Method for predicting remaining fuel quantity in vehicle fuel tank based on data-driven and non-linear model

Abstract

The present invention discloses a method for predicting the remaining fuel quantity in an automobile fuel tank based on data-driven and non-linear models. First, using the physical model of the fuel tank and the position of the oil level in the fuel tank at different remaining fuel quantities, the volume of the remaining space in the fuel tank at different oil level positions is calculated, and the corresponding remaining fuel quantity is determined based on the total volume of the fuel tank, so as to determine a number of measurement points. Using these measurement points, a non-linear model between the remaining fuel quantity in the fuel tank when the vehicle is stationary and the oil level position is constructed; then, the vehicle-mounted data set related to fuel consumption is input into a BP neural network for training to obtain a data-driven model for predicting the real-time fuel consumption of the vehicle; when the above non-linear model and data-driven model act jointly, the remaining fuel quantity in the fuel tank can be predicted in real time. The present invention can accurately predict the remaining fuel quantity in the fuel tank in real time through the non-linear model and the data-driven model, avoiding the problem of inaccurate prediction of the oil level information when the vehicle is driving on a non-stationary and non-horizontal ground, and improving the oil level prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the field of vehicle engineering, and particularly to a precise prediction method for the remaining fuel quantity in a fuel tank under the combined action of data driving and non - linear interpolation. Background Art

[0002] Due to the influence of the driving range on electric vehicles, fuel - powered vehicles are still relatively common means of transportation at present. Fuel consumption is a key indicator for evaluating fuel economy. Accurately estimating the fuel consumption of a vehicle and obtaining the remaining fuel quantity in the vehicle's fuel tank play an important role in improving the fuel economy of the vehicle. In addition, drivers can predict the remaining driving range of the vehicle based on the remaining fuel quantity and fuel consumption, so as to adjust their driving strategies.

[0003] In the urban traffic environment, there are many factors affecting fuel consumption prediction, such as the driving speed, acceleration, traffic congestion, and vehicle performance of the vehicle. Affected by various factors, the fuel consumption of the vehicle actually has no regular pattern, and it is very difficult to directly calculate the fuel consumption with a mathematical formula, making it difficult to obtain the real - time remaining fuel quantity of the vehicle, and preventing the driver from predicting the remaining driving range of the vehicle. At present, the prediction of the remaining fuel quantity in the vehicle's fuel tank only shows the current remaining grid number through the fuel gauge pointer, and the data used in this technology is the oil level position measured by the oil level sensor. The prediction accuracy is acceptable when the vehicle is stationary, but when the vehicle is driving, accelerating, decelerating, or driving on slopes and bumpy roads, the oil level position will fluctuate greatly, making it difficult to accurately display the current remaining fuel quantity of the vehicle. Coupled with the irregular shape of the current vehicle fuel tank, it makes the prediction of the remaining fuel quantity more difficult, and it is very difficult to estimate the remaining fuel quantity based on the oil level position when the vehicle is driving. In addition, for the real - time prediction of fuel consumption, existing methods such as linear regression analysis and support vector machines predict fuel consumption by establishing a linear relationship between input and output, which are usually suitable for fuel consumption data sets with small scale and clear linear relationships. For large - scale data of vehicle operation, due to the highly non - linear relationship between operation data and fuel consumption, it is very difficult for these methods to accurately predict the fuel consumption of the vehicle. The BP neural network has strong adaptability, strong fault tolerance and self - learning ability, can arbitrarily adapt to complex non - linear models, and can represent the accuracy of multi - dimensional data to any desired data, making it very suitable for describing the complex non - linear process of vehicle fuel consumption, facilitating the acquisition of real - time fuel consumption under the vehicle driving state, and then predicting the remaining fuel quantity in the vehicle's fuel tank.

[0004] The present invention uses a method combining data-driven and non-linear models to predict the remaining fuel quantity in a vehicle fuel tank in real time. This method first constructs a non-linear model for predicting the remaining fuel quantity of the vehicle in a stationary state based on the physical model of the fuel tank, then constructs a corresponding fuel consumption data set based on in-vehicle sensor data, inputs the data set into a BP neural network model for training to obtain the real-time fuel consumption of the vehicle during driving, and then uses the remaining fuel quantity in the stationary state and the real-time fuel consumption data to predict the remaining fuel quantity in the fuel tank at the current moment. The operation is simple, the prediction accuracy is high, and it has high reliability. Summary of the Invention

