Wind power generation coupled PEM electrolytic hydrogen production predictive control method
By using the MLP-BiLSTM-TCN combined model to predict wind power generation and combining it with the PEM state strategy controller to achieve energy coordinated control, the power mismatch problem between the wind power generation system and the PEM electrolysis hydrogen production system was solved, and the energy utilization rate and system stability were improved.
Patent Information
- Application Number
- CN202510711190.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-29
- Publication Date
- 2025-09-12
AI Technical Summary
The intermittent and fluctuating nature of wind power generation systems results in a power mismatch between them and the PEM electrolysis hydrogen production system, leading to frequent starts and stops and reduced energy utilization.
A machine learning model based on the MLP-BiLSTM-TCN combination is used to predict wind power generation power, and the PEM state strategy controller is used to achieve energy coordinated control to match the power of the wind power generation system and the PEM electrolysis hydrogen production system.
The energy utilization rate of the wind power generation coupled PEM electrolysis hydrogen production system is improved, the frequent start and stop of the electrolysis system is reduced, and the stability and efficiency of the system are improved.
Smart Images
Figure CN120638291A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of PEM electrolysis hydrogen production, and in particular relates to a wind power generation coupled PEM electrolysis hydrogen production prediction and control method. Background Art
[0002] Developing wind power and hydrogen energy is a crucial path to achieving the goals of "carbon peak and carbon neutrality." Furthermore, solar and wind power, among other sources, are experiencing explosive growth. In the future, wind power will gradually replace traditional fossil fuels and assume a dominant position in the energy sector. Furthermore, hydrogen, as a high-energy-density energy carrier, is often considered the best means of storing energy from wind power. PEM water electrolysis hydrogen production technology utilizes wind power during the hydrogen production process, produces no carbon compounds, and produces high-concentration hydrogen. This mature, clean, and efficient hydrogen production technology holds a crucial position in future hydrogen energy applications and economic and technological development.
[0003] Due to the inherent intermittent and fluctuating nature of wind power generation, its large-scale grid-connected generation can cause stability issues in the power system and lead to wind curtailment. Although PEM electrolysis hydrogen production systems have the advantages of fast start-up and shutdown speeds, a wide load range, and suitability for coupling with renewable energy, there is still a power mismatch with wind power generation systems, resulting in frequent start-up and shutdown of the electrolysis system. Therefore, it is necessary to add a power prediction system to the existing wind power generation system and, from the perspective of energy utilization, add a set of coupling control strategies for wind power generation and PEM electrolysis hydrogen production systems to achieve the effect of improving the energy utilization rate of the wind power-coupled PEM electrolysis hydrogen production system and reducing the frequent start-up and shutdown of the electrolysis system. However, current prediction models mostly use a single deep learning model, which has low prediction accuracy and lacks universal applicability, making it relatively limited in practical applications. Summary of the Invention
[0004] In order to solve the above technical problems, the present invention provides a wind power generation coupled with PEM electrolysis hydrogen production prediction control method.
[0005] The specific plan is as follows: A wind power generation coupled with PEM electrolysis hydrogen production prediction and control method includes a wind power generation system, a wind power generation power predictor, a PEM state strategy controller, a PEM electrolysis hydrogen production system and a accumulator; wherein, The wind power forecaster uses a machine learning model based on the MLP-BiLSTM-TCN combination to predict wind power. The wind power forecaster predicts the power generation at the next moment, obtains the predicted value of the combined model, and outputs the predicted value of the combined model to the PEM state strategy controller. The PEM state strategy controller adopts the energy coordination control mode according to the predicted value of the combined model to distribute the electric energy of the wind power generation system between the PEM electrolysis hydrogen production system and the storage battery, so that the power of the wind power generation system matches the power of the PEM electrolysis hydrogen production system.
