New energy automobile electricity hydrogen demand spatial-temporal distribution prediction method

Abstract

The invention discloses a new energy automobile electricity hydrogen demand spatial-temporal distribution prediction method. A new energy automobile electricity hydrogen demand spatial-temporal distribution prediction model is realized by combining a diffusion model and a CNN-LSTM deep learning model. The diffusion model is suitable for predicting the long-term change trend of the sequence data, and can effectively predict the annual inventory of the electric vehicle and the hydrogen fuel vehicle; the CNN-LSTM deep learning model has strong learning ability for the short-term change process of the sequence, and can predict the daily charging demand change of the electric vehicle and the hydrogen fuel vehicle; by combining the annual inventory increase trend of the electric vehicle and the hydrogen fuel vehicle with the daily energy charging demand change prediction, the energy charging demand change prediction under the target year is realized, and a technical support is provided for the planning of the electricity-hydrogen production and charging integrated station.

Description

Technical Field

[0001] The present invention belongs to the technical field of prediction of electric hydrogen demand for new energy vehicles, and specifically relates to a method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles. Background Art

[0002] Transportation is a basic, leading, strategic industry and an important service industry in the national economy, and it is also one of the important areas of carbon emissions. In order to better achieve carbon reduction and emission reduction in the transportation industry, building a new transportation system with new energy vehicles as the main body is the only way to achieve the "dual carbon" goal. At present, the number of electric vehicles and hydrogen fuel vehicles in my country is increasing rapidly, but the construction of charging piles and hydrogen refueling stations is relatively lagging, and the market volume is still small. And how to integrate charging with hydrogen refueling and build an integrated station for electric hydrogen production and filling will become an important direction for the transformation and development of low-carbon energy. However, the planning of integrated stations for electric hydrogen production and filling requires accurate prediction of the temporal and spatial distribution of electric hydrogen demand for new energy vehicles. At present, although some progress has been made in the field of electric hydrogen demand prediction for new energy vehicles, there is still a lack of prediction of the temporal and spatial distribution of electric hydrogen demand for new energy vehicles for the planning of integrated stations for electric hydrogen production and filling. Therefore, it is necessary to propose an innovative and more reliable prediction method to provide data support for the planning of integrated stations for electric hydrogen production and filling. Summary of the invention

[0003] The purpose of the present invention is to propose a method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles, to predict the growth trend of the ownership of electric vehicles and hydrogen fuel vehicles, and to predict the changes in daily charging demand for electric vehicles and hydrogen fuel vehicles through a deep learning model.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles is as follows:

[0006] S1: Extract the historical annual ownership data of electric vehicles and hydrogen fuel cell vehicles, historical meteorological data, and historical daily charging demand data of electric vehicles and hydrogen fuel cell vehicles at different traffic nodes in the city where the planning area is located;

[0007] S2: Fitting historical electric vehicle ownership data using different diffusion models;

[0008] S3: According to the mean absolute errors of the different diffusion models and the electric vehicles, the optimal diffusion model with the smallest error is selected;

[0009] S4: Using the optimal diffusion model in S3, a prediction model for the annual ownership of electric vehicles and hydrogen fuel vehicles is established to predict the annual ownership of electric vehicles and hydrogen fuel vehicles;

[0010] S5: Based on the annual stock of electric vehicles and hydrogen fuel cell vehicles, calculate the annual growth rate of electric vehicles and hydrogen fuel cell vehicles;

[0011] S6: Pearson correlation coefficient is used to conduct correlation analysis on historical daily vehicle charging demand data and meteorological data at different traffic nodes, and meteorological factors with low correlation coefficient are eliminated;

[0012] S7: Build a CNN-LSTM model based on convolutional neural network (CNN) and long short-term memory (LSTM) network;

[0013] S8: Based on S6, the daily charging demand data and meteorological data of electric vehicles and hydrogen fuel vehicles are normalized, and the normalized data are input into the CNN-LSTM model to establish a spatiotemporal distribution prediction model for the charging demand of electric vehicles and hydrogen fuel vehicles;

[0014] S9: train the prediction model in S8 and predict the charging demand of electric vehicles and hydrogen fuel vehicles at different traffic nodes on a typical day;

[0015] S10: Combine the annual growth rate of electric vehicles and hydrogen fuel cell vehicles in S5 with the daily charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes to predict the charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes on a typical day in the target year.