[0005] Aiming at the problem of inaccurate prediction of the remaining fuel quantity in the fuel tank due to the irregularity of the vehicle fuel tank and different vehicle motion states, the present invention proposes a real-time accurate prediction method for the remaining fuel quantity in the fuel tank under the combined action of data-driven and non-linear models. First, based on the physical model of the fuel tank, a non-linear model for predicting the remaining fuel quantity in the fuel tank in a stationary state is established. Then, based on vehicle operation data and the relationship between vehicle driving state and fuel consumption, a data-driven model for predicting fuel consumption in the vehicle driving state is established using a BP neural network. Finally, by combining the non-linear model and the data-driven model, the remaining fuel quantity in the vehicle fuel tank is predicted in real time.

[0006] A method for predicting the remaining fuel quantity in a vehicle fuel tank based on data-driven and non-linear models includes the following steps:

[0007] S1 By measuring the dimensional parameters of the physical model of the vehicle fuel tank, including the height h, total length l, width d, thickness t, etc. of the fuel tank, accurately establish a 3D CAD model of the fuel tank using 3D modeling software, providing a CAD model for the establishment of a non-linear model for predicting the remaining fuel quantity in the fuel tank in a stationary state of the vehicle;

[0008] S2 Construct a non-linear model for predicting the remaining fuel quantity in the fuel tank in a stationary state;

[0009] S2.1 Obtain multiple measurement points for calculating the remaining fuel quantity and establish a horizontal plane

[0010] Detect the situations when the oil level is at 1 / 8h, 1 / 4h, 3 / 8h, 1 / 2h, 5 / 8h, 3 / 4h, 7 / 8h respectively through an oil level sensor, obtain 7 measurement points for calculating the remaining fuel quantity, and then establish a horizontal plane at the positions of each measurement point. Based on the shape and structural parameters of the fuel tank below the horizontal plane, obtain the space occupied by the part of the fuel tank at the positions of the 7 measurement points and below.

[0011] S2.2 Calculate the remaining fuel quantity in the fuel tank at the measurement points

[0012] According to the horizontal plane, tank structure and size parameters established at the measurement point, the CAD model of the tank is imported into the finite element analysis software to calculate the volume of the space below the liquid level position and the horizontal plane at different measurement points. The volume of the liquid level position and the space below the oil tank at different measuring points is calculated by grid division and numerical integration technology, so as to obtain the remaining oil volume V1, V2, V3, V4, V5, V6, V7 corresponding to each liquid level measuring point;

[0013] S2.3 Establish a nonlinear model of the remaining fuel in the tank when the car is stationary

[0014] Using the above 7 measurement points and their corresponding remaining oil volume, a nonlinear model of the remaining oil volume in the fuel tank of the vehicle at rest is established. The establishment process of the nonlinear model is as follows: First, using the known liquid level height at the measurement point and the corresponding remaining oil volume, an interval nonlinear model is established through nonlinear interpolation to obtain a nonlinear model between the liquid level height and the remaining oil volume, which is expressed as follows:

[0015]

[0016] Wherein, f(V) is the remaining oil volume to be predicted, hx is the height of the oil level to be predicted, and hi (i=1-7) is the height of the liquid level of the 7 measurement points known in S2.2;

[0017] S3 establishes a prediction model for the remaining fuel in the fuel tank when the car is in motion

[0018] S3.1 Construct a data-driven model for automobile fuel consumption, that is, use BP neural network to establish the mapping relationship between vehicle driving status data and fuel consumption

[0019] S3.1.1 Determine the input layer data

[0020] The input layer contains 3 neurons. The input layer inputs the speed, acceleration and engine torque data related to the vehicle status and fuel consumption, and the output layer is the fuel consumption result. The speed, acceleration and engine speed data come from the vehicle's on-board sensor data. The collected data is preprocessed. First, the amplitude filtering algorithm is used to remove abnormal values in the speed, acceleration and engine speed data. After normalization, the samples are summarized and unified into the input layer. The normalization method uses the following formula

[0021]

[0022] Among them, TV and CV are the sample data before and after normalization, respectively, and MinV and MaxV are the minimum and maximum data values.

[0023] S3.1.2 Determine the hidden layer data

[0024] The number of hidden layers is selected according to the following formula;

[0025]

[0026] where h m is the number of hidden layers, n i is the number of input layers, l o is the number of output layers, and α is a constant that can take values from 1 to 10.

[0027] The expression for calculating the output of the hidden layer is as follows

[0028]

[0029] where hy i is the output result of the hidden layer, f is the activation function, w ij is the weight function, θ j represents the threshold function, and x i is the data of the input layer.

[0030] S3.1.3 Compare the result of the output layer with the expected value and calculate the error. Determine whether the feedback termination condition is satisfied based on the error. If not, further correct the weights and thresholds between the layers

[0031] where the expected value is the original fuel consumption data of the vehicle or the fuel consumption data measured by a fuel consumption meter. The error of the output layer is calculated through the following expression

[0032]

[0033] where Ev k is the expected output result, Op k is the result of the output layer, and w jk and θ k are the weight and threshold functions between the hidden layer and the output layer.

[0034] When the error does not meet the set value, the weights and thresholds between the layers need to be corrected. The correction values are calculated by taking the derivatives of the weights and thresholds respectively using the error expression. The correction expressions are as shown in Eqs. (6) and (7)

[0035]

[0036]

[0037] where n represents the number of all output nodes;

[0038] S3.2 Further correct the data-driven model

[0039] After the vehicle enters the stationary state from the driving state, the non-linear model for predicting the remaining fuel volume in the stationary state is used to estimate the remaining fuel volume after the vehicle has traveled a certain distance again, calculate the fuel consumption value of the vehicle during this driving mileage, and use this fuel consumption value to correct the data-driven model.

[0040] S3.3 Construct a real-time and accurate prediction model for the remaining fuel volume

[0041] After the data-driven model for fuel consumption prediction has been corrected multiple times, the real-time fuel consumption prediction model of the vehicle can be obtained, so as to achieve the purpose of real-time and accurate prediction of the vehicle's real-time fuel consumption. The remaining fuel volume of the vehicle in the stationary state can be calculated using the non-linear model in S2. The remaining fuel volume of the vehicle during driving is calculated by subtracting the real-time fuel consumption of the vehicle from the remaining fuel volume in the stationary state, and then the remaining fuel volume in the fuel tank of the vehicle in the driving state can be predicted.

[0042] Advantages of the present invention:

[0043] 1. Using the mean value of the oil level heights measured by the oil level sensor multiple times and the structural parameters of the fuel tank to establish a physical model, and then using the non-linear relationship between the physical model and the remaining volume of the fuel tank to establish a non-linear model to predict the remaining fuel volume in the fuel tank and obtain the remaining fuel volume in the fuel tank of the vehicle in the stationary state. The present invention constructs a non-linear model for predicting the remaining fuel volume starting from the fuel tank CAD model, and the prediction accuracy is higher;

[0044] 2. Based on the automotive CAN bus data, a BP neural network is established with the speed, acceleration, and engine speed data sets as inputs, and then a data-driven model for fuel consumption prediction in the vehicle driving state is obtained after being corrected by the non-linear model, further improving the prediction accuracy. Then, by updating the data sample set, the fuel consumption in the vehicle driving state is predicted in real time, and the remaining fuel volume in the fuel tank of the vehicle in the driving state is obtained. The network structure adopted conforms very well to the non-linear characteristics of fuel consumption. By selecting the appropriate number of neurons, the efficiency and fitting accuracy of the model can be improved, and then the accuracy of fuel consumption and remaining fuel volume prediction in the fuel tank can be improved. Description of the drawings

[0045] Figure 1 It is a flow chart for constructing a non-linear model for predicting the remaining fuel volume;

[0046] Figure 2 It is a flow chart of the BP neural network structure;

[0047] Figure 3 It is a general flow chart for jointly predicting the remaining fuel volume by the data-driven and non-linear models;

[0048] Figure 4 It is a front view of the fuel tank;

[0049] Figure 5It is a side view of the fuel tank. Specific implementation mode

[0050] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0051] A method for predicting the remaining fuel quantity of an automobile fuel tank based on data driving and a non-linear model, as Figures 1-5 shown, includes the following steps:

[0052] S1 By measuring the dimensional parameters of the physical model of the automobile fuel tank, including the height h, total length l, width d, thickness t, etc. of the fuel tank, accurately establish a three-dimensional CAD model of the fuel tank using three-dimensional modeling software, and provide a CAD model for the establishment of a non-linear model for predicting the remaining fuel quantity in the fuel tank when the vehicle is stationary. Among them, the CAD model of the fuel tank is as Figures 4-5 shown;

[0053] S2 Construct a non-linear model for predicting the remaining fuel quantity in the fuel tank when it is stationary;

