Hydrogen production system based on neural network model and control method

Abstract

The invention relates to a hydrogen production system based on a neural network model and a control method, and belongs to the technical field of hydrogen preparation, and the system integrates a power supply module, a water replenishing module, an electrolytic hydrogen production module, a hydrogen absorption module and a control module. The power supply system is respectively connected with the water replenishing device and the electrolytic hydrogen production device, and the water replenishing device supplies fresh water to the electrolytic hydrogen production device through the control valve. And hydrogen generated by the electrolytic hydrogen production device is processed or stored by the hydrogen absorption module. The control module comprises a PID (Proportion Integration Differentiation) controller and an LSTM (Long Short Term Memory) neural network, and is responsible for monitoring current, voltage, liquid level, flow and valve opening, adjusting PID parameters through the LSTM network, dynamically controlling the valve opening and ensuring the balance of water replenishing and electrolysis water consumption. According to the method, PID parameters are adjusted after LSTM network training, so that the opening degree of the valve is dynamically controlled, the balance between water replenishing and electrolysis water consumption is accurately ensured, the efficient and accurate hydrogen production control process is achieved, the hydrogen production efficiency and quality are effectively improved, and resource waste is reduced.

Description

Technical Field

[0001] The present invention relates to a hydrogen production system and a control method based on a neural network model, belonging to the technical field of hydrogen production. Background Art

[0002] With the continuous growth of global energy demand and the increasing emphasis on environmental protection, hydrogen, as a clean and efficient energy carrier, has received extensive attention. As one of the important methods for producing high-purity hydrogen, the water electrolysis hydrogen production technology has broad application prospects. However, there are some problems in the operation of existing hydrogen production systems. On the one hand, the traditional hydrogen production system is not precise enough in water replenishment control. Usually, simple liquid level switches or timing control methods are used to adjust the operation of the water replenishment device, and it is unable to adjust the water replenishment amount in real time according to the actual water consumption of the electrolytic hydrogen production device, which easily leads to excessive or insufficient water replenishment, affecting the hydrogen production efficiency and the stability of the system. On the other hand, the existing hydrogen production system does not make full use of operation data, lacks effective data analysis and processing means, and it is difficult to comprehensively and accurately monitor and predict the operation state of the system, which is not conducive to the optimized operation and fault prevention of the system.

[0003] The patent document with the patent number "CN115572990A" discloses a hydrogen production system and its control method. The problems of this method are as follows: The overall technology is relatively traditional, mainly relying on preset thresholds and rules to control the on-off of the electrolytic cell, the control logic is relatively simple, lacking the ability of deep learning and intelligent analysis of the system operation state, and it is difficult to cope with complex working condition changes and system characteristic changes. The control accuracy is relatively limited. It mainly determines the number of electrolytic cells to be turned on based on the preset hydrogen demand threshold. This control method is relatively rough and cannot achieve precise dynamic control of the balance between water replenishment and electrolysis water consumption. In actual operation, the situation of mismatch between water replenishment amount and water consumption amount may occur, affecting the hydrogen production efficiency and the stability of the system. Summary of the Invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a hydrogen production system and a control method based on a neural network model.

[0005] The technical solution of the present invention is as follows:

[0006] On the one hand, the present invention provides a hydrogen production system based on a neural network model, including a power supply system, a water replenishment device, an electrolytic hydrogen production device, a hydrogen consumption module, and a control module;

[0007] The output end of the power supply system is respectively connected to the input end of the water replenishment device and the input end of the electrolytic hydrogen production device. The output end of the water replenishment device is connected to the input end of the electrolytic hydrogen production device through a control valve. The input end of the electrolytic hydrogen production device is connected to the input end of the hydrogen consumption module;

[0008] The input end of the control module is connected to the water replenishing device and the electrolytic hydrogen production device, and the output end of the control module is connected to the water replenishing device;

[0009] The water replenishing device provides fresh water for the electrolytic hydrogen production device;

[0010] The hydrogen consumption module is used to consume or store the hydrogen produced by the electrolytic hydrogen production device;

[0011] The control module includes a PID controller and a long short-term memory network LSTM, which is used to collect the current, voltage, liquid level, flow rate of the hydrogen production system and the opening degree of the control valve in real time, and input them into the long short-term memory network LSTM for training. The parameters of the PID controller are adjusted in real time through the long short-term memory network LSTM to control the opening and closing of the control valve, so as to achieve the balance between the water replenishment amount and the electrolytic water consumption amount of the hydrogen production system.

[0012] As a preferred embodiment, the water replenishing device includes a box body, an impurity filter screen, an inlet and outlet water pump, a reverse osmosis membrane module and a raw material buffer tank;

[0013] The box body includes a raw water tank and a finished water tank. The impurity filter screen is arranged at the water inlet of the raw water tank. The output end of the raw water tank is connected to the input end of the finished water tank through the inlet and outlet water pump, the raw material buffer tank and the reverse osmosis membrane module. The output end of the finished water tank is connected to the input end of the electrolytic hydrogen production device;

[0014] The flowmeter is arranged at the water outlet of the inlet and outlet water pump, and liquid level gauges are arranged inside the raw water tank and the finished water tank.

[0015] As a preferred embodiment, the electrolytic hydrogen production device is an alkaline electrolytic water hydrogen production system, an AEM hydrogen production system or a PEM hydrogen production system.

[0016] As a preferred embodiment, the hydrogen consumption module includes a combustion device, a hydrogen using device and a hydrogen storage device;

[0017] The output end of the electrolytic hydrogen production device is respectively connected to the input end of the combustion device, the input end of the hydrogen using device and the input end of the hydrogen storage device;

[0018] The output end of the hydrogen storage device is connected to the input ends of the input ends of the combustion device and the hydrogen using device.

[0019] On the other hand, the present invention also provides a control method for a hydrogen production system based on a neural network model, including the following steps:

[0020] Obtain the operation data of the hydrogen production system, where the operation data includes the current, voltage, cell temperature, pressure, flow rate, temperature before the valve, temperature after the valve, pressure before the valve, pressure after the valve parameters of the electrolytic hydrogen production device, and the liquid level, flow rate, pressure of the water replenishment device, the opening degree of the control valve, and the environmental data;

[0021] Set the initial parameters for the PID controller according to the operation data;

[0022] Add Gaussian white noise to the operation data and repeat the addition operation multiple times. Normalize the operation data after adding Gaussian white noise and use it as the data set;

[0023] Use the data set to train the long short-term memory network LSTM;

[0024] Start the water replenishment device and detect the liquid level through the liquid level gauge. When the liquid level reaches the preset height, start the electrolytic hydrogen production device to produce hydrogen;

[0025] Real-time predict the opening degree of the control valve through the long short-term memory network LSTM, and input the predicted opening degree of the control valve and the data set into the PID controller to update the initial parameters;

[0026] Control the opening degree of the control valve through the PID controller to achieve the balance between the water replenishment volume of the water replenishment device and the water consumption of the electrolytic hydrogen production device.

[0027] As a preferred embodiment, the method for determining the initial parameters is as follows:

[0028] Calculate the comprehensive performance metric:

[0029] J = α∫|e(t)|dt + β∫u 2 (t)dt + γ·OS;

[0030] where α, β, γ represent preset weights, OS represents the overshoot, e(t) represents the error between the water replenishment volume and the water consumption at time t, u 2 (t) represents the square of the operation data u(t) at time t, and J represents the comprehensive performance metric;

[0031] Based on the comprehensive performance metric J, use the multi-objective genetic algorithm to search for the Pareto optimal solution set that makes J < 0.15 in the parameter space (K p ∈[0, 5], K i ∈[0, 0.1], K d ∈[0, 0.5]);

[0032] Take the proportional gain K p , integral gain K i and derivative gain K d determined in the Pareto optimal solution set as the initial parameters of the PID controller.

[0033] As a preferred embodiment, the loss function of the long short-term memory network LSTM is the mean square error function.