[0016] Furthermore, the annual ownership data in S1 includes the annual ownership data of electric vehicles and the annual ownership data of hydrogen fuel vehicles; the historical meteorological data includes temperature, humidity, air pressure, and wind speed; the daily charging demand data includes the historical daily charging demand of electric vehicles at different traffic nodes and the daily charging demand of hydrogen fuel vehicles at different traffic nodes.

[0017] Furthermore, in S1, a prediction model for the number of electric vehicles and hydrogen fuel vehicles is established based on the Bass, Logistic, and Gompertz diffusion models; the Bass, Logistic, and Gompertz diffusion models are fitted using historical electric vehicle ownership data;

[0018] The Bass diffusion model is as follows:

[0019]

[0020] Where: N Bass (d) is the number of cars fitted by Bass in year d; e is the natural constant; p is the innovation coefficient; q is the imitation coefficient; m is the potential market demand;

[0021] The logistic diffusion model is as follows:

[0022]

[0023] Where: N Logistic (d) is the number of cars in the logistic model in year d; a is the slope factor, b is the growth rate; c is the potential market demand; and e is a natural constant.

[0024] The Gompertz diffusion model is as follows:

[0025]

[0026] Where: N Gompertz (d) is the number of cars in the country fitted by Gompertz in year d; j is the growth rate; k is the time point when half of the potential market demand is reached; n is the potential market demand; and e is a natural constant.

[0027] Furthermore, in S3, the cumulative absolute errors of the Bass, Logistic, and Gompertz diffusion models in fitting the number of electric vehicles and hydrogen fuel vehicles are calculated; according to the size of the cumulative absolute errors, the diffusion model with the smallest cumulative absolute error is selected as the optimal diffusion model;

[0028] The calculation method of the cumulative absolute error of the holdings is as follows:

[0029]

[0030] Where: E is the cumulative absolute error; D is the number of forecast years; N′(d) is the actual value of the number of cars in year d, and N(d) is the forecast value of the number of cars in year d.

[0031] Furthermore, the specific method in S6 is: the first step: calculating the Pearson correlation coefficient of the historical daily charging demand data of electric vehicles and hydrogen fuel vehicles and the historical meteorological data; the second step: according to the calculation result of the Pearson correlation coefficient, eliminating the meteorological factors with a correlation coefficient less than 0.2.

[0032] Furthermore, the method in S7 is constructed sequentially as follows:

[0033] 1) Construct the convolutional layer of the CNN-LSTM model;

[0034]

[0035] Where: x t is the input feature of the CNN-LSTM model; F t For x t Features extracted by the convolutional layer; is the convolution operation; W g is the weight matrix of the convolutional layer; g is the bias vector; fRelu (x) is the Relu activation function;

[0036] 2) Construct the forget gate of the CNN-LSTM model;

[0037] f t =σ(W f *[F t ,h t-1 ] T +B f )(8)

[0038] Where: X t represents the input value of the hidden layer at time t; h t-1 represents the output value of the hidden layer at time t-1; σ is the activation function sigmoid, and the output value is between 0 and 1; W f is the weight matrix of the forget gate; B f is the forget gate bias; f t is the output value of the forget gate at time t, and T is used to represent the matrix transpose;

[0039] 3) Construct the input gate of the CNN-LSTM model;

[0040]

[0041] Where tanh represents the hyperbolic tangent activation function, and the output value is between -1 and 1; W i is the weight matrix of the input gate; B i is the input gate bias term; i t is the output value of the input gate at time t; C t ' is the temporary state of the memory unit at time t; B C is the memory unit bias term; it represents the corresponding multiplication operation of matrix elements; C t is the output of the memory unit at time t, and T is used to represent the matrix transpose;

[0042] 4) Construct the output gate of the CNN-LSTM model;

[0043]

[0044] Where: W o is the weight matrix of the output gate; B o is the output gate bias term; O t is the output value of the output gate at time t, and T is used to represent matrix transpose.

