A method for predicting instantaneous energy consumption of fuel cell vehicles based on integrated model

Abstract

The present invention discloses a method for predicting instantaneous energy consumption of fuel cell vehicles based on an integrated model. First, sensors are used to acquire collected information, which is then processed to obtain calculated data. The collected information and calculated data are used together as preprocessed data. Second, the preprocessed data is time-aligned to obtain time-aligned data, which is then divided into an initial feature set and model labels. Feature extraction and data normalization are then performed on the initial feature set to obtain model features. The model features and model labels are then imported into the original integrated model for network training to obtain an optimized integrated model. Finally, the output of the optimized integrated model is post-processed using a hydrogen-to-electricity conversion method to obtain instantaneous energy consumption values ​​at the fuel cell end and the power battery end. This method solves the problem of difficult instantaneous energy consumption prediction for fuel cell vehicles and power battery ends, providing a basis for accurately estimating the driving range of fuel cell vehicles.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cell vehicles, and in particular to a method for predicting instantaneous energy consumption of fuel cell vehicles based on an integrated model. Background Art

[0002] Fuel cell vehicles offer advantages such as environmental friendliness, long driving range, and high energy conversion efficiency. Currently, most fuel cell vehicles utilize a hybrid drive system, combining a fuel cell with a power battery as a secondary power source. The additional power provided by the power battery pack enables the vehicle to handle a variety of complex operating conditions, but it also presents the challenge of difficult energy consumption prediction. Traditional fuel cell vehicles and pure electric vehicles have a single energy source, making energy consumption calculation units easily standardized and energy consumption prediction relatively simple. However, fuel cell vehicles have complex energy management strategies, resulting in variations in energy distribution under different operating conditions. Furthermore, under the same operating conditions, energy distribution varies due to the varying states of charge of the power battery pack, making accurate energy consumption prediction difficult.

[0003] Chinese patent publication CN110549915B discloses a range estimation method for hydrogen fuel cell hybrid new energy vehicles. This method estimates the range of a vehicle by real-time monitoring of the energy status of the power battery, supercapacitor, and fuel cell, combined with the vehicle's energy consumption rate. This method estimates energy consumption from the fuel cell, power battery, and supercapacitor sides, addressing user range anxiety. However, this method only estimates the average energy consumption of fuel cell vehicles and cannot predict instantaneous energy consumption.

[0004] Chinese patent publication number CN110789352A discloses a method and device for displaying average and instantaneous energy consumption for fuel cell vehicles. This device calculates and displays vehicle energy consumption for each operating condition, based on pure electric operation (with the fuel cell disabled) and hybrid operation (with the fuel cell enabled). This device predicts fuel cell vehicle energy consumption under multiple operating conditions. However, this method relies solely on internal vehicle sensor data, calculating energy consumption under different operating conditions using a formula. It also fails to consider the impact of the vehicle's external environment and road conditions, making it unsuitable for practical application in open road testing. Summary of the Invention

[0005] In order to avoid the shortcomings of the above-mentioned existing technologies, the present invention provides a method for predicting the instantaneous energy consumption of fuel cell vehicles based on an integrated model, which realizes the instantaneous energy consumption prediction of fuel cell vehicles on open roads and provides a basis for accurately estimating the driving range of fuel cell vehicles.

[0006] The present invention adopts the following technical solutions to achieve the purpose of the invention:

[0007] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is characterized by comprising the following steps:

[0008] Step 1: Use sensors to obtain collected information, including: collecting road elevation information by a laser rangefinder, collecting location information and altitude information by the vehicle-mounted GPS, collecting the air flow at the fuel cell end by the air flow sensor, and collecting various electrical signals by the vehicle's internal CAN protocol. The electrical signals are: fuel cell end output current I1, fuel cell end output voltage U1, power battery end output current I2, and power battery end output voltage U2; for the location information and altitude information, calculate the speed, acceleration, and slope information based on the location difference and altitude difference between two adjacent time intervals; and calculate the fuel cell end instantaneous power P according to formula (1) FC and the instantaneous power P at the power battery end BAT ;