[0034] As a preferred embodiment, the number of input nodes of the long short-term memory network LSTM is 8;

[0035] The number of hidden layers is 4, and the number of nodes in each hidden layer is 100. The Dropout rate in the hidden layer is set to 40%;

[0036] The optimizer is set to the Adam optimizer. The exponential decay rate beta1 for the first-order momentum estimation is set to 0.9, and the exponential decay rate beta2 for the second-order momentum estimation is set to 0.999;

[0037] The learning rate lr is set to 0.001.

[0038] As a preferred embodiment, the training method of the long short-term memory network LSTM is:

[0039] The forget gate is expressed as:

[0040] f t = σ(W f [h t-1 ,c t +b f );

[0041] Among them, f t represents the activation value of the forget gate, W f is the weight matrix of the forget gate, b f represents the bias term of the forget gate;

[0042] The input gate is expressed as:

[0043] i t = σ(W i [h t-1 ,x t +b i );

[0044] Among them, i t represents the activation value of the input gate, W i represents the weight matrix of the input gate, σ represents the sigmoid activation function, b i represents the bias term of the input gate, h t-1 represents the hidden state at time t-1, x t represents the data set at time t;

[0045] The candidate memory cell at time t is expressed as:

[0046]

[0047] Among them, represents the candidate memory cell at time t, tanh represents the hyperbolic tangent activation function, W c represents the weight matrix of the candidate memory cell, b c represents the bias term of the candidate memory cell;

[0048] The cell state at time t is expressed as:

[0049]

[0050] Among them, c t represents the cell state at time t, c t-1 represents the cell state at time t - 1;

[0051] The output gate is expressed as:

[0052] o t = σ(W o [h t-1 , x t +b o );

[0053] Among them, o t represents the activation value of the output gate, W o represents the weight matrix of the input gate, b o is the bias term of the input gate;

[0054] The hidden state at time t is expressed as:

[0055] h t = o t ·tanh(c t );

[0056] Among them, h t represents the hidden state at time t;

[0057] The method for predicting the opening of the control valve is:

[0058] v t = linear(h t );

[0059] Among them, v t represents the opening of the control valve at time t, linear represents the linear layer function.

[0060] As a preferred embodiment, the method of the normalization process is:

[0061]

[0062] Among them, x newrepresents the operating data after normalization, \(x\) represents the operating data before normalization, \(x\) min represents the minimum value of the operating data before normalization, \(x\) max represents the maximum value of the operating data before normalization.

[0063] The present invention has the following beneficial effects:

[0064] The present invention integrates a power supply, water replenishment, electrolytic hydrogen production, hydrogen consumption, and control module, and uses a PID controller and an LSTM neural network in the control module to monitor key parameters of the hydrogen production system in real time, such as current, voltage, liquid level, flow rate, and valve opening. After training with the LSTM network, the PID parameters are adjusted to dynamically control the valve opening, accurately ensuring the balance between water replenishment and electrolytic water consumption, realizing an efficient and accurate hydrogen production control process, effectively improving the hydrogen production efficiency and quality, and reducing resource waste. Before training the LSTM neural network, Gaussian white noise is added to the operating data and normalization processing is performed. Adding Gaussian white noise can simulate various interference factors in actual working conditions, enabling the model to have a certain anti-interference ability during the training process; normalization processing unifies the data to the same scale range, which helps to accelerate the convergence speed of the model, improve the training efficiency, and avoid the problem of unstable model training caused by too large differences in data dimensions, thereby enhancing the robustness and generalization ability of the LSTM model, enabling it to more accurately predict and control the valve opening in practical applications, and ensuring the stable operation of the hydrogen production system. Description of the Drawings

[0065] Figure 1 is the system connection diagram of the present invention.

[0066] Figure 2 is the flowchart of the control method of the present invention. Detailed Embodiments

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0068] It should be understood that the step numbers used in the text are only for convenient description and do not limit the execution order of the steps.

[0069] It should be understood that the terms used in the specification of the present invention are merely for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0070] The terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0071] The term "and / or" refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0072] Embodiment 1:

[0073] See Figure 1 , the present invention provides a hydrogen production system based on a neural network model, including a power supply system, a water replenishing device, an electrolytic hydrogen production device, a hydrogen consumption module and a control module;

[0074] The output end of the power supply system is respectively connected to the input end of the water replenishing device and the input end of the electrolytic hydrogen production device. The output end of the water replenishing device is connected to the input end of the electrolytic hydrogen production device through a control valve. The input end of the electrolytic hydrogen production device is connected to the input end of the hydrogen consumption module;

[0075] The input end of the control module is connected to the water replenishing device and the electrolytic hydrogen production device, and the output end of the control module is connected to the water replenishing device;

[0076] The water replenishing device provides fresh water for the electrolytic hydrogen production device;

[0077] The hydrogen consumption module is used to consume or store the hydrogen produced by the electrolytic hydrogen production device;

[0078] The control module includes a PID controller and a long short-term memory network LSTM, which are used to collect the current, voltage, liquid level, flow rate of the hydrogen production system and the opening degree of the control valve in real time, and input them into the long short-term memory network LSTM for training. The parameters of the PID controller are adjusted in real time through the long short-term memory network LSTM to control the opening and closing of the control valve, so as to achieve the balance between the water replenishment amount and the electrolytic water consumption amount of the hydrogen production system.

[0079] As a preferred embodiment, the water replenishing device includes a box body, an impurity filter screen, an inlet and outlet water pump, a reverse osmosis membrane module and a raw material buffer tank;

[0080] The box body includes a raw water tank and a finished water tank. The raw water tank is used to store the water to be treated, and the finished water tank is used to store the treated water. The impurity filter screen is arranged at the water inlet of the raw water tank to remove large particle impurities. The output end of the raw water tank is connected to the input end of the finished water tank through a water inlet and outlet pump, a raw material buffer tank, and a reverse osmosis membrane module. The raw material buffer tank is used to supply chemical agents to the reverse osmosis membrane module. The output end of the finished water tank is connected to the input end of the electrolytic hydrogen production device; the reverse osmosis membrane module removes dissolved salts and minute impurities through a semi-permeable membrane.

[0081] The liquid level gauge monitors the water levels of the raw water tank and the finished water tank to prevent overflow or dry pumping, and automatically controls the start and stop of the water inlet and outlet pump according to the liquid level signal.

[0082] The flowmeter is arranged at the water outlet of the water inlet and outlet pump, and liquid level gauges are arranged inside the raw water tank and the finished water tank.

[0083] As a preferred embodiment, the electrolytic hydrogen production device is an alkaline electrolytic water hydrogen production system, an AEM hydrogen production system, or a PEM hydrogen production system.

[0084] As a preferred embodiment, the hydrogen consumption module includes a combustion device, a hydrogen using device, and a hydrogen storage device;

[0085] The output end of the electrolytic hydrogen production device is respectively connected to the input end of the combustion device, the input end of the hydrogen using device, and the input end of the hydrogen storage device;

[0086] The output end of the hydrogen storage device is connected to the input ends of the input ends of the combustion device and the hydrogen using device.

[0087] Embodiment 2:

[0088] See Figure 2 , the present invention also provides a control method for a hydrogen production system based on a neural network model, including the following steps:

[0089] Obtain the operation data of the hydrogen production system. The operation data includes the current, voltage, cell temperature, pressure, flow rate, temperature before the valve, temperature after the valve, pressure before the valve, pressure after the valve parameters of the electrolytic hydrogen production device, and the liquid level, flow rate, pressure, opening degree of the control valve of the water replenishing device, and environmental data;

[0090] Set the initial parameters for the PID controller according to the operation data;

[0091] Add Gaussian white noise to the operation data and repeat the addition operation multiple times. Normalize the operation data after adding Gaussian white noise and use it as a data set;

[0092] Use the data set to train the long short-term memory network LSTM;

[0093] Start the water replenishing device and detect the liquid level through the liquid level gauge. When the liquid level reaches the preset height, start the electrolytic hydrogen production device to produce hydrogen;

[0094] Real-time predict and control the opening of the valve through the long short-term memory network LSTM, and input the predicted opening of the control valve and the data set into the PID controller to update the initial parameters;

[0095] Control the opening of the control valve through the PID controller to achieve the balance between the water replenishment volume of the water replenishing device and the water consumption of the electrolytic hydrogen production device.

[0096] As a preferred implementation method, the method for determining the initial parameters is:

[0097] Calculate the comprehensive performance metric:

[0098] J = α∫|e(t)|dt + β∫u 2 (t)dt + γ·OS;

[0099] Where α, β, and γ represent preset weights. In this embodiment, α is 0.6, β is 0.3, γ is 0.1, OS represents the overshoot, e(t) represents the error between the water replenishment volume and the water consumption at time t, and u 2 (t) represents the square of the operating data u(t) at time t, and J represents the comprehensive performance metric;

[0100] The overshoot OS is the calculated value of the response curve after simulating the control system under different conditions:

[0101] OS = (c peak - c steady ) / c steady * 100%;

[0102] Where C peak represents the maximum peak value of the response curve, and C steady represents the final steady-state value.

[0103] Based on the comprehensive performance metric J, use the multi-objective genetic algorithm to search for the Pareto optimal solution set that makes J < 0.15 in the parameter space (K p ∈ [0, 5], K i ∈ [0, 0.1], K d ∈ [0, 0.5]);

[0104] Take the proportional gain K p , integral gain K i and derivative gain K d determined in the Pareto optimal solution set as the initial parameters of the PID controller.

[0105] As a preferred embodiment, the loss function of the long short-term memory network (LSTM) is the mean square error function.

[0106] As a preferred embodiment, the number of input nodes of the long short-term memory network (LSTM) is 8;

[0107] The number of hidden layers is 4, and the number of nodes in each hidden layer is 100. The Dropout rate in the hidden layer is set to 40%;

[0108] The optimizer is set to the Adam optimizer. The exponential decay rate beta1 for the first-order momentum estimation is set to 0.9, and the exponential decay rate beta2 for the second-order momentum estimation is set to 0.999;

[0109] The learning rate lr is set to 0.001.

[0110] As a preferred embodiment, the training method of the long short-term memory network (LSTM) is as follows:

[0111] The forget gate is expressed as:

[0112] f t = σ(W f [h t-1 , x t + b f );

[0113] Among them, f t represents the activation value of the forget gate, W f is the weight matrix of the forget gate, b f represents the bias term of the forget gate;

[0114] The input gate is expressed as:

[0115] i t = σ(W i [h t-1 , x y + b i );

[0116] Among them, i t represents the activation value of the input gate, W i represents the weight matrix of the input gate, σ represents the sigmoid activation function, b i represents the bias term of the input gate, h t-1 represents the hidden state at time t-1, and x t represents the data set at time t;

[0117] The candidate memory cell at time t is expressed as:

[0118]

[0119] Among them, represents the candidate memory cell at time t, tanh represents the hyperbolic tangent activation function, W c represents the weight matrix of the candidate memory cell, b c represents the bias term of the candidate memory cell;

[0120] The cell state at time t is expressed as:

[0121]

[0122] Among them, c t represents the cell state at time t, c t-1 represents the cell state at time t - 1;

[0123] The output gate is expressed as:

[0124] o t = σ(W o [h t-1 , x t + b o );

[0125] Among them, o t represents the activation value of the output gate, W o represents the weight matrix of the input gate, b o is the bias term of the input gate;

[0126] The hidden state at time t is expressed as:

[0127] h t = o t · thnh(c t );

[0128] Among them, h t represents the hidden state at time t;

[0129] The method for predicting the opening of the control valve is:

[0130] v t = linear(h t );

[0131] Among them, v t represents the opening of the control valve at time t, linear represents the linear layer function.