[0045] Furthermore, the annual growth rate of electric vehicles and hydrogen fuel cell vehicles in S5 is calculated as follows:

[0046]

[0047] Where: α d is the annual growth rate of car ownership; N(d) and N(d-1) are the car ownership in year d and year d-1.

[0048] Further, the Pearson correlation coefficient is as follows;

[0049]

[0050] Where: t is a certain moment; T is the time series X t , Y t The total number of data; The time series X t , Y t The average value of ρ XY ∈[-1,1] and the larger the absolute value, the stronger the correlation.

[0051] Furthermore, the specific steps in S8 are:

[0052] 1) Based on S6, the daily charging demand data and meteorological data of electric vehicles and hydrogen fuel vehicles are normalized;

[0053]

[0054] In the formula: x represents the data after normalization; Indicates the actual value of a certain type of data; and Indicates the maximum and minimum values ​​of a certain type of data;

[0055] 2) The normalized data is input into CNN-LSTM to establish a spatiotemporal distribution prediction model for the charging demand of electric vehicles and hydrogen fuel vehicles; the model input includes the meteorological factors at time t on the predicted day, the charging demand and hydrogen filling demand of a certain traffic node at time t 7 days before the predicted day, and the model output is the predicted charging demand and hydrogen filling demand of a certain traffic node at time t on the predicted day.

[0056] Furthermore, the annual growth rate of electric vehicles and hydrogen fuel cell vehicles in S5 is combined with the daily charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes to predict the charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes on a typical day in the target year:

[0057] x d (t) = [1 + α d ]x d-1 (t)(12)

[0058] In the formula, x d (t) and x d-1 (t) is the predicted vehicle charging demand at a certain traffic node at time t in year d and year d-1; αd is the annual growth rate of car ownership in year d

[0059] The following beneficial effects can be obtained through the above technical solution:

[0060] The present invention combines the diffusion model and the CNN-LSTM deep learning model to realize a spatiotemporal distribution prediction model for the electric hydrogen demand of new energy vehicles. The diffusion model is suitable for predicting the long-term change trend of sequence data, and can effectively predict the annual ownership of electric vehicles and hydrogen fuel vehicles; the CNN-LSTM deep learning model has a strong learning ability for the short-term change process of the sequence, and can predict the daily charging demand changes of electric vehicles and hydrogen fuel vehicles; by combining the annual ownership growth trend of electric vehicles and hydrogen fuel vehicles with the prediction of daily charging demand changes, the prediction of charging demand changes in the target year is realized, which provides technical support for the planning of electric hydrogen production and filling integrated stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is a flow chart of the method in the present invention;

[0062] Figure 2 The number of electric vehicles in a city predicted by the optimal diffusion model;

[0063] Figure 3 The number of electric vehicles in a city;

[0064] Figure 4 The number of hydrogen fuel cell vehicles in a city predicted by the optimal diffusion model;

[0065] Figure 5 The predicted charging demand for electric vehicles at each traffic node on a typical day in a target year in a certain region; Figure 6 The predicted hydrogen filling demand for hydrogen fuel cell vehicles at each traffic node on a typical day in a target year in a certain region. DETAILED DESCRIPTION

[0066] The following is attached Figure 1-5 The present invention is further described with reference to the following embodiments:

[0067] This implementation example provides a new energy vehicle electric hydrogen demand spatiotemporal distribution prediction model, such as Figure 1 As shown, the following steps are included:

[0068] Step 1: Extract the historical annual ownership data of electric vehicles and hydrogen fuel cell vehicles, historical meteorological data, and historical daily charging demand data of electric vehicles and hydrogen fuel cell vehicles at different traffic nodes in the city where the planning area is located;

[0069] Extract the historical annual ownership data of electric vehicles and hydrogen fuel cell vehicles in the city where the planning area is located, including: annual ownership data of electric vehicles and annual ownership data of hydrogen fuel cell vehicles; extract the historical meteorological data of the planning area, including: temperature, humidity, air pressure, wind speed, etc.; extract the historical daily charging demand data of electric vehicles and hydrogen fuel cell vehicles at different traffic nodes in the planning area, including: historical daily charging demand of electric vehicles at different traffic nodes, and daily charging demand of hydrogen fuel cell vehicles at different traffic nodes.