[0054] S2.1 Obtain multiple measurement points for calculating the remaining fuel quantity and establish a horizontal plane

[0055] Detect the situations when the oil level is at 1 / 8h, 1 / 4h, 3 / 8h, 1 / 2h, 5 / 8h, 3 / 4h, and 7 / 8h respectively through an oil level sensor, obtain 7 measurement points for calculating the remaining fuel quantity, and then use the positions of each measurement point to establish a horizontal plane. According to the shape and structural parameters of the fuel tank below the horizontal plane, obtain the space occupied by the part of the fuel tank at the positions of the 7 measurement points and below;

[0056] S2.2 Calculate the remaining fuel quantity in the fuel tank at the measurement points

[0057] According to the horizontal plane established based on the positions of the measurement points, the fuel tank structure and dimensional parameters, import the CAD model of the fuel tank into finite element analysis software, calculate the volumes of the different measurement point liquid level positions and the space below their horizontal planes, and use the volume calculation formula Calculate the volumes of the spaces of the fuel tank parts at different measurement point liquid level positions and below through mesh generation and numerical integration techniques, so as to obtain the remaining fuel quantities V1, V2, V3, V4, V5, V6, and V7 corresponding to each liquid level measurement point position;

[0058] S2.3 Establish a non-linear model for the remaining fuel quantity in the fuel tank when the vehicle is stationary

[0059] Using the above 7 measurement points and their corresponding remaining fuel amounts, a non-linear model of the remaining fuel amount in the fuel tank in the stationary state of the vehicle is established. Through this model, the remaining fuel amount in the fuel tank at any liquid level position can be predicted. The establishment process of the non-linear model is as follows: First, using the known liquid level heights of the measurement points and the corresponding remaining fuel amounts, an interval non-linear model is established through non-linear interpolation to obtain a non-linear model between the liquid level height and the remaining fuel amount. The expression is

[0060]

[0061] where f(V) is the remaining fuel amount to be predicted, hx is the height of the fuel level position to be predicted, and hi (i = 1 to 7) are the liquid level heights of the 7 known measurement points in S2.2. Compared with the traditional linear regression model, this non-linear model can well reflect the non-linearity of vehicle fuel consumption and can more accurately predict the remaining fuel amount in the fuel tank in the stationary state. In addition, by collecting the liquid level height values of the fuel level sensor multiple times and substituting their average values into the non-linear model to predict the remaining fuel amount in the fuel tank, the accuracy of fuel level position detection and remaining fuel amount prediction can be improved, so as to predict the remaining fuel amount in the fuel tank at any liquid level height.

[0062] S3 Establish a prediction model for the remaining fuel amount in the fuel tank in the driving state of the vehicle

[0063] When the vehicle is in the driving state, the liquid level in the fuel tank will change with factors such as vehicle speed, acceleration, and road environment. The liquid level height measured by the fuel level sensor is no longer reliable. If the liquid level height is directly applied to the remaining fuel amount prediction method in the stationary state, the accuracy of the predicted result will be greatly reduced. To overcome this problem, the present invention starts from the aspect of fuel consumption. First, based on a data-driven method, by predicting the fuel consumption of the vehicle in the driving state, the non-linear model in the stationary state and the vehicle fuel consumption model are used to predict the remaining fuel amount in real time. Finally, the non-linear model in the stationary state is used to correct and verify the prediction result. The specific technical solution is as follows:

[0064] S3.1 Construct a data-driven model for vehicle fuel consumption, that is: use a BP neural network to establish a mapping relationship between vehicle driving state data and fuel consumption

[0065] The BP neural network can describe the non-linear relationship between input and output and is suitable for predicting vehicle fuel consumption. The present invention uses a BP neural network to predict vehicle fuel consumption. This network includes an input layer, a hidden layer, and an output layer. First, it is necessary to clarify the data of the input layer, hidden layer, and output.

[0066] S3.1.1 Determine the input layer data

[0067] The input layer contains 3 neurons. Since the fuel consumption of the vehicle can be reflected from its operating state data, the input layer inputs the speed, acceleration, and engine torque data related to the vehicle state and fuel consumption, and the output layer is the fuel consumption result. Among them, the data of speed, acceleration, and engine speed are from the on-vehicle sensor data of the car. In addition, in order to improve the training efficiency of the neural network, accelerate the convergence of network training, and preprocess the collected data, first, the amplitude filtering algorithm is used to remove the outliers in the speed, acceleration, and engine speed data. After normalization processing, they are summarized and unified into samples and placed in the input layer. The normalization method uses the following formula

[0068]

[0069] where TV and CV are the sample data after and before normalization respectively, and MinV and MaxV are the minimum and maximum data values.