[0006] The wind power prediction method using a machine learning model based on the MLP-BiLSTM-TCN combination in the wind power forecaster includes the following steps: Step 1: Use the Pearson correlation coefficient method to perform correlation analysis on the original data set, remove meteorological information that is extremely weakly correlated or irrelevant to wind power, and select meteorological information with strong, moderate, and weak correlation as the input value after preliminary processing; Step 2: The dataset processed by the Pearson correlation coefficient method is further preprocessed to remove outliers in the dataset and fill the blanks with null values using the mean method. The dataset is divided into training set, validation set, and test set according to the ratio of 8:1:1. Finally, the dataset is normalized and denormalized. Step 3: Generate three single neural network models: MLP, BiLSTM, and TCN. For each wind power prediction value at a point, the six actual wind power values before the prediction point, as well as the values selected after preprocessing and Pearson correlation coefficient analysis, are used as input values for experimental calculation. Step 4: During the experiment, different parameters were set for the hidden layer of each single model for training. Four parameters with good training effects were selected, resulting in a total of 12 different parameters for the three models. These parameters were: four different sets of model parameters for the three hidden layers of MLP, four different sets of model parameters for BiLSTM, and four different sets of model parameters for TCN. The prediction accuracy of each model was different at different times, and the prediction results of the 12 models were used to combine the models. Step 5: A linear programming method was used to solve the 12 different models using a CPLEX solver in the validation set to determine different weights. The linear programming method includes an objective function and constraints. The objective function is to determine the weights of each model so that the sum of the results of the models with different weights is closest to the true value. After the weights were determined in the validation set, they were tested in the test set to achieve the optimal result. Finally, the product of the obtained weights and the corresponding predicted values was added to obtain the predicted value of the combined model. Step 6: Perform error analysis on the prediction results to verify the prediction accuracy of the combined model MLP-BiLSTM-TCN. The model error metrics are the mean absolute error (MAE) and the root mean square error (RMSE). The root mean square error (RMSE) is used to measure the degree of error dispersion, and the mean absolute error (MAE) is used to evaluate the average magnitude of the prediction error. The formula for the mean absolute error (MAE) is: in, Represents the absolute mean error of any model of a single model MLP, BiLSTM or TCN, k = 1, 2, 3, that is, represents the absolute mean error of the MLP model, represents the absolute mean error of the BiLSTM model, Represents the absolute mean error of the TCN model; is the total number of data points, is the i-th observation, is the i-th predicted value; The formula for the root mean square error (RMSE) is: in, is the total number of data points, is the i-th observation, is the i-th predicted value.
[0007] In step 1, the extremely weak correlation, irrelevance, strong correlation, moderate correlation, and weak correlation of wind power are determined based on the Pearson correlation coefficient. The Pearson correlation coefficient is used to describe the degree of correlation between two variables, that is, the degree of correlation between meteorological characteristics and wind power. The meteorological characteristics include wind speed, wind direction, and temperature. The calculation formula of the Pearson correlation coefficient is: in, is the Pearson correlation coefficient, n is the length of the sample, , is the actual sample value, , is the mean of the actual sample; The value range of is (-1, 1). When the value is positive, it indicates positive correlation. When the value is negative, it indicates negative correlation. The closer the absolute value of is to 1, the higher the correlation between the influencing factors of this meteorological characteristic and wind power.
[0008] The energy coordination control mode of the PEM state strategy controller includes six modes, namely mode 1, mode 2, mode 3, mode 4, mode 5 and mode 6, among which, Mode 1: Power P of wind power generation system ne Less than the rated power P of the PEM electrolysis hydrogen production system eleWhen the battery state of charge SOC is less than 10%, that is, P ne <20%P ele , SOC<10%, the PEM state strategy controller stores all the electricity generated by the wind power generation system in the storage device and does not provide electricity to the PEM electrolysis hydrogen production system; Mode 2: Power P of wind power generation system ne Less than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC is greater than 10%, that is, P ne <20%P ele , SOC>10%, the PEM state strategy controller provides all the electricity generated by the wind power generation system to the PEM electrolysis hydrogen production system, so that the PEM electrolysis hydrogen production system is maintained at the minimum power, and the insufficient electricity is provided by the storage device; Mode 3: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 20% of the rated power P ele 120% of the ele <P ne <120%P ele , the PEM state strategy controller provides all the electricity generated by the wind power generation system to the PEM electrolysis hydrogen production system; Mode 4: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC reaches 100%, that is, P ne >120%P ele , SOC=100%, the PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power. Since the battery reaches the maximum energy storage level, the PEM state strategy controller discards the excess energy generated; Mode 5: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 120% of the battery capacity, the battery state of charge SOC does not reach 100%, and the battery charging power P b Greater than the maximum charging power P of the battery b,max When P ne >120%P ele , SOC<100%, P b >P b,maxThe PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power, and part of it to the storage device for storage to maintain the charging power of the storage device at maximum power, and discards the excess electric energy; Mode 6: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 120% of the battery capacity, the battery state of charge SOC does not reach 100%, and the battery charging power P b Less than the maximum charging power P of the battery b,max When P ne >120%P ele , SOC<100%, P b <P b,max , the PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power, and the wind power generation system provides the excess energy generated to the storage device.