[0070] Step 2: Use different diffusion models to fit historical electric vehicle and hydrogen fuel vehicle ownership data;

[0071] 1) Based on the Bass, Logistic, and Gompertz diffusion models, a prediction model for the number of electric vehicles and hydrogen fuel cell vehicles is established;

[0072]

[0073] Where: N Bass (d) is the number of cars fitted by Bass in year d; e is the natural constant; p is the innovation coefficient; q is the imitation coefficient; and m is the potential market demand.

[0074]

[0075] Where: N Logistic (d) is the number of cars in the logistic model in year d; a is the slope factor, b is the growth rate; c is the potential market demand.

[0076]

[0077] Where: N Gompertz (d) is the number of cars in the country fitted by Gompertz in year d; j is the growth rate; k is the time point when half of the potential market demand is reached; n is the potential market demand, and e is a natural constant.

[0078] 2) Using historical electric vehicle and hydrogen fuel cell vehicle ownership data, the Bass, Logistic, and Gompertz diffusion models are fitted.

[0079] Step 3: According to the mean absolute errors of the different diffusion models and the electric vehicle fitting, the optimal diffusion model with the smallest error is selected;

[0080] 1) Calculate the cumulative absolute error of the Bass, Logistic, and Gompertz diffusion models in fitting the number of electric vehicles and hydrogen fuel vehicles;

[0081]

[0082] Where: E is the mean absolute error; D is the number of forecast years; N′(d) is the true value of the inventory in year d.

[0083] 2) According to the size of the cumulative absolute error, the diffusion model with the smallest cumulative absolute error is selected as the optimal diffusion model;

[0084] Taking the historical ownership data of electric vehicles and hydrogen fuel vehicles in a certain region as an example, the above model is used for fitting. According to the mean absolute error, Gompertz has the smallest mean absolute error, so Gompertz is used as the optimal diffusion model for ownership prediction.

[0085] Table 1 Cumulative absolute errors of different models for fitting electric vehicles and hydrogen fuel cell vehicles

[0086] category Bass Logistic Gompertz Electric Vehicles 11,900 vehicles 7,500 vehicles 7,300 vehicles Hydrogen fuel cell vehicles 0.026 thousand vehicles 0.021 million vehicles 0.017 thousand vehicles

[0087] Step 4: Select the optimal Gompertz diffusion model mentioned above, establish an annual ownership prediction model for electric vehicles and hydrogen fuel vehicles, and predict the annual ownership of electric vehicles and hydrogen fuel vehicles. Figure 2 The number of electric vehicles in a city is predicted by the optimal diffusion model. Figure 3 The number of hydrogen fuel cell vehicles in a city predicted by the optimal diffusion model.

[0088] Step 5: Calculate the annual growth rate of electric vehicles and hydrogen fuel cell vehicles based on the annual number of electric vehicles and hydrogen fuel cell vehicles;

[0089]

[0090] Where: α d is the annual growth rate of car ownership; N(d) and N(d-1) are the car ownership in year d and year d-1.

[0091] Step 6: Use the Pearson correlation coefficient to conduct a correlation analysis on the historical daily vehicle charging demand data and meteorological data at different traffic nodes, and eliminate meteorological factors with low correlation coefficients;

[0092] 1) Calculate the Pearson correlation coefficient between the daily charging demand data of historical electric vehicles and hydrogen fuel vehicles and the historical meteorological data;

[0093]

[0094] Where: t is a certain moment; T is the time series X t , Y t The total number of data; The time series X t , Y t The average value of ρXY ∈[-1,1] and the larger the absolute value, the stronger the correlation.