[0009]

[0010] The collected information, speed, acceleration, slope information and instantaneous power P of the fuel cell end are FC and the instantaneous power P at the power battery end BAT Together as preprocessed data;

[0011] Step 2: Time-aligning the preprocessed data to obtain time-aligned data, and dividing the time-aligned data into an initial feature set and a model label;

[0012] Step 3: Perform feature extraction and data normalization on the initial feature set to obtain model features;

[0013] Step 4: Import the model features and model labels into the original integrated model for network training to obtain an optimized integrated model;

[0014] The original integration model is constructed as follows:

[0015] A formula constraint method is used to integrate prior knowledge into a fully connected neural network to obtain an improved fully connected neural network algorithm that includes prior knowledge; a time series cross-validation method is used to perform hyperparameter tuning on a random forest regression algorithm, a support vector regression algorithm, and the improved fully connected neural network algorithm to obtain a tuned model, and a weighted average method is used to obtain the original integrated model based on the tuned model;

[0016] Step 5: The output results of the optimization integrated model are subjected to a post-processing method of hydrogen-to-electricity conversion to obtain the instantaneous energy consumption values ​​at the fuel cell end and the instantaneous energy consumption values ​​at the power battery end.

[0017] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is also characterized by:

[0018] In step 1, the speed, acceleration and slope information are calculated as follows:

[0019] The distance difference between adjacent track points is calculated based on the position information of adjacent track points at UTC time captured by the vehicle-mounted GPS, the slope information between adjacent track points is calculated based on the distance difference and altitude information, the speed information between adjacent track points is calculated based on the distance difference and time difference between adjacent track points, and the acceleration information between adjacent track points is calculated based on the speed information and time difference between adjacent track points.

[0020] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is also characterized by:

[0021] In step 2, time alignment is performed on the preprocessed data to obtain time-aligned data, and dividing the time-aligned data into an initial feature set and a model label means:

[0022] Time-align data of different sampling frequencies: select the information with the lowest sampling frequency and keep it unchanged, and downsample all other information at the lowest sampling frequency, converting high-frequency information to low-frequency information;

[0023] The same-frequency road elevation information, location information, altitude information, speed, acceleration, slope information and fuel cell air flow in the time-aligned data are used as the initial feature set, and the same-frequency fuel cell instantaneous power P FC and the instantaneous power P at the power battery end BAT as model labels.

[0024] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is also characterized by:

[0025] In step 3, the model features are obtained as follows:

[0026] The Pearson correlation coefficient method represented by formula (2) is used to obtain the characteristic parameters highly correlated with the instantaneous power of the fuel cell vehicle on the initial feature set:

[0027]

[0028] In formula (2):

[0029] Pearson is the Pearson correlation coefficient; high correlation means: Pearson ≥ 0.7;

[0030] COV(X,Y) is the covariance of the instantaneous power and characteristic parameters of the fuel cell vehicle;

[0031] σ X is the standard deviation of the characteristic parameter, σ Yis the standard deviation of the instantaneous power of the fuel cell vehicle;

[0032] The characteristic parameters include: road elevation information, location information, altitude information, speed, acceleration and slope information, and fuel cell end air flow;

[0033] The feature parameters are normalized using formula (3) using min-max, the original data X of the feature parameters are linearly transformed to obtain X′ mapped to between 0 and 1, and the original data X of the feature parameters is replaced by X′ as the model feature;

[0034]

[0035] In formula (3):

[0036] X min is the minimum value in the original data X; X max is the maximum value of the original data X.