[0009] The wind power generation system includes a solar photovoltaic power generation module, a wind power generation module, an inverter module and a DC-DC conversion device. The solar photovoltaic power generation module and the wind power generation module are both converted into direct current through the DC-DC conversion device and converted into alternating current through the inverter module.
[0010] The present invention discloses a predictive control method for wind power generation coupled with PEM electrolysis hydrogen production. The method is composed of a linear programming combination of the models MLP (Multi-layer Perceptron)-BiLSTM (Long Short-Term Memory Network)-TCN (Temporal Convolutional Network). The data is processed through the Pearson coefficient, mean filling method, normalization, etc. The processed data eliminates data with low correlation, reduces the amount of calculation, and improves the efficiency of calculation.
[0011] By predicting the wind power generation, an energy control strategy is added to the PEM electrolysis hydrogen production system. By detecting the wind power generation parameters and matching the operating status of the PEM electrolysis hydrogen production system, the problem of the output power of the wind power generation system failing to be maintained between the maximum power and the minimum power of the electrolysis hydrogen production system is solved. The electrolysis hydrogen production system is also unable to effectively utilize the energy generated by the wind power generation to convert it into hydrogen energy, resulting in the abandonment of this part of the energy, the frequent start and stop of the electrolysis hydrogen production system, and the low energy utilization rate is solved. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 It is a schematic diagram of the framework of the wind power generation coupled with PEM electrolysis hydrogen production system in the present invention.
[0013] Figure 2This is a flow chart of the machine learning wind power prediction process using the MLP-BiLSTM-TCN combination in the present invention.
[0014] Figure 3 This is a table showing the relationship between the value range of the Pearson correlation coefficient and the degree of correlation.
[0015] Figure 4 This is a flow chart of the control strategy for the wind power generation coupled with PEM electrolysis hydrogen production system in the present invention. DETAILED DESCRIPTION
[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the implementation of the present invention, not the entire implementation. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.
[0017] like Figure 1 As shown, a wind power generation coupled with PEM electrolysis hydrogen production prediction control method includes a wind power generation system, a wind power generation power predictor, a PEM state strategy controller, a PEM electrolysis hydrogen production system and a storage device, wherein, The wind power forecaster uses a machine learning model based on the MLP-BiLSTM-TCN combination to predict wind power. The wind power forecaster predicts the power generation at the next moment, obtains the predicted value of the combined model, and outputs the predicted value of the combined model to the PEM state strategy controller. The PEM state strategy controller adopts the energy coordination control mode according to the predicted value of the combined model to distribute the electric energy of the wind power generation system between the PEM electrolysis hydrogen production system and the storage battery, so that the power of the wind power generation system matches the power of the PEM electrolysis hydrogen production system.
[0018] The algorithm within the wind power predictor uses machine learning wind power prediction based on the MLP-BiLSTM-TCN combination. Data is processed through the Pearson coefficient, mean filling method, normalization, etc. The processed data removes data with low correlation, reduces the amount of calculation, and improves calculation efficiency. A parameter adjustment experiment is conducted on a single model, and the optimal four parameters are selected for each model to predict wind power. Finally, the 12 model parameters are combined and predicted using the linear programming method. Compared to the more common combination models currently available, this embodiment uses the prediction results of 12 models to form a combined model, which can effectively reduce the impact of deviations in a single model in the traditional combination model. It also has strong flexibility, adaptability, high prediction accuracy, and reduces prediction deviations.