[0095] The Pearson correlation coefficient between the charging demand of electric vehicles and hydrogen fuel vehicles in a certain area and meteorological factors is used as an example for analysis. From the Pearson correlation coefficient, it can be seen that electric vehicles and hydrogen fuel vehicles have a certain correlation with temperature, humidity, and air pressure, but have a low correlation with wind speed, so wind speed is excluded.

[0096] Table 1 Pearson correlation coefficient between vehicle charging demand and meteorological factors

[0097] Car Category Temperature Relative humidity Air pressure Wind speed Electric Vehicles 0.5681 0.3884 0.2137 -0.0146 Hydrogen fuel cell vehicles 0.5422 0.3269 0.2376 -0.0096

[0098] 2) Based on the calculation results of the Pearson correlation coefficient, meteorological factors with correlation coefficients less than 0.2 were eliminated.

[0099] Step 7: Build a CNN-LSTM model based on convolutional neural network (CNN) and long short-term memory network (LSTM); the specific construction method is as follows:

[0100] 1) Construct the convolutional layer of the CNN-LSTM model;

[0101]

[0102] Where: x t is the input feature of the CNN-LSTM model; F t For x t Features extracted by the convolutional layer; is the convolution operation; W g is the weight matrix of the convolutional layer; g is the bias vector; f Relu (x) is the Relu activation function.

[0103] 2) Construct the forget gate of the CNN-LSTM model;

[0104] f t =σ(W f *[F t ,h t-1 ] T +B f )(8)

[0105] Where: X t represents the input value of the hidden layer at time t; h t-1 represents the output value of the hidden layer at time t-1; σ is the activation function sigmoid, and the output value is between 0 and 1; W f is the weight matrix of the forget gate; B f is the forget gate bias; f t is the output value of the forget gate at time t.

[0106] 3) Construct the input gate of the CNN-LSTM model;

[0107]

[0108] Where tanh represents the hyperbolic tangent activation function, and the output value is between -1 and 1; W i is the weight matrix of the input gate; B i is the input gate bias term; i t is the output value of the input gate at time t; C t ′ is the temporary state of the memory unit at time t; W C is the weight matrix of the memory unit; B C is the memory unit bias term; it represents the corresponding multiplication operation of matrix elements; C t is the output of the memory unit at time t, C t-1 Represents the output of the LSTM memory unit at time t-1. At time t, LSTM can combine the output obtained by training at time t with the output at time t-1, thereby effectively capturing the relationship in long time series, which is one of the characteristics of LSTM.

[0109] 4) Construct the output gate of the CNN-LSTM model;

[0110]

[0111] Where: W o is the weight matrix of the output gate; B o is the output gate bias term; O t is the output value of the output gate at time t.

[0112] Step 8: Based on step 6, normalize the daily charging demand data and meteorological data of electric vehicles and hydrogen fuel vehicles;

[0113]

[0114] In the formula, x represents the normalized data. Represents the actual value of a certain type of data. and Indicates the maximum and minimum values ​​of a certain type of data;

[0115] The normalized data is input into CNN-LSTM to establish a spatiotemporal distribution prediction model for the charging demand of electric vehicles and hydrogen fuel vehicles. The model input includes the meteorological factors at time t on the day of prediction, the charging demand and hydrogen filling demand of a certain traffic node at time t 7 days before the day of prediction, and the model output is the charging demand and hydrogen filling demand of a certain traffic node at time t on the day of prediction.

[0116] Step 9: Train CNN-LSTM based on the prediction model established in step 8, and predict the charging demand of electric vehicles and hydrogen fuel vehicles at different traffic nodes on a typical day.

[0117] Step 10: Combine the annual growth rate of electric vehicles and hydrogen fuel cell vehicles in step 5 with the daily charging demand of electric vehicles and hydrogen fuel cell vehicles at transportation nodes to predict the charging demand of electric vehicles and hydrogen fuel cell vehicles at transportation nodes on a typical day in the target year.