[0132] As a preferred embodiment, the method for normalization processing is:

[0133]

[0134] Among them, x new represents the running data after normalization processing, x represents the running data before normalization processing, xmin represents the minimum value of the running data before normalization, x max represents the maximum value of the running data before normalization.

[0135] Merge the newly acquired running data every hour with the aforementioned running data to construct a new dataset, delete several historical samples in the dataset so that the sample size is maintained at 20,000, and asynchronously train the long short-term memory network (LSTM) while maintaining the operation of the PID controller. Read the weights of the long short-term memory network (LSTM) after asynchronous training to update the long short-term memory network (LSTM).

[0136] In the embodiments of the present application, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects and indicates that there can be three relationships. For example, A and / or B can represent the cases of A existing alone, A and B existing simultaneously, and B existing alone. Where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one of the following" and its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0137] Those of ordinary skill in the art can realize that the units and algorithm steps described in the embodiments disclosed herein can be implemented by a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present application.

[0138] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0139] In several embodiments provided by the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0140] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A hydrogen production system based on a neural network model, characterized in that, It includes a power supply system, a water replenishing device, an electrolytic hydrogen production device, a hydrogen consumption module, and a control module; The output end of the power supply system is respectively connected to the input end of the water replenishing device and the input end of the electrolytic hydrogen production device. The output end of the water replenishing device is connected to the input end of the electrolytic hydrogen production device through a control valve. The input end of the electrolytic hydrogen production device is connected to the input end of the hydrogen consumption module; The input end of the control module is connected to the water replenishing device and the electrolytic hydrogen production device, and the output end of the control module is connected to the water replenishing device; The water replenishing device provides fresh water for the electrolytic hydrogen production device; The hydrogen consumption module is used to consume or store the hydrogen produced by the electrolytic hydrogen production device; The control module includes a PID controller and a long short-term memory network LSTM, which is used to collect the current, voltage, liquid level, flow rate of the hydrogen production system and the opening degree of the control valve in real time, and input them into the long short-term memory network LSTM for training. The parameters of the PID controller are adjusted in real time through the long short-term memory network LSTM to control the opening and closing of the control valve, so as to achieve the balance between the water replenishment amount and the electrolytic water consumption amount of the hydrogen production system.

2. The hydrogen production system based on a neural network model according to claim 1, wherein, The water replenishing device includes a box body, an impurity filter screen, an inlet and outlet water pump, a reverse osmosis membrane module, and a raw material buffer tank; The box body includes a raw water tank and a finished water tank. The impurity filter screen is arranged at the water inlet of the raw water tank. The output end of the raw water tank is connected to the input end of the finished water tank through the inlet and outlet water pump, the raw material buffer tank, and the reverse osmosis membrane module. The output end of the finished water tank is connected to the input end of the electrolytic hydrogen production device; The flow meter is arranged at the outlet of the inlet and outlet water pump, and liquid level gauges are arranged inside the raw water tank and the finished water tank.

3. The hydrogen production system based on a neural network model according to claim 1, wherein The electrolytic hydrogen production device is an alkaline electrolytic water hydrogen production system, an AEM hydrogen production system, or a PEM hydrogen production system.

4. The hydrogen production system based on a neural network model according to claim 3, wherein The hydrogen consumption module includes a combustion device, a hydrogen using device, and a hydrogen storage device; The output end of the electrolytic hydrogen production device is respectively connected to the input end of the combustion device, the input end of the hydrogen using device, and the input end of the hydrogen storage device; The output end of the hydrogen storage device is connected to the input end of the combustion device and the input end of the hydrogen using device.

5. A control method for a hydrogen production system based on a neural network model, characterized in that, It includes the following steps: Obtain the operation data of the hydrogen production system. The operation data includes the current, voltage, cell temperature, pressure, flow rate, temperature before the valve, temperature after the valve, pressure before the valve, pressure after the valve parameters of the electrolytic hydrogen production device, and the liquid level, flow rate, pressure, opening degree of the control valve of the water replenishing device, and environmental data; Set the initial parameters for the PID controller according to the operation data; Add Gaussian white noise to the operation data and repeat the addition operation multiple times. Normalize the operation data after adding Gaussian white noise and use it as a data set; Use the data set to train the long short-term memory network LSTM; Start the water replenishing device and detect the liquid level through the liquid level gauge. When the liquid level reaches the preset height, start the electrolytic hydrogen production device to produce hydrogen; Real-time predict the opening degree of the control valve through the long short-term memory network LSTM, and input the predicted opening degree of the control valve and the data set into the PID controller to update the initial parameters; Control the opening degree of the control valve through the PID controller to achieve the balance between the water replenishment amount of the water replenishing device and the water consumption amount of the electrolytic hydrogen production device.

6. The hydrogen production system control method based on a neural network model according to claim 5, characterized in that, The determination method of the initial parameters is: Calculate the comprehensive performance metric: J = α∫|e(t)|dt + β∫u 2 (t)dt + γ·OS; Among them, α, β, and γ represent preset weights, OS represents the overshoot, e(t) represents the error between the water replenishment and water consumption at time t, and u 2 (t) represents the square of the operating data u(t) at time t, and J represents the comprehensive performance metric; Using a multi-objective genetic algorithm based on the comprehensive performance metric J to search for the Pareto optimal solution set that makes J < 0.15 in the parameter space (K p ∈ [0, 5], K i ∈ [0, 0.1], K d ∈ [0, 0.5]); Determine the proportional gain K, integral gain K, and derivative gain K in the Pareto optimal solution set as the initial parameters of the PID controller. p Integral gain K i And derivative gain K d As the initial parameters of the PID controller.

7. The hydrogen production system control method based on a neural network model according to claim 5, wherein The loss function of the long short-term memory network LSTM is the mean square error function.

8. The hydrogen production system control method based on a neural network model according to claim 5, wherein The number of input nodes of the long short-term memory network LSTM is 8; The number of hidden layers is 4, and the number of nodes in each hidden layer is 100. The Dropout rate in the hidden layer is set to 40%; The optimizer is set to the Adam optimizer. The exponential decay rate beta1 for first-order momentum estimation is set to 0.9, and the exponential decay rate beta2 for second-order momentum estimation is set to 0.999; The learning rate lr is set to 0.

001.

9. The hydrogen production system control method based on a neural network model according to claim 5, wherein The training method of the long short-term memory network LSTM is as follows: The forget gate is expressed as: f t = σ(W f [h t-1 ,c t + b f ); Among them, f t represents the activation value of the forget gate, and W f is the weight matrix of the forget gate, and b f represents the bias term of the forget gate; The input gate is expressed as: i t = σ(W i [h t-1 , x t + b i ); Among them, i t represents the activation value of the input gate, W i represents the weight matrix of the input gate, σ represents the sigmoid activation function, b i represents the bias term of the input gate, h t-1 represents the hidden state at time t-1, x t represents the data set at time t; The candidate memory cell at time t is expressed as: Among them, represents the candidate memory cell at time t, tanh represents the hyperbolic tangent activation function, W c represents the weight matrix of the candidate memory cell, b c represents the bias term of the candidate memory cell; The cell state at time t is expressed as: where, c t represents the cell state at time t, and c t-1 represents the cell state at time t-1; The output gate is expressed as: o t = σ(W o [h t-1 , x t + b o ); Among them, o t represents the activation value of the output gate, W o represents the weight matrix of the input gate, b o is the bias term of the input gate; The hidden state at time t is expressed as: h t = o t ·tanh(c t ); Among them, h t represents the hidden state at time t; The prediction method for the opening degree of the control valve is: v t = linear(h t ); Among them, v t represents the opening degree of the control valve at time t, and linear represents the linear layer function.

10. The control method of the hydrogen production system based on the neural network model according to claim 5, wherein, The method of the normalization process is: Among them, x new represents the operation data after normalization, x represents the operation data before normalization, and x min represents the minimum value of the operation data before normalization, and x max represents the maximum value of the operation data before normalization.

Images