[0037] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is also characterized by:

[0038] In step 4, an improved fully connected neural network algorithm including prior knowledge is obtained as follows:

[0039] The formula constraint method is used to integrate the prior knowledge represented by formula (4) into the fully connected neural network:

[0040]

[0041] In formula (4):

[0042] P theory is the theoretical input power of the motor, r is the motor resistance, and K is the armature constant;

[0043] P FC_max 、P BAT_max , m, k, f rl and R are vehicle parameters, which are determined by different vehicle models. FC_max is the maximum power of the fuel cell end of the selected vehicle model, P BAT_max is the maximum power of the power battery of the selected model, m is the curb weight of the selected model, k is the air resistance constant of the selected model, f rl is the rolling resistance constant of the selected model, and R is the tire radius of the selected model;

[0044] v is the vehicle speed, a is the vehicle acceleration, sinθ is the sine of the slope angle;

[0045] The formula constraint method refers to the process of performing weight sharing and limiting synaptic weight selection on the improved fully connected neural network model during the training phase of the improved fully connected neural network model, using the motor theoretical input power limit, the fuel cell instantaneous power limit, and the power battery instantaneous power limit represented by formula (4) as constraints, to constrain the model output results.

[0046] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is also characterized by:

[0047] In step 4, the time series cross validation refers to:

[0048] The model features and model labels are used as model input data. The model input data is divided into 70% of the model input data as the training set and 30% of the model input data as the test set. Time series cross-validation is performed on the test set. Each fold of the data set is divided according to the time sequence. A certain step size is selected to advance according to the time sequence. The verification method is as follows:

[0049] First fold: training[N1], test[N2]

[0050] Second fold: training[N1 N2], test[N3]

[0051] Fold 3: training[N1 N2 N3], test[N4]

[0052] …

[0053] Fold N: training[N1 N2 N3……Nx-1], test[Nx]

[0054] Among them, N1, N2, N3...Nx are the data contained in each step.

[0055] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is also characterized by:

[0056] In step 4, the weighted average method is used to obtain the original integrated model for the tuned model in the following manner:

[0057] The mean absolute error (MAE) is used as the evaluation index for the partition weight, and the evaluation index is calculated according to formula (5):

[0058]

[0059] In formula (5):

[0060] is the true value data, is the predicted value of a single tuning model, and n is the number of data samples;

[0061] The smaller the value of the evaluation index MAE, the higher the prediction accuracy, and the greater the weight is given. The three models are weighted in turn according to the size of the evaluation index MAE;

[0062] The weighted average method represented by formula (6) is used to calculate the prediction value y of the comprehensive tuning model output The comprehensive tuning model includes three models: random forest regression optimization model, support vector regression optimization model and improved fully connected neural network optimization model. The order of the models is represented by the i-th model, then: i=1, 2, 3;

[0063]

[0064] In formula (6):

[0065] ω i is the weight of the i-th model, s i is the prediction result of the i-th model;

[0066] The instantaneous energy consumption prediction method of a fuel cell vehicle based on an integrated model of the present invention is also characterized by:

[0067] In step 5, the output result of the optimization integrated model is subjected to the post-processing method of hydrogen-to-electricity conversion to obtain the instantaneous energy consumption value of the fuel cell end and the instantaneous energy consumption value of the power battery end in the following manner:

[0068] The fuel cell end energy E is calculated according to formula (7): FC , power battery end energy E BAT and total energy E D :

[0069]

[0070] The fuel cell end contribution mileage D is calculated according to formula (8): FC And the power battery contributes to the mileage D BAT ;

[0071]

[0072] In formula (8): FC is the fuel cell output ratio, η BAT is the power battery output ratio, D is the total mileage;

[0073] According to hydrogen consumption The instantaneous energy consumption of the fuel cell is calculated by formula (9): and the instantaneous energy consumption of the power battery C E :

[0074]

[0075] Instantaneous energy consumption of fuel cell The unit is kg / 100km, the instantaneous energy consumption of the power battery is C E The unit is kWh / 100km.

[0076] The present invention solves the problem of difficulty in predicting instantaneous energy consumption at the fuel cell and power battery ends of fuel cell vehicles. Compared with existing technologies, the present invention has the following beneficial effects:

[0077] 1. The present invention adopts a formula constraint method to integrate prior knowledge into a fully connected neural network; constructing an improved fully connected neural network model with prior knowledge helps to improve the efficiency and accuracy of the improved fully connected neural network model.

[0078] 2. The present invention uses a random forest regression optimization model, a support vector regression optimization model, and an improved fully connected neural network optimization model that have undergone hyperparameter tuning to construct an integrated model using a weighted average method to achieve instantaneous power prediction of the fuel cell and power battery ends of fuel cell vehicles with high accuracy.

[0079] 3. The present invention adopts the hydrogen-electricity conversion method to convert the instantaneous power value predicted by the integrated model into the instantaneous energy consumption value of the fuel cell end and the power battery end, thereby realizing the prediction of the instantaneous energy consumption of fuel cell vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 This is a flow chart of the instantaneous energy consumption prediction method for fuel cell vehicles based on the integrated model of the present invention;

[0081] Figure 2 This is a schematic diagram of the fully connected neural network model in the method of the present invention;

[0082] Figure 3 Schematic diagram of a fully connected neural network containing prior knowledge in the method of the present invention;

[0083] Figure 4 This is the principle diagram of the original integrated model in the method of the present invention;

[0084] Figure 5 This is a diagram showing the instantaneous power prediction result of the power battery terminal in the method of the present invention;

[0085] Figure 6 This is a diagram showing the instantaneous power prediction result of the fuel cell end in the method of the present invention;

[0086] Figure 7 This is a diagram showing the instantaneous energy consumption prediction result of the power battery terminal in the method of the present invention;

[0087] Figure 8 This is a diagram showing the prediction results of the instantaneous energy consumption at the fuel cell end in the method of the present invention. DETAILED DESCRIPTION

[0088] Figure 1 The flowchart of the instantaneous energy consumption prediction method in this embodiment is shown, which includes the following steps:

[0089] Step 1: Use sensors to obtain collected information, including: road elevation information collected by the laser rangefinder, location information and altitude information collected by the vehicle GPS, air flow sensor to collect the air flow at the fuel cell end, and various electrical signals collected by the vehicle internal CAN protocol. The electrical signals are: fuel cell end output current I1, fuel cell end output voltage U1, power battery end output current I2, and power battery end output voltage U2; for the location information and altitude information, calculate the speed, acceleration, and slope information based on the location difference and altitude difference between two adjacent time intervals; and calculate the instantaneous power P of the fuel cell end according to formula (1) FC and the instantaneous power P at the power battery end BAT ;

[0090]

[0091] The collected information, speed, acceleration, slope information and instantaneous power P of the fuel cell end FC and the instantaneous power P at the power battery end BAT Together as preprocessed data;

[0092] The distance difference between adjacent track points is calculated based on the position information of adjacent track points in UTC time captured by the vehicle-mounted GPS. The slope information between adjacent track points is calculated based on the distance difference and altitude information. The speed information between adjacent track points is calculated based on the distance difference and time difference between adjacent track points. The acceleration information between adjacent track points is calculated based on the speed information and time difference between adjacent track points.

[0093] Step 2: Time-align the preprocessed data to obtain time-aligned data, and divide the time-aligned data into the initial feature set and model labels:

[0094] Time-align data of different sampling frequencies: select the information with the lowest sampling frequency and keep it unchanged, and downsample all other information at the lowest sampling frequency, converting high-frequency information to low-frequency information;

[0095] The same-frequency road elevation information, location information, altitude information, speed, acceleration, slope information and fuel cell air flow in the time-aligned data are used as the initial feature set, and the same-frequency fuel cell instantaneous power PFC and the instantaneous power P at the power battery end BAT as model labels.

[0096] Step 3: Perform feature extraction and data normalization on the initial feature set to obtain model features:

[0097] The Pearson correlation coefficient method represented by formula (2) is used to obtain the characteristic parameters that are highly correlated with the instantaneous power of fuel cell vehicles:

[0098]

[0099] In formula (2):

[0100] Pearson is the Pearson correlation coefficient; high correlation means: Pearson ≥ 0.7;

[0101] COV(X,Y) is the covariance of the instantaneous power and characteristic parameters of the fuel cell vehicle;

[0102] σ X is the standard deviation of the characteristic parameter, σ Y is the standard deviation of the instantaneous power of the fuel cell vehicle;

[0103] The characteristic parameters include: road elevation information, location information, altitude information, speed, acceleration and slope information, as well as the air flow at the fuel cell end;

[0104] The feature parameters are normalized using formula (3) using min-max, and the original data X of the feature parameters are linearly transformed to obtain X′ mapped to between 0 and 1. The original data X of the feature parameters is replaced by X′ as the model feature;

[0105]

[0106] In formula (3):

[0107] X min is the minimum value in the original data X; X max is the maximum value of the original data X.