[0118] x d (t) = [1 + α d ]x d-1 (t)(12)

[0119] In the formula, x d (t) and x d-1 (t) is the predicted vehicle charging demand at a certain traffic node at time t in year d and year d-1; α d is the annual growth rate of car ownership in year d. Figure 4 The predicted charging demand for electric vehicles at each traffic node on a typical day in a target year in a certain region. Figure 5 The predicted hydrogen filling demand for hydrogen fuel vehicles at each traffic node on a typical day in a target year in a certain region.

[0120] The above are all preferred embodiments of the present invention. For ordinary technicians in this technical field, without departing from the principle of the present invention, various equivalent modifications to the present invention belong to the protection scope of the claims attached to this application.

Claims

1. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles. Features: Here’s how: S1: Extract the historical annual ownership data of electric vehicles and hydrogen fuel cell vehicles, historical meteorological data, and historical daily charging demand data of electric vehicles and hydrogen fuel cell vehicles at different traffic nodes in the city where the planning area is located; S2: Fitting historical electric vehicle ownership data using different diffusion models; S3: According to the mean absolute errors of the different diffusion models and the electric vehicles, the optimal diffusion model with the smallest error is selected; S4: Using the optimal diffusion model in S3, a prediction model for the annual ownership of electric vehicles and hydrogen fuel vehicles is established to predict the annual ownership of electric vehicles and hydrogen fuel vehicles; S5: Based on the annual stock of electric vehicles and hydrogen fuel cell vehicles, calculate the annual growth rate of electric vehicles and hydrogen fuel cell vehicles; S6: Pearson correlation coefficient is used to conduct correlation analysis on historical daily vehicle charging demand data and meteorological data at different traffic nodes, and meteorological factors with low correlation coefficient are eliminated; S7: Build a CNN-LSTM model based on convolutional neural network (CNN) and long short-term memory (LSTM) network; S8: Based on S6, the daily charging demand data and meteorological data of electric vehicles and hydrogen fuel vehicles are normalized, and the normalized data are input into the CNN-LSTM model to establish a spatiotemporal distribution prediction model for the charging demand of electric vehicles and hydrogen fuel vehicles; S9: train the prediction model in S8 and predict the charging demand of electric vehicles and hydrogen fuel vehicles at different traffic nodes on a typical day; S10: Combine the annual growth rate of electric vehicles and hydrogen fuel cell vehicles in S5 with the daily charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes to predict the charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes on a typical day in the target year.

2. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1, Features: The annual ownership data in S1 includes the annual ownership data of electric vehicles and the annual ownership data of hydrogen fuel vehicles; the historical meteorological data includes temperature, humidity, air pressure, and wind speed; the daily charging demand data includes the historical daily charging demand of electric vehicles at different traffic nodes and the daily charging demand of hydrogen fuel vehicles at different traffic nodes.

3. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1, Features: In the S1, a prediction model for the number of electric vehicles and hydrogen fuel cell vehicles is established based on the Bass, Logistic and Gompertz diffusion models; and the Bass, Logistic and Gompertz diffusion models are fitted using historical electric vehicle ownership data.

4. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 3, Features: The Bass diffusion model is as follows: Where: N Bass (d) is the number of cars fitted by Bass in year d; e is the natural constant; p is the innovation coefficient; q is the imitation coefficient; and m is the potential market demand.

5. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 3, Features: The logistic diffusion model is as follows: Where: N Logistic (d) is the number of cars in the logistic model in year d; a is the slope factor, b is the growth rate; c is the potential market demand, and e is a natural constant.

6. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 3, Features: The Gompertz diffusion model is as follows: Where: N Gompertz (d) is the number of cars in the country fitted by Gompertz in year d; j is the growth rate; k is the time point when half of the potential market demand is reached; n is the potential market demand, and e is a natural constant.

7. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1, Features: In S3, the cumulative absolute error of the Bass, Logistic, and Gompertz diffusion models in fitting the number of electric vehicles and hydrogen fuel vehicles is calculated; according to the size of the cumulative absolute error, the diffusion model with the smallest cumulative absolute error is selected as the optimal diffusion model; the cumulative absolute error calculation method of the number of vehicles is as follows: Where: E is the cumulative absolute error; D is the number of forecast years; N′(d) is the actual value of the number of cars in year d, and N(d) is the forecast value of the number of cars in year d.

8. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1, Features: The specific method in S6 is as follows: Step 1: Calculate the Pearson correlation coefficient between the historical daily charging demand data of electric vehicles and hydrogen fuel vehicles and the historical meteorological data; Step 2: Based on the calculation results of the Pearson correlation coefficient, eliminate meteorological factors with correlation coefficients less than 0.

2.

9. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1, Features: The method in S7 is constructed sequentially as follows: 1) Construct the convolutional layer of the CNN-LSTM model; Where: x t It is the input feature of the CNN-LSTM model; F t For x t Features extracted by the convolutional layer; is the convolution operation; W g is the weight matrix of the convolutional layer; g is the bias vector; f Relu (x) is the Relu activation function; 2) Construct the forget gate of the CNN-LSTM model; f t =σ(W f *[F t ,h t-1 ] T +B f ) (8) Where: X t represents the input value of the hidden layer at time t; h t-1 represents the output value of the hidden layer at time t-1; σ is the activation function sigmoid, and the output value is between 0 and 1; W f is the weight matrix of the forget gate; B f is the forget gate bias; f t is the output value of the forget gate at time t, and T is used to represent the matrix transpose; 3) Construct the input gate of the CNN-LSTM model; Where tanh represents the hyperbolic tangent activation function, and the output value is between -1 and 1; W i is the weight matrix of the input gate; B i is the input gate bias term; i t is the output value of the input gate at time t; C′ t is the temporary state of the memory unit at time t; B C is the memory unit bias term; ⊙ represents the corresponding multiplication operation of matrix elements; C t is the output of the memory unit at time t, and T is used to represent the matrix transpose; 4) Construct the output gate of the CNN-LSTM model; Where: W o is the weight matrix of the output gate; B o is the output gate bias term; O t is the output value of the output gate at time t, and T is used to represent matrix transpose.

10. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1 or 3, Features: The annual growth rate of electric vehicles and hydrogen fuel cell vehicles in S5 is calculated as follows: Where: α d is the annual growth rate of car ownership; N(d) and N(d-1) are the car ownership in year d and year d-1.

11. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1 or 8, Features: The Pearson correlation coefficient is as follows; Where: t is a certain moment; T is the time series X t , Y t The total number of data; The time series X t , Y t The average value of ρ XY ∈[-1,1] and the larger the absolute value, the stronger the correlation.

12. A method for predicting the temporal and spatial distribution of electric hydrogen demand for new energy vehicles according to claim 1, Features: The specific steps in S8 are: 1) Based on S6, the daily charging demand data and meteorological data of electric vehicles and hydrogen fuel vehicles are normalized; In the formula: x represents the data after normalization; Indicates the actual value of a certain type of data; and Indicates the maximum and minimum values ​​of a certain type of data; 2) The normalized data is input into CNN-LSTM to establish a spatiotemporal distribution prediction model for the charging demand of electric vehicles and hydrogen fuel vehicles; the model input includes the meteorological factors at time t on the predicted day, the charging demand and hydrogen filling demand of a certain traffic node at time t 7 days before the predicted day, and the model output is the predicted charging demand and hydrogen filling demand of a certain traffic node at time t on the predicted day.

13. A method for predicting the spatiotemporal distribution of electric hydrogen demand for new energy vehicles according to claim 1, Features: Combining the annual growth rate of electric vehicles and hydrogen fuel cell vehicles in S5 with the daily charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes, the charging demand of electric vehicles and hydrogen fuel cell vehicles at traffic nodes on a typical day in the target year is predicted: x d (t)=[1+α d ]x d-1 (t) (12) In the formula, x d (t) and x d-1 (t) is the predicted vehicle charging demand at a certain traffic node at time t in year d and year d-1; α d is the annual growth rate of car ownership in year d.

Images