[0019] like Figure 2As shown in FIG, the method for learning wind power prediction using a machine learning model based on the MLP-BiLSTM-TCN combination in the wind power forecaster includes the following steps: Step 1: Use the Pearson correlation coefficient method to perform correlation analysis on the original data set, remove meteorological information that is extremely weakly correlated or irrelevant to wind power, and select meteorological information with strong, moderate, and weak correlation as the input value after preliminary processing; Step 2: The dataset processed by the Pearson correlation coefficient method is further preprocessed to remove outliers in the dataset and fill the blanks with null values using the mean method. The dataset is divided into training set, validation set, and test set according to the ratio of 8:1:1. Finally, the dataset is normalized and denormalized. Step 3: Generate three single neural network models: MLP, BiLSTM, and TCN. For each wind power prediction value at a point, the six actual wind power values before the prediction point, as well as the values selected after preprocessing and Pearson correlation coefficient analysis, are used as input values for experimental calculation. Step 4: During the experiment, different parameters were set for the hidden layer of each single model for training. Four parameters with good training effects were selected, resulting in a total of 12 different parameters for the three models. These parameters were: four different sets of model parameters for the three hidden layers of MLP, four different sets of model parameters for BiLSTM, and four different sets of model parameters for TCN. The prediction accuracy of each model was different at different times, and the prediction results of the 12 models were used to combine the models. Step 5: A linear programming method was used to solve the 12 different models using a CPLEX solver in the validation set to determine different weights. The linear programming method includes an objective function and constraints. The objective function is to determine the weights of each model so that the sum of the results of the models with different weights is closest to the true value. After the weights were determined in the validation set, they were tested in the test set to achieve the optimal result. Finally, the product of the obtained weights and the corresponding predicted values was added to obtain the predicted value of the combined model. Step 6: Perform error analysis on the prediction results to verify the prediction accuracy of the combined model MLP-BiLSTM-TCN. The model error metrics are the mean absolute error (MAE) and the root mean square error (RMSE). The root mean square error (RMSE) is used to measure the degree of error dispersion, and the mean absolute error (MAE) is used to evaluate the average magnitude of the prediction error. The formula for the mean absolute error (MAE) is: in, represents the absolute mean error of any model of a single model MLP, BiLSTM or TCN, k = 1, 2, 3, that is, represents the absolute mean error of the MLP model, represents the absolute mean error of the BiLSTM model, Represents the absolute mean error of the TCN model; is the total number of data points, is the i-th observation, is the i-th predicted value; The formula for the root mean square error (RMSE) is: in, is the total number of data points, is the i-th observation, is the i-th predicted value.
[0020] In step 1, the extremely weak correlation, irrelevance, strong correlation, moderate correlation, and weak correlation of wind power are determined based on the Pearson correlation coefficient. The Pearson correlation coefficient is used to describe the degree of correlation between two variables, that is, the degree of correlation between meteorological characteristics and wind power. The meteorological characteristics include wind speed, wind direction, and temperature. The calculation formula of the Pearson correlation coefficient is: in, is the Pearson correlation coefficient, n is the length of the sample, , is the actual sample value, , is the mean of the actual sample; The value range of is (-1, 1). When the value is positive, it indicates positive correlation. When the value is negative, it indicates negative correlation. The closer the absolute value of is to 1, the higher the correlation between the influencing factors of this meteorological characteristic and wind power. The relationship between the value range of the Pearson correlation coefficient and the degree of correlation is shown in the table below: Figure 3 shown.
[0021] In this embodiment, the Represents meteorological characteristics, Represents wind power. Since meteorological characteristics include wind speed, wind direction and temperature, in actual calculations, it is necessary to take wind speed, wind direction and temperature as meteorological characteristic variables and calculate their correlation with wind power.