[0108] Step 4: Import the model features and model labels into the original ensemble model for network training to obtain the optimized ensemble model;

[0109] The original integration model is built as follows. The principle of the original integration model is as follows Figure 4 As shown:

[0110] The formula constraint method is used to integrate prior knowledge into the fully connected neural network, such as Figure 3As shown, an improved fully connected neural network algorithm containing prior knowledge is obtained; the random forest regression algorithm, support vector regression algorithm and the improved fully connected neural network algorithm are hyperparameter tuned using the time series cross-validation method to obtain a tuned model, and the weighted average method is used to obtain the original integrated model for the tuned model;

[0111] A random forest regression algorithm was constructed, which contained 20 decision trees. The maximum depth of each tree was set to 50, and the subset feature data was set to 5. A support vector regression algorithm was constructed, which used a Gaussian kernel function with a kernel function coefficient of 0.2 and a penalty coefficient of 100.

[0112] A fully connected neural network is constructed, which contains 3 hidden layers with 32 nodes each. The road elevation information, speed, acceleration, slope information and fuel cell air flow in the model features are used as input, and the output is the instantaneous power of the fuel cell and power battery, such as Figure 2 As shown;

[0113] In step 4, the improved fully connected neural network algorithm including prior knowledge is obtained as follows:

[0114] The formula constraint method is used to integrate the prior knowledge represented by formula (4) into the fully connected neural network:

[0115]

[0116] In formula (4):

[0117] P theory is the theoretical input power of the motor, r is the motor resistance, and K is the armature constant;

[0118] P FC_max 、P BAT_max , m, k, f rl and R are vehicle parameters, which are determined by different vehicle models. FC_max is the maximum power of the fuel cell end of the selected vehicle model, P BAT_max is the maximum power of the power battery of the selected model, m is the curb weight of the selected model, k is the air resistance constant of the selected model, f rl is the rolling resistance constant of the selected model, and R is the tire radius of the selected model;

[0119] v is the vehicle speed, a is the vehicle acceleration, sinθ is the sine of the slope angle;

[0120] The theoretical input power of the motor is calculated based on the basic theory of vehicle dynamics, ignoring the mechanical efficiency of the transmission system. It is used to constrain the lower limit of the sum of the instantaneous power at the fuel cell end and the instantaneous power at the power battery end.

[0121] The formula constraint method refers to the process of using the theoretical input power limit of the motor, the instantaneous power limit of the fuel cell end, and the instantaneous power limit of the power battery end represented by formula (4) as constraints during the training phase of the improved fully connected neural network model to perform weight sharing and restrict the selection of synaptic weights on the improved fully connected neural network model to constrain the model output results.

[0122] Time series cross-validation means: using model features and model labels as model input data, dividing the model input data, using 70% of the model input data as the training set and 30% of the model input data as the test set, and performing time series cross-validation on the test set. Each fold of the data set is divided in chronological order, and every 5 data points are selected as a step size. The validation is carried out in a time series manner. The following is the method:

[0123] First fold: training[N1], test[N2]

[0124] Second fold: training[N1 N2], test[N3]

[0125] Fold 3: training[N1 N2 N3], test[N4]

[0126] …

[0127] Fold N: training[N1 N2 N3……Nx-1], test[Nx]

[0128] Among them, N1, N2, N3...Nx are the data contained in each step.

[0129] The mean absolute error (MAE) is used as the evaluation index for the partition weight, and the evaluation index is calculated according to formula (5):

[0130]

[0131] In formula (5):

[0132] is the true value data, is the predicted value of a single tuning model, and n is the number of data samples;

[0133] The smaller the value of the evaluation index MAE, the higher the prediction accuracy, and the larger the weight is given. The three models are weighted in turn according to the size of the evaluation index MAE.