[0070] S3.1.2 Determine the hidden layer data

[0071] The sample data set is passed through the hidden layer in the neural network structure to calculate the output result. The output result is compared with the expected result to determine whether the convergence condition is met. If not, the weights and thresholds between layers are corrected through backpropagation. After multiple cycles, an output value that meets the convergence condition is obtained. In the present invention, by setting the training accuracy, the error of the model is continuously reduced through feedback loops, and the accuracy of the fuel consumption prediction model is gradually improved. In this process, the selection of the number of hidden layers is very important, which can determine the stability and efficiency of the network architecture. The number of hidden layers is selected according to the following formula. After verification, it is found that a three-layer non-linear network can approximate any function more accurately, can better approximate the target function, and can accurately identify the fuel consumption state of the car.

[0072]

[0073] where h m is the number of hidden layers, n i is the number of input layers, l o is the number of output layers, and α is a constant that can take values from 1 to 10.

[0074] The expression for calculating the hidden layer output is as follows

[0075]

[0076] where hy i is the hidden layer output result, f is the activation function, w ij is the weight function, θ j represents the threshold function, and x i is the input layer data.

[0077] S3.1.3 Compare the results of the output layer with the expected values and calculate the error. Determine whether the feedback termination condition is met based on the error. If not, further correct the weights and thresholds between the layers.

[0078] The expected value is the original fuel consumption data of the vehicle or the fuel consumption data measured by a fuel consumption meter. The error of the output layer is calculated through the following expression

[0079]

[0080] where Ev k is the expected output result, Op k is the result of the output layer, w jk and θ k are the weight and threshold functions between the hidden layer and the output layer.

[0081] When the error does not meet the set value, the weights and thresholds between the layers need to be corrected. The correction values are calculated by taking the derivatives of the weights and thresholds respectively using the error expression. The correction expressions are as shown in equations (6) and (7)

[0082]

[0083]

[0084] where n represents the number of all output nodes. After the network structure is corrected through multiple feedbacks, it can be used to accurately predict the fuel consumption of the vehicle during driving.

[0085] S3.2 Further correct the data-driven model

[0086] After the vehicle enters the stationary state from the driving state, use the non-linear model for predicting the remaining fuel in the stationary state to estimate the remaining fuel of the vehicle after driving a certain distance again, calculate the fuel consumption value of the vehicle during this driving mileage, and use this fuel consumption value to correct the data-driven model to further improve the accuracy of fuel consumption prediction.

[0087] S3.3 Construct a real-time accurate prediction model for the remaining fuel

[0088] The data-driven model for fuel consumption prediction can obtain the real-time fuel consumption prediction model of the vehicle after being corrected multiple times, so as to achieve the purpose of real-time and accurate prediction of the real-time fuel consumption of the vehicle. The remaining fuel of the vehicle in the stationary state can be calculated using the non-linear model in S2. The remaining fuel of the vehicle during driving is calculated by subtracting the real-time fuel consumption of the vehicle from the remaining fuel in the stationary state, and then the remaining fuel in the fuel tank of the vehicle during driving is predicted.

