0183] Furthermore, in step (4), to solve the optimization problem, the integrated problem (55) can be directly solved using an optimization solver, thereby obtaining the optimal control instructions. Issuing the instructions can achieve the integrated optimal operation scheduling of the system.

[0184] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field according to the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art shall fall within the protection scope determined by the claims.

Claims

1. An integrated operation control method for an electric-hydrogen-thermal system, characterized in that, Including: (1) Construction of an integrated optimal operation model for the electricity-hydrogen-heat system; (2) Construction of an electricity-hydrogen-heat neural network model for the electro-hydrogen conversion device; (3) Equivalently embedding the neural network model into the optimization problem; (4) Solving the optimization problem.

2. The integrated operation control method of the electric-hydrogen-thermal system according to claim 1, characterized in that The construction of the optimal operation model described in step (1) includes determining the operation constraints on the electricity side, and the operation constraints on the electricity side include: E0 ≤ E T Among them, respectively represent the electricity quantities purchased from the power grid, generated by renewable energy, released by the electrical energy storage, charged into the electrical energy storage, consumed by the electrical load, and consumed by electrolytic hydrogen production at time t of the system; respectively represent the electricity quantity stored by the electrical energy storage at time t and the energy efficiency of the electrical energy storage for storing and releasing electrical energy.

3. The integrated operation control method of the electric-hydrogen-thermal system according to claim 1, characterized in that, The construction of the optimal operation model described in step (1) includes determining the operation constraints on the heat energy side, and the operation constraints on the heat energy side include: Among them, respectively represent the heat generated by the boiler, the fuel consumed, and its conversion efficiency at time t; and respectively represent the heat load of the system and the heat utilized from the waste heat of electrolytic hydrogen production at time t.

4. The integrated operation control method of the electric-hydrogen-thermal system according to claim 1, wherein, The construction of the optimal operation model described in step (1) includes determining the operation constraints of the electrolytic hydrogen production device and the hydrogen energy side, and the operation constraints of the electrolytic hydrogen production device and the hydrogen energy side include: Among them, respectively represent the electrolyzer temperature, ambient temperature, total power consumption of the electrolyzer, heat generation of the electrolyzer, heat dissipation of the electrolyzer, and heat utilization of the electrolyzer waste heat at time t; and respectively represent the hydrogen production amount and the hydrogen energy unit conversion coefficient; respectively represent the hydrogen production amount, hydrogen load, and absorption / release amount of the hydrogen storage device during electrolytic hydrogen production at time t; respectively represent the hydrogen storage energy and the energy efficiency of hydrogen storage charging / discharging at time t.

5. The integrated operation control method of the electric-hydrogen-thermal system according to claim 1, characterized in that, The specific steps of step (2) include: S1. Generate simulation samples based on the integration of the electrolytic hydrogen production temperature and efficiency of the model; S2. Construct a data set; S3. Train the neural network.

6. The integrated operation control method of the electric-hydrogen-thermal system according to claim 5, wherein, The specific method of S1 is: S1.

1. Dynamic model of electrolyzer temperature; Assume that the device operates at a constant power during the scheduling period, and for the heat dissipation power of the electrolyzer and the heat power utilized from the waste heat of the electrolyzer there is where τ represents the time of a scheduling time slot; The dynamic temperature change of the electrolyzer follows the following differential equation: Among which C t is the lumped thermal resistance; S1.

2. Calculation of the heat production of electrolytic hydrogen production based on the dynamic model; For the electrolyzer, the electrolytic hydrogen production power and the heat production of electrolytic hydrogen production have the following relationship: Among them, is the electrolytic cell voltage, is the electrolytic hydrogen production power; To calculate the total heat production during the statistical period, considering the dynamic process of temperature change, based on the finite element difference solution of temperature dynamics, the heat production power at each moment is calculated And the total power value is calculated by summation; where τ represents the total time and Δt represents the time step; Preferably, the construction of the data set in S2 specifically includes: Random sampling is carried out within the operating state range to be fitted to obtain different previous moment temperatures Ambient temperature Total power of electrolytic cell Heat dissipation of electrolytic cell Heat utilization of waste heat from electrolytic cell Value; Obtain the current temperature of the electrolytic cell based on the S1 method Heat generation of electrolytic hydrogen production As a sample label; Repeatedly execute operations to obtain data samples, so as to realize the construction of the data set; Preferably, the training of the neural network in S3 includes: Construct an artificial neural network, and realize the fitting of the non-linear function through the data set constructed in S2.

7. The integrated operation control method of the electric-hydrogen-thermal system according to claim 1, characterized in that The equivalent embedding of the neural network model into the optimization problem described in step (3) includes equivalently replacing the non-linear function of the operation control and management optimization problem in step (1) with the neural network model trained in step (2), and ensuring that the results are equivalent.

8. The integrated operation control method of the electric-hydrogen-thermal system according to claim 7, characterized in that, Equivalently replacing the non-linear function of the operation control and management optimization problem in step (1) is specifically the operation constraints of the electrolytic hydrogen production device and the hydrogen energy side, that is: Among them, respectively represent the electrolytic cell temperature, ambient temperature, total power consumption of the electrolytic cell, heat generation of the electrolytic cell, heat dissipation of the electrolytic cell, and heat utilization of the waste heat of the electrolytic cell at time t.

9. The integrated operation control method of the electric-hydrogen-thermal system according to claim 1, characterized in that, In step (3), the process of equivalent embedding optimization for the multi-layer perceptron neural network based on ReLU includes: The multi-layer perceptron is composed of multiple fully connected layers. The operation result of a single neuron is as follows, and the neuron calculation is equivalently represented as a set of constraints: y≥x, y ≤ x - M L (1 - z), y ≤ M U z, Among them, [L] and [N l represent the number of layers of the neural network and the number of neurons in the l-th layer respectively, and represent the output and input of the activation function of the i-th neuron in the l-th layer; and represent the weight and bias from the i-th neuron in the (l - 1)-th layer to the current neuron respectively; where M L < 0 and M U > 0 are the lower and upper bounds of all possible x respectively; z is the newly added auxiliary binary decision variable; There is also the following relationship:

10. The integrated operation control method of the electric-hydrogen-thermal system according to claim 1, characterized in that In step (3), the equivalent embedding optimization problem of the integrated operation control and management method for the electricity-hydrogen-heat system is integrally represented as:

Images