[0134] The weighted average method represented by formula (6) is used to calculate the prediction value y of the comprehensive tuning model outputThe comprehensive tuning model includes three models: random forest regression optimization model, support vector regression optimization model and improved fully connected neural network optimization model. The order of the models is represented by the i-th model, i=1,2,3.

[0135]

[0136] In formula (6):

[0137] ω i is the weight of the i-th model, s i is the prediction result of the i-th model;

[0138] The improved fully connected neural network optimization model has the highest accuracy and is assigned a weight of 50%. The random forest regression optimization model and the support vector regression optimization model are assigned weights of 30% and 20%, respectively. The final model output values ​​are calculated: the instantaneous power at the fuel cell end and the instantaneous power at the power battery end.

[0139] The ensemble model is trained with an upper limit of 2000 training times and a batch size of 1024 (since it is a pure data model without images, the batch size can be appropriately increased to improve the training speed). Pytorch is used as the deep learning framework.

[0140] Record the loss value for each training iteration. If the loss value is less than 0.001, terminate the training and save the model. Otherwise, continue training until 2000 iterations and save the model.

[0141] After loading the model, due to the different prediction speeds between different sub-models, the sub-models in the integrated model are processed in parallel to improve the real-time performance of the model calculation. The prediction results of the sub-models from the same time point are saved to the same location, then weighted and output as the instantaneous power prediction result.

[0142] In this embodiment, the instantaneous power values ​​of the power battery and fuel cell predicted by the integrated model are as follows: Figure 5 and Figure 6 As shown, Figure 5 The middle curve a1 is the true value of the instantaneous power at the power battery end, and the curve a2 is the predicted value of the instantaneous power at the power battery end; Figure 6 The middle curve b1 is the true value of the instantaneous power at the fuel cell end, and the curve b2 is the predicted value of the instantaneous power at the fuel cell end.

[0143] Curve a1 and curve a2 basically overlap, and curve b1 and curve b2 basically overlap, indicating that the integrated model can accurately predict the instantaneous power of fuel cell vehicles, and the evaluation index MAE reaches 0.08, and the percentage error reaches 0.53%.

[0144] Step 5: Based on the output results of the optimized integrated model, the instantaneous energy consumption values ​​of the fuel cell side and the power battery side are obtained using the hydrogen-to-electricity conversion post-processing method as shown below:

[0145] The fuel cell end energy E is calculated according to formula (7): FC , power battery end energy E BAT and total energy E D :

[0146]

[0147] The fuel cell end contribution mileage D is calculated according to formula (8): FC And the power battery contributes to the mileage D BAT ;

[0148]

[0149] In formula (8): FC is the fuel cell output ratio, η BAT is the power battery output ratio, D is the total mileage;

[0150] According to hydrogen consumption The instantaneous energy consumption of the fuel cell is calculated by formula (9): and the instantaneous energy consumption of the power battery C E :

[0151]

[0152] Instantaneous energy consumption of fuel cell The unit is kg / 100km, the instantaneous energy consumption of the power battery is C E The unit is kWh / 100km.

[0153] The instantaneous energy consumption values ​​predicted by the hydrogen-to-electricity conversion method in this embodiment are as follows: Figure 7 and Figure 8 As shown, Figure 8 The middle curve c1 is the true value of the instantaneous energy consumption at the fuel cell end, and the curve c2 is the predicted value of the instantaneous energy consumption at the fuel cell end.

[0154] Figure 7 The real value and predicted value of instantaneous energy consumption of the power battery are basically overlapped. Figure 8 The middle curve c1 and curve c2 basically overlap, indicating that the hydrogen-to-electricity conversion method can accurately predict the instantaneous energy consumption of fuel cell vehicles, and the evaluation index MAE reaches 0.12 and the percentage error reaches 0.8%.

[0155] By comprehensively judging the prediction results of the integrated model of this embodiment and the prediction results of the hydrogen-to-electricity conversion method, the present invention can accurately predict the instantaneous energy consumption of fuel cell vehicles.