[0022] Through steps 1 to 6 above, the wind power generation power predictor outputs the power generation power at the next moment to the PEM state strategy controller. The PEM state strategy controller matches the hydrogen production power according to the built-in state control strategy, thereby improving the energy utilization rate of the PEM electrolysis hydrogen production system and reducing the number of starts and stops, reducing the start-stop energy loss of the electrolysis hydrogen production system, and making it adapt to the power fluctuation characteristics of wind power generation.
[0023] like Figure 4 As shown in FIG, the energy coordination control mode of the PEM state strategy controller includes six modes, namely mode 1, mode 2, mode 3, mode 4, mode 5 and mode 6, wherein: Mode 1: Power P of wind power generation system ne Less than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC is less than 10%, that is, P ne <20%P ele , SOC<10%, the PEM state strategy controller stores all the electricity generated by the wind power generation system in the storage device and does not provide electricity to the PEM electrolysis hydrogen production system; Mode 2: Power P of wind power generation system ne Less than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC is greater than 10%, that is, P ne <20%P ele , SOC>10%, the PEM state strategy controller provides all the electricity generated by the wind power generation system to the PEM electrolysis hydrogen production system, so that the PEM electrolysis hydrogen production system is maintained at the minimum power, and the insufficient electricity is provided by the storage device; Mode 3: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 20% of the rated power P ele 120% of the ele <P ne <120%P ele , the PEM state strategy controller provides all the electricity generated by the wind power generation system to the PEM electrolysis hydrogen production system; Mode 4: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC reaches 100%, that is, P ne >120%P ele, SOC=100%, the PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power. Since the battery reaches the maximum energy storage level, the PEM state strategy controller discards the excess energy generated; Mode 5: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 120% of the battery capacity, the battery state of charge SOC does not reach 100%, and the battery charging power P b Greater than the maximum charging power P of the battery b,max When P ne >120%P ele , SOC<100%, P b >P b,max The PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power, and part of it to the storage device for storage to maintain the charging power of the storage device at maximum power, and discards the excess electric energy; Mode 6: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 120% of the battery capacity, the battery state of charge SOC does not reach 100%, and the battery charging power P b Less than the maximum charging power P of the battery b,max When P ne >120%P ele , SOC<100%, P b <P b,max , the PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power, and the wind power generation system provides the excess energy generated to the storage device.
[0024] The wind power generation system includes a solar photovoltaic power generation module, a wind power generation module, an inverter module and a DC-DC conversion device. The solar photovoltaic power generation module and the wind power generation module are both converted into direct current through the DC-DC conversion device and converted into alternating current through the inverter module.
[0025] The present invention discloses a predictive control method for wind power generation coupled with PEM electrolysis hydrogen production. The method is composed of a linear programming combination of the models MLP (Multi-layer Perceptron)-BiLSTM (Long Short-Term Memory Network)-TCN (Temporal Convolutional Network). The data is processed through the Pearson coefficient, mean filling method, normalization, etc. The processed data eliminates data with low correlation, reduces the amount of calculation, and improves the efficiency of calculation.
[0026] A predictive control method for wind power generation coupled with PEM electrolysis for hydrogen production consists of two parts. The first part combines three single models, namely, MLP, TCN, and BiLSTM, through linear programming. Each deep learning model is assigned different hidden parameters. Six real values before the prediction point, mixed with other meteorological conditions, are used as inputs and fed into different parameters of different models. Each model then selects the four best prediction results, generating 12 models for power prediction. Finally, linear programming is used to determine different weights to optimize the results and determine the predicted power of the combined MLP-TCN-BiLSTM model. Each model has different prediction accuracy at different times. Combining the prediction results of the 12 models effectively mitigates the impact of individual model issues and reduces prediction bias.
[0027] The second part is the energy control strategy of the wind power generation coupled with PEM electrolysis hydrogen production system, including a wind power generation system, a state strategy controller, a PEM electrolysis hydrogen production system, a accumulator, and a hydrogen storage tank; the wind power generation system includes a solar photovoltaic power generation module and a wind power generation module, which respectively capture solar energy and wind energy and convert them into electrical energy. The state strategy controller supplies part of the electrical energy to the PEM electrolysis hydrogen production system and stores the other part in the accumulator. Under the action of electrical energy, water is electrolyzed into hydrogen in the PEM electrolysis hydrogen production system and stored in the hydrogen storage tank.
[0028] The problem that the wind power generation power cannot meet the minimum power of the PEM electrolysis hydrogen production system, the PEM electrolysis hydrogen production system will be unable to produce hydrogen and will be forced to shut down; the wind power generation power is too high, greater than the maximum power of the PEM electrolysis hydrogen production system, reaching overload, and being in an overload state for a long time will cause the system components to overheat, resulting in damage to the proton exchange membrane, affecting the stability and service life of the electrolysis hydrogen production system, and causing safety hazards.
[0029] The technical means disclosed in the solutions of the present invention are not limited to those disclosed in the above-mentioned embodiments, but also include technical solutions composed of any combination of the above-mentioned technical features. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A predictive control method for wind power generation coupled with PEM electrolysis hydrogen production, characterized by: It includes a wind power generation system, a wind power generation power predictor, a PEM state strategy controller, a PEM electrolysis hydrogen production system and a storage device; wherein, The wind power forecaster uses a machine learning model based on the MLP-BiLSTM-TCN combination to predict wind power. The wind power forecaster predicts the power generation at the next moment, obtains the predicted value of the combined model, and outputs the predicted value of the combined model to the PEM state strategy controller. The PEM state strategy controller adopts the energy coordination control mode according to the predicted value of the combined model to distribute the electric energy of the wind power generation system between the PEM electrolysis hydrogen production system and the storage battery, so that the power of the wind power generation system matches the power of the PEM electrolysis hydrogen production system.
2. The wind power generation coupled with PEM electrolysis hydrogen production prediction control method according to claim 1, characterized in that: The wind power prediction method using a machine learning model based on the MLP-BiLSTM-TCN combination in the wind power forecaster includes the following steps: Step 1: Use the Pearson correlation coefficient method to perform correlation analysis on the original data set, remove meteorological information that is extremely weakly correlated or irrelevant to wind power, and select meteorological information with strong, moderate, and weak correlation as the input value after preliminary processing; Step 2: The dataset processed by the Pearson correlation coefficient method is further preprocessed to remove outliers in the dataset and fill the blanks with null values using the mean method. The dataset is divided into training set, validation set, and test set according to the ratio of 8:1:
1. Finally, the dataset is normalized and denormalized. Step 3: Generate three single neural network models: MLP, BiLSTM, and TCN. For each wind power prediction value at a point, the six actual wind power values before the prediction point, as well as the values selected after preprocessing and Pearson correlation coefficient analysis, are used as input values for experimental calculation. Step 4: During the experiment, different parameters were set for the hidden layer of each single model for training. Four parameters with good training effects were selected, resulting in a total of 12 different parameters for the three models. These parameters were: four different sets of model parameters for the three hidden layers of MLP, four different sets of model parameters for BiLSTM, and four different sets of model parameters for TCN. The prediction accuracy of each model was different at different times, and the prediction results of the 12 models were used to combine the models. Step 5: Use linear programming method to solve different weights for 12 different models using CPLEX solver in the validation set; The linear programming method includes an objective function and constraints. The objective function is to find the weights of each model so that the sum of the results of the models with different weights is closest to the true value. After the weights are found in the validation set, they are tested in the test set to achieve the optimal result. Finally, the product of the found weights and the corresponding predicted values is added to obtain the predicted value of the combined model. Step 6: Perform error analysis on the prediction results to verify the prediction accuracy of the combined model MLP-BiLSTM-TCN. The model error metrics are the mean absolute error (MAE) and the root mean square error (RMSE). The root mean square error (RMSE) is used to measure the degree of error dispersion, and the mean absolute error (MAE) is used to evaluate the average magnitude of the prediction error. The formula for the mean absolute error (MAE) is: in, Represents the absolute mean error of any model of a single model MLP, BiLSTM or TCN, k = 1, 2, 3, that is, represents the absolute mean error of the MLP model, represents the absolute mean error of the BiLSTM model, Represents the absolute mean error of the TCN model; is the total number of data points, is the i-th observation, is the i-th predicted value; The formula for the root mean square error (RMSE) is: in, is the total number of data points, is the i-th observation, is the i-th predicted value.