[0089] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the remaining fuel quantity in an automobile fuel tank based on data-driven and non-linear models, characterized in that, It includes the following steps: S1 By measuring the dimensional parameters of the physical model of the vehicle fuel tank, including the height h, total length l, width d, and thickness t of the fuel tank, accurately establish a three-dimensional CAD model of the fuel tank using three-dimensional modeling software, providing a CAD model for the establishment of a non-linear model for predicting the remaining fuel volume in the fuel tank under the vehicle's stationary state; S2 Construct a non-linear model for predicting the remaining fuel volume in the fuel tank under the stationary state; S3 Establish a model for predicting the remaining fuel volume in the fuel tank under the vehicle's driving state; Step S2 is specifically as follows: S2.1 Obtain multiple measurement points for calculating the remaining fuel volume and establish a horizontal plane Detect the situations when the oil level is at 1 / 8h, 1 / 4h, 3 / 8h, 1 / 2h, 5 / 8h, 3 / 4h, and 7 / 8h respectively through the oil level sensor to obtain 7 measurement points for calculating the remaining fuel volume. Then, use the positions of each measurement point to establish a horizontal plane. Based on the shape and structural parameters of the fuel tank below the horizontal plane, obtain the space occupied by the part of the fuel tank at the 7 measurement points and below; S2.2 Calculate the remaining fuel volume in the fuel tank at the measurement points According to the horizontal plane, fuel tank structure and dimensional parameters established based on the location of the measurement points, the CAD model of the fuel tank is imported into the finite element analysis software to calculate the volume of the space below the liquid level position of different measurement points and its horizontal plane. Using the volume calculation formula Through mesh generation and numerical integration techniques, calculate the volume of the space of the fuel tank below the liquid level position of different measurement points, so as to obtain the remaining fuel volumes V1, V2, V3, V4, V5, V6, V7 corresponding to the position of each liquid level measurement point; S2.3 Establish a non-linear model for the remaining fuel volume in the fuel tank under the vehicle's stationary state Using the above 7 measurement points and their corresponding remaining fuel volumes, establish a non-linear model for the remaining fuel volume in the fuel tank under the vehicle's stationary state. The establishment process of the non-linear model is as follows: First, use the known liquid level heights of the measurement points and the corresponding remaining fuel volumes to establish an interval non-linear model through non-linear interpolation to obtain a non-linear model between the liquid level height and the remaining fuel volume. The expression is where f(V) is the remaining fuel volume to be predicted, hx is the height of the oil level position to be predicted, and hi (i = 1 to 7) are the liquid level heights of the 7 known measurement points in S2.2; Step S3 is specifically as follows: S 3.1 Construct a data-driven model for vehicle fuel consumption, that is, establish a mapping relationship between vehicle driving state data and fuel consumption using a BP neural network; S 3.2 Further correct the data-driven model After the vehicle enters the stationary state from the driving state, use the non-linear model for predicting the remaining fuel volume under the stationary state to estimate the remaining fuel volume after the vehicle has traveled a certain distance again, calculate the fuel consumption value of the vehicle during this driving mileage, and use this fuel consumption value to correct the data-driven model; S 3.3 Construct a real-time and accurate prediction model for the remaining fuel volume The data-driven model for fuel consumption prediction can obtain the real-time fuel consumption prediction model of the vehicle after being corrected multiple times, so as to achieve the purpose of real-time and accurate prediction of the vehicle's real-time fuel consumption. The remaining fuel volume of the vehicle under the stationary state is calculated using the non-linear model in S2. The remaining fuel volume of the vehicle during driving is calculated by subtracting the vehicle's real-time fuel consumption from the remaining fuel volume under the stationary state, and then the remaining fuel volume in the fuel tank under the vehicle's driving state is predicted.

2. The method for predicting the remaining fuel quantity in an automobile fuel tank based on data-driven and non-linear models according to claim 1, wherein, Step S3.1 is specifically as follows: S 3.1.1 Determine the input layer data The input layer contains 3 neurons. The input of the input layer is the speed, acceleration, and engine torque data related to the vehicle state and fuel consumption, and the output layer is the fuel consumption result. Among them, the data of speed, acceleration, and engine speed are sourced from the vehicle on-board sensor data. And the collected data is preprocessed. First, the outliers in the speed, acceleration, and engine speed data are removed through the amplitude filtering algorithm. After normalization, the data is summarized and unified into samples and placed in the input layer. The normalization method uses the following formula where TV and CV are the sample data after and before normalization respectively, and MinV and MaxV are the minimum and maximum data values; S 3.1.2 Determine the data of the hidden layer The selection of the number of hidden layers is based on the following formula; where h m is the number of hidden layers, n i is the number of input layers, l o is the number of output layers, and α is a constant with a value ranging from 1 to 10; The expression for calculating the output of the hidden layer is as follows Among them, hy i is the output result of the hidden layer, f is the activation function, and w ij is the weight function, θ j represents the threshold function, and x i is the data of the input layer; S 3.1.3 The result of the output layer is compared with the expected value and the error is calculated. According to the error, it is judged whether the feedback termination condition is met. If not, the weights and thresholds between each layer are further corrected where the expected value is the original fuel consumption data of the vehicle or the fuel consumption data measured by the fuel consumption meter. The error of the output layer is calculated through the following expression Among them, Ev k is the expected output result, Op k is the output layer result, w jk and θ k are the weight and threshold functions between the hidden layer and the output layer; When the error does not meet the set value, the weights and thresholds between each layer need to be corrected. The correction values are calculated by taking the derivatives of the weights and thresholds respectively using the error expression. The correction expressions are as follows: where n represents the number of all output nodes.

Images