A Hierarchical Prediction Method for the Driving Range of Electric Vehicles Driven by the Collaboration of Knowledge and Data

Abstract

The present invention is applicable to the technical field of electric vehicles, and provides a hierarchical prediction method for the driving range of electric vehicles driven by knowledge data collaboration. The present invention not only considers the influence of energy consumption on the driving range of electric vehicles, but also considers the influence of available energy on the driving range, thereby improving the prediction accuracy of the driving range of electric vehicles; the hierarchical prediction model is adopted to transform the problem of predicting the driving range of electric vehicles into the problem of predicting the energy consumption rate and the actual energy coefficient of electric vehicles, improving the flexibility and scalability of the model, reducing the development time and cost, and greatly improving the accuracy of predicting the driving range of electric vehicles. In addition, through the model driven by data and mechanism collaboration, the limitations of the physical model are overcome, the interpretability is enhanced, the dependence on large-scale data is reduced, the problem of model overfitting is effectively avoided, the trust degree of users is enhanced, and the promotion of electric vehicles is facilitated.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric vehicles, and particularly relates to a hierarchical prediction method for the driving range of electric vehicles driven by the collaboration of knowledge and data. Background Art

[0002] In recent years, the rapid development of electric vehicles has benefited from the promotion of national policies, the growth of market demand, and the rapid progress of technology. The electrification of transportation is regarded as a key way to improve energy efficiency and reduce carbon emissions. However, during the popularization process of electric vehicles, driving range anxiety has become the main concern of consumers, fearing that the battery power will run out during long-distance driving due to insufficient battery life, and it is difficult to find charging facilities, which affects the completion of the journey.

[0003] The driving range refers to the maximum distance that an electric vehicle can continuously travel under different road or environmental conditions when the battery is fully charged. It is an important indicator to measure the battery performance and vehicle energy efficiency of electric vehicles, usually measured in kilometers (km) or miles. However, under cold climates and complex driving conditions, the battery performance and energy consumption rate of electric vehicles will fluctuate significantly, which makes it a major challenge for the current industry to accurately predict the driving range of electric vehicles. The traditional methods for predicting the driving range of electric vehicles mainly have the following problems:

[0004] First of all, traditional methods for predicting the driving range of electric vehicles usually only consider the impact of energy consumption of electric vehicles, and often ignore the impact of available energy on the driving range of electric vehicles. For example, battery degradation, low winter temperature resulting in smaller available energy of the battery, etc. These factors all have a certain impact on the driving range. This kind of neglect leads to low prediction accuracy of traditional models.

[0005] Secondly, traