Hydrogen fuel cell temperature control method, control device and computer-readable storage medium based on neural network algorithm

Abstract

The present invention relates to the technical field of fuel cell temperature control, and particularly relates to a hydrogen fuel cell temperature control method, control device, and computer-readable storage medium based on a neural network algorithm, including the following steps: S1. Neural network model construction; S2. Operating condition identification; S3. Data acquisition; S4. Neural network model operation; S5. Temperature regulation; The present invention only adjusts the PID parameter value by the neural network model under variable load conditions, with low computing power and strong robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cell temperature control, and particularly to a hydrogen fuel cell temperature control method, a control device and a computer-readable storage medium based on a neural network algorithm. Background Art

[0002] A key factor in hydrogen fuel cell control is temperature, which directly affects the efficiency and stability of the fuel cell's output electrical performance. The more precise the temperature control, the more beneficial it is to improve various performance indicators of the hydrogen fuel cell.

[0003] The temperature control of hydrogen fuel cells mostly uses a PID controller. When the PID controller performs temperature control, different PID control quantities need to be calibrated, and the whole process is relatively cumbersome, and it performs poorly in a complex non-linear hydrogen fuel cell system. To overcome the deficiencies of the PID controller, existing temperature control methods use a neural network model to adaptively adjust the PID control quantity. However, the neural network model does not distinguish the operating conditions of the hydrogen fuel cell when adjusting the PID control quantity. In the steady-state operating condition (i.e., non-varying load), the PID controller can continue to operate according to the corresponding PID control quantity without adjustment, resulting in excessive computing power, high energy consumption, and poor robustness of the PID controller. At the same time, in each hidden layer of the neural network model, there is no connection between the weights of the neurons in the current layer and the weights of the neurons in the previous layer, resulting in unreliable weight values of the neurons in the current layer, and thus the output results of the output layer are not precise enough, and there is still room for improvement in robustness. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: to overcome the deficiencies in the prior art and provide a hydrogen fuel cell temperature control method, a control device and a computer-readable storage medium based on a neural network algorithm, in which the neural network model only adjusts the PID parameter values under variable load conditions, with small computing power and strong robustness.

[0005] The technical solution adopted by the present invention to solve its technical problems is: a hydrogen fuel cell temperature control method based on a neural network algorithm, including the following steps:

[0006] S1. Neural network model construction: Build an input layer, a hidden layer and an output layer. Among them, the hidden layer is set to be multiple layers, and the number of neurons on each hidden layer is arranged alternately in sequence;

[0007] S2. Operating condition identification: Collect the real-time power. If the real-time power is within the preset power range, it is non-varying load, and step S2 is repeatedly executed; otherwise, it is variable load, and step S3 is executed;

[0008] S3. Data acquisition: Collect the previous PID parameter value and calculate the target temperature, where the target temperature is the average value of multiple acquisitions of the current real-time temperature;

[0009] S4. Neural network model operation: Input the acquired data into the neural network model, calculate the number of iterations of the neural network model until the predetermined number of iterations is reached, and fix the PID parameter values;

[0010] S5. Temperature regulation: Perform PID control on the fan with the fixed PID parameter values and return to step S2.

[0011] Further, the calculation formula for the neurons on each hidden layer in step S1 is:

[0012]

[0013]

[0014] In the formula, represents the value of the th neuron on the th layer, represents the weight of the connection between the neurons in the current layer and the neurons in the previous layer, represents the number of neurons in the previous layer, represents the th neuron on the th layer corresponding correction matrix, and the value range of each element in is 0 - 1, represents the th neuron on the th layer corresponding learning rate matrix, and the value range of each element in is 0 - 1, represents the matrix weight of the connection of the previous neuron, represents the average value of the weight of the connection between the current neuron and the previous neuron.

[0015] Further, if , then discard the neuron value.

[0016] Further, in step S1, the hidden layer is 5 layers, and the number of neurons on each hidden layer is 4, 5, 4, 5, and 4 in sequence.

[0017] Further, in step S2, the preset power range is 0.98P - 1.02P, where P represents the current net power.

[0018] Further, in step S3, the number of times to collect the current real-time temperature is 5 times.

[0019] Further, in step S4, the predetermined number of iterations is 1000 times.

[0020] A control device includes a memory and a processor. The memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the above control method.

[0021] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of the above control method.

[0022] The beneficial effects of the present invention are as follows:

[0023] (1) The present invention identifies the working conditions through a preset power range. Only under variable load conditions does the neural network model adjust the PID parameter values, reducing computing power and energy consumption, improving robustness. At the same time, combined with a predetermined number of iterations, when the neural network model iterates to the predetermined number of iterations, the PID parameter values are fixed, further reducing computing power and energy consumption, thereby slowing down the temperature fluctuations under non-variable load and variable load.

[0024] (2) The present invention connects the current neuron weight with the neuron weight of the previous layer through a neuron calculation formula, and uses the slope and a fixed parameter matrix , to correct the weight value of the current layer neuron, adjust the numerical values of each neuron in the current layer more quickly and accurately, improve the reliability of the weight value of the current layer neuron, and further improve the accuracy of the output result of the output layer, thereby further improving the robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below in conjunction with the drawings and embodiments.

[0026] Figure 1 is a flowchart of the control method of the present invention;

[0027] Figure 2 is an architecture diagram of the neural network model in the present invention;

[0028] Figure 3 is a schematic diagram of the PID control in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] The present invention will now be further described in conjunction with the drawings and preferred embodiments. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0030] Embodiment 1

[0031] As Figures 1 - 3 shown, a hydrogen fuel cell temperature control method based on a neural network algorithm includes the following steps:

[0032] S1. Construction of neural network model: construct input layer, hidden layer and output layer, where the hidden layer is set to multiple layers, and the number of neurons in each hidden layer is arranged alternately in sequence.

[0033] Specifically, there are 5 hidden layers, and the number of neurons in each hidden layer is 4, 5, 4, 5 and 4 respectively. The number of neurons in each layer is arranged in a staggered manner, which is conducive to making the neural network better adapt to data and processing tasks, and also enables each layer to learn abstract representations of data at different levels; each layer in the neural network model is equipped with a different perceptron, which is composed of parameter weights, bias, and activation function. It can be forward propagated and backward propagated at the same time, and the parameter weights of the perceptron are adjusted. This is the prior art.

[0034] S2. Working condition identification: collect real-time power. If the real-time power is within the preset power range, it is non-variable load and step S2 is repeated; otherwise, it is variable load and step S3 is executed.

[0035] Specifically, the preset power range is 0.98P-1.02P, where P represents the current net power.

[0036] S3. Data acquisition: Collect the previous PID parameter value and calculate the target temperature, where the target temperature is the average value of the current real-time temperature collected multiple times.

[0037] Specifically, the current real-time temperature is collected 5 times; the PID parameters are parameter P, parameter I and parameter D.

[0038] S4. Neural network model operation: input the acquired data into the neural network model, calculate the number of iterations of the neural network model, until the predetermined number of iterations is reached, and fix the PID parameter value.

[0039] Specifically, the predetermined number of iterations is 1000.

[0040] S5. Temperature control: Use the fixed PID parameter value to perform PID control on the fan and return to step S2.

[0041] Specifically, Figure 3 The actuator refers to the fan speed.

[0042] The operating conditions are identified by pre-setting the power range, and the neural network model adjusts the PID parameter values ​​only under variable load conditions, which reduces computing power and energy consumption and improves robustness. At the same time, combined with the predetermined number of iterations, the PID parameter values ​​are fixed when the neural network model iterations reach the predetermined number of iterations, which further reduces computing power and energy consumption, thereby mitigating temperature fluctuations under non-variable load and variable load conditions.

[0043] The calculation formula for the neurons in each hidden layer in step S1 is:

[0044]

[0045]

[0046] In the formula, represents the value of the th neuron in the layer, represents the weight of the connection between the neurons in the current layer and the neurons in the previous layer, represents the number of neurons in the previous layer, represents the th correction matrix corresponding to the th neuron in the layer, and the value range of each element in is 0 - 1, represents the matrix weight of the connection with the previous neuron,

[0047] The current neuron weight is related to the neuron weight of the previous layer through the neuron calculation formula, and the slope and the fixed parameter matrix , are used to correct the weight value of the neurons in the current layer, adjust the values of each neuron in the current layer more quickly and accurately, improve the reliability of the weight values of the neurons in the current layer, and further improve the accuracy of the output result of the output layer, thereby further improving the robustness. Among them, the correction matrix can adjust the influence degree of the current neuron by the previous neuron; the learning rate matrix can increase the weight influence of the previous neuron to the next neuron, control the step size of the bias update, avoid skipping the optimal solution due to too fast update or resulting in too long learning time due to too slow update, and a more suitable value can be found through a predetermined number of iterations.

[0048] Specifically, if , then discard the value of this neuron.

[0049] Embodiment 2

[0050] A control device includes a memory and a processor. The memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the control method described in Embodiment 1.

[0051] Embodiment 3

[0052] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of the control method described in Embodiment 1.

[0053] The above embodiments are only for illustrating the technical concept and features of the present invention, and the purpose is to enable those skilled in the art to understand the content of the present invention and implement it. However, the protection scope of the present invention cannot be limited thereby. Any equivalent changes or modifications made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A hydrogen fuel cell temperature control method based on a neural network algorithm, characterized in that: The steps include: S1. Construction of neural network model: construct input layer, hidden layer and output layer, where the hidden layer is set to multiple layers, and the number of neurons in each hidden layer is staggered in sequence; S2. Working condition identification: collect real-time power. If the real-time power is within the preset power range, it is non-variable load, and repeat step S2; otherwise, it is variable load, and execute step S3; S3. Data acquisition: collect the previous PID parameter value and calculate the target temperature, where the target temperature is the average value of the current real-time temperature collected multiple times; S4. Neural network model operation: input the acquired data into the neural network model, calculate the number of iterations of the neural network model, until the predetermined number of iterations is reached, and fix the PID parameter value; S5. Temperature control: PID control the fan with a fixed PID parameter value, and return to step S2; The calculation formula of neurons in each hidden layer in step S1 is: ； In the formula, Indicates Tier The value of a neuron, Represents the weight of the connection between the current layer of neurons and the previous layer of neurons, represents the number of neurons in the previous layer, Indicates Tier The correction matrix corresponding to the neurons, and The value range of each element in is 0-1. Indicates Tier The learning rate matrix corresponding to the neurons, and The value range of each element in is 0-1. represents the matrix weight of the previous neuron connection, Represents the average value of the weights connecting the current neuron and the previous neuron.

2. The hydrogen fuel cell temperature control method based on the neural network algorithm according to claim 1 is characterized in that: like , then the neuron value is discarded.

3. The hydrogen fuel cell temperature control method based on the neural network algorithm according to claim 1 is characterized in that: In step S1, there are 5 hidden layers, and the number of neurons in each hidden layer is 4, 5, 4, 5 and 4 respectively.

4. The hydrogen fuel cell temperature control method based on the neural network algorithm according to claim 1 is characterized in that: The preset power range in step S2 is 0.98P-1.02P, where P represents the current net power.

5. The hydrogen fuel cell temperature control method based on the neural network algorithm according to claim 1 is characterized in that: The number of times the current real-time temperature is collected in step S3 is 5 times.

6. The hydrogen fuel cell temperature control method based on the neural network algorithm according to claim 1 is characterized in that: The predetermined number of iterations in step S4 is 1000.

7. A control device, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of any one of the control methods of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the control method according to any one of claims 1 to 6 are implemented.

Images