Claims

1. A method for predicting instantaneous energy consumption of fuel cell vehicles based on an integrated model, characterized by The following steps are included: Step 1: Use sensors to obtain collected information, including: collecting road elevation information by a laser rangefinder, collecting location information and altitude information by the vehicle-mounted GPS, collecting the air flow at the fuel cell end by the air flow sensor, and collecting various electrical signals by the vehicle's internal CAN protocol. The electrical signals are: fuel cell end output current I1, fuel cell end output voltage U1, power battery end output current I2, and power battery end output voltage U2; for the location information and altitude information, calculate the speed, acceleration, and slope information based on the location difference and altitude difference between two adjacent time intervals; and calculate the fuel cell end instantaneous power P according to formula (1) FC and the instantaneous power P at the power battery end BAT ; The collected information, speed, acceleration, slope information and instantaneous power P of the fuel cell end are FC and the instantaneous power P at the power battery end BAT Together as preprocessed data; Step 2: Time-aligning the preprocessed data to obtain time-aligned data, and dividing the time-aligned data into an initial feature set and a model label; Step 3: Perform feature extraction and data normalization on the initial feature set to obtain model features; Step 4: Import the model features and model labels into the original integrated model for network training to obtain an optimized integrated model; The original integration model is constructed as follows: A formula constraint method is used to integrate prior knowledge into a fully connected neural network to obtain an improved fully connected neural network algorithm that includes prior knowledge; a time series cross-validation method is used to perform hyperparameter tuning on a random forest regression algorithm, a support vector regression algorithm, and the improved fully connected neural network algorithm to obtain a tuned model, and a weighted average method is used to obtain the original integrated model based on the tuned model; The improved fully connected neural network algorithm containing prior knowledge is obtained as follows: The formula constraint method is used to integrate the prior knowledge represented by formula (4) into the fully connected neural network: In formula (4): P theory is the theoretical input power of the motor, r is the motor resistance, and K is the armature constant; P FC_max 、P BAT_max , m, k, f rl and R are vehicle parameters, which are determined by different vehicle models. FC_max is the maximum power of the fuel cell end of the selected vehicle model, P BAT_max is the maximum power of the power battery of the selected model, m is the curb weight of the selected model, k is the air resistance constant of the selected model, f rl is the rolling resistance constant of the selected model, and R is the tire radius of the selected model; v is the vehicle speed, a is the vehicle acceleration, sinθ is the sine of the slope angle; The formula constraint method refers to the process of performing weight sharing and limiting synaptic weight selection on the improved fully connected neural network model during the training phase of the improved fully connected neural network model, using the motor theoretical input power limit, the fuel cell terminal instantaneous power limit, and the power battery terminal instantaneous power limit represented by formula (4) as constraints, to constrain the model output results. Step 5: The output results of the optimization integrated model are subjected to a post-processing method of hydrogen-to-electricity conversion to obtain the instantaneous energy consumption values ​​at the fuel cell end and the instantaneous energy consumption values ​​at the power battery end.

2. The method for predicting instantaneous energy consumption of a fuel cell vehicle based on an integrated model according to claim 1 is characterized by: In step 1, the speed, acceleration and slope information are calculated as follows: The distance difference between adjacent track points is calculated based on the position information of adjacent track points at UTC time captured by the vehicle-mounted GPS, the slope information between adjacent track points is calculated based on the distance difference and altitude information, the speed information between adjacent track points is calculated based on the distance difference and time difference between adjacent track points, and the acceleration information between adjacent track points is calculated based on the speed information and time difference between adjacent track points.

3. The method for predicting instantaneous energy consumption of a fuel cell vehicle based on an integrated model according to claim 1 is characterized by: In step 2, time alignment is performed on the preprocessed data to obtain time-aligned data, and dividing the time-aligned data into an initial feature set and a model label means: Time-align data of different sampling frequencies: select the information with the lowest sampling frequency and keep it unchanged, and downsample all other information at the lowest sampling frequency, converting high-frequency information to low-frequency information; The same-frequency road elevation information, location information, altitude information, speed, acceleration, slope information and fuel cell air flow in the time-aligned data are used as the initial feature set, and the same-frequency fuel cell instantaneous power P FC and the instantaneous power P at the power battery end BAT as model labels.