3. The wind power generation coupled with PEM electrolysis hydrogen production prediction control method according to claim 2, characterized in that: In step 1, the extremely weak correlation, irrelevance, strong correlation, moderate correlation, and weak correlation of wind power are determined based on the Pearson correlation coefficient. The Pearson correlation coefficient is used to describe the degree of correlation between two variables, that is, the degree of correlation between meteorological characteristics and wind power. The meteorological characteristics include wind speed, wind direction, and temperature. The calculation formula of the Pearson correlation coefficient is: in, is the Pearson correlation coefficient, n is the length of the sample, , is the actual sample value, , is the mean of the actual sample; The value range of is (-1, 1). When the value is positive, it indicates positive correlation. When the value is negative, it indicates negative correlation. The closer the absolute value of is to 1, the higher the correlation between the influencing factors of this meteorological characteristic and wind power.
4. The wind power generation coupled with PEM electrolysis hydrogen production prediction control method according to claim 1, characterized in that: The energy coordination control mode of the PEM state strategy controller includes six modes, namely mode 1, mode 2, mode 3, mode 4, mode 5 and mode 6, among which, Mode 1: Power P of wind power generation system ne Less than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC is less than 10%, that is, P ne <20%P ele , SOC<10%, the PEM state strategy controller stores all the electricity generated by the wind power generation system in the storage device and does not provide electricity to the PEM electrolysis hydrogen production system; Mode 2: Power P of wind power generation system ne Less than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC is greater than 10%, that is, P ne <20%P ele , SOC>10%, the PEM state strategy controller provides all the electricity generated by the wind power generation system to the PEM electrolysis hydrogen production system, so that the PEM electrolysis hydrogen production system is maintained at the minimum power, and the insufficient electricity is provided by the storage device; Mode 3: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 20% of the rated power P ele 120% of the ele <P ne <120%P ele , the PEM state strategy controller provides all the electricity generated by the wind power generation system to the PEM electrolysis hydrogen production system; Mode 4: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele When the battery state of charge SOC reaches 100%, that is, P ne >120%P ele , SOC=100%, the PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power. Since the battery reaches the maximum energy storage level, the PEM state strategy controller discards the excess energy generated; Mode 5: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 120% of the battery capacity, the battery state of charge SOC does not reach 100%, and the battery charging power P b Greater than the maximum charging power P of the battery b,max When P ne >120%P ele , SOC<100%, P b >P b,max The PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power, and part of it to the storage device for storage to maintain the charging power of the storage device at maximum power, and discards the excess electric energy; Mode 6: Power P of wind power generation system ne Greater than the rated power P of the PEM electrolysis hydrogen production system ele 120% of the battery capacity, the battery state of charge SOC does not reach 100%, and the battery charging power P b Less than the maximum charging power P of the battery b,max When P ne >120%P ele , SOC<100%, P b <P b,max , the PEM state strategy controller provides part of the electric energy generated by the wind power generation system to the PEM electrolysis hydrogen production system to maintain it at maximum power, and the wind power generation system provides the excess energy generated to the storage device.
5. The wind power generation coupled with PEM electrolysis hydrogen production prediction and control method according to claim 1, characterized in that: The wind power generation system includes a solar photovoltaic power generation module, a wind power generation module, an inverter module and a DC-DC conversion device. The solar photovoltaic power generation module and the wind power generation module are both converted into direct current through the DC-DC conversion device and converted into alternating current through the inverter module.
6. The wind power generation coupled with PEM electrolysis hydrogen production prediction control method according to claim 1, characterized in that: The PEM state strategy controller supplies part of the electrical energy to the PEM electrolysis hydrogen production system and stores the other part in the accumulator. Water is electrolyzed into hydrogen in the PEM electrolysis hydrogen production system. The main structure of the PEM electrolysis hydrogen production system is a PEM electrolyzer, which also includes a hydrogen separator, a dryer and a hydrogen storage tank. After passing through the hydrogen separator and the dryer, the hydrogen is stored in the hydrogen storage tank, and the excess water and oxygen are directly discharged.