4. The method for predicting instantaneous energy consumption of a fuel cell vehicle based on an integrated model according to claim 1, characterized in that: In step 3, the model features are obtained as follows: The Pearson correlation coefficient method represented by formula (2) is used to obtain the characteristic parameters highly correlated with the instantaneous power of the fuel cell vehicle on the initial feature set: In formula (2): Pearson is the Pearson correlation coefficient; high correlation means: Pearson ≥ 0.7; COV(X,Y) is the covariance of the instantaneous power and characteristic parameters of the fuel cell vehicle; σ X is the standard deviation of characteristic parameters, σ Y is the standard deviation of the instantaneous power of the fuel cell vehicle; The characteristic parameters include: road elevation information, location information, altitude information, speed, acceleration and slope information, and fuel cell end air flow; The feature parameters are normalized using formula (3) using min-max, the original data X of the feature parameters are linearly transformed to obtain X′ mapped to between 0 and 1, and the original data X of the feature parameters is replaced by X′ as the model feature; In formula (3): X min is the minimum value in the original data X; X max is the maximum value of the original data X.

5. The method for predicting instantaneous energy consumption of a fuel cell vehicle based on an integrated model according to claim 1, characterized in that: In step 4, the time series cross validation refers to: The model features and model labels are used as model input data. The model input data is divided into 70% of the model input data as the training set and 30% of the model input data as the test set. Time series cross-validation is performed on the test set. Each fold of the data set is divided according to the time sequence. A certain step size is selected to advance according to the time sequence. The verification method is as follows: First fold: training[N1], test[N2] Second fold: training[N1 N2], test[N3] Fold 3: training[N1 N2 N3], test[N4] …… Fold N: training[N1 N2 N3……Nx-1], test[Nx] Among them, N1, N2, N3...Nx are the data contained in each step.

6. The method for predicting instantaneous energy consumption of a fuel cell vehicle based on an integrated model according to claim 1, characterized in that: In step 4, the weighted average method is used to obtain the original integrated model for the tuned model in the following manner: The mean absolute error (MAE) is used as the evaluation index for the partition weight, and the evaluation index is calculated according to formula (5): In formula (5): is the true value data, is the predicted value of a single tuning model, and n is the number of data samples; The smaller the value of the evaluation index MAE, the higher the prediction accuracy, and the greater the weight is given. The three models are weighted in turn according to the size of the evaluation index MAE; The weighted average method represented by formula (6) is used to calculate the prediction value y of the comprehensive tuning model output The comprehensive tuning model includes three models: random forest regression optimization model, support vector regression optimization model and improved fully connected neural network optimization model. The order of the models is represented by the i-th model, then: i=1, 2, 3; In formula (6): ω i is the weight of the i-th model, s i is the prediction result of the i-th model.

7. The method for predicting instantaneous energy consumption of a fuel cell vehicle based on an integrated model according to claim 1, characterized in that: In step 5, the output result of the optimization integrated model is subjected to the post-processing method of hydrogen-to-electricity conversion to obtain the instantaneous energy consumption value of the fuel cell end and the instantaneous energy consumption value of the power battery end in the following manner: The fuel cell end energy E is calculated according to formula (7): FC , power battery end energy E BAT and total energy E D : The fuel cell end contribution mileage D is calculated according to formula (8): FC And the power battery contributes to the mileage D BAT ; In formula (8): FC is the fuel cell output ratio, η BAT is the power battery output ratio, D is the total mileage; According to hydrogen consumption The instantaneous energy consumption of the fuel cell is calculated by formula (9): and the instantaneous energy consumption of the power battery C E : Instantaneous energy consumption of fuel cell The unit is kg / 100km, the instantaneous energy consumption of the power battery is C E The unit is kWh / 100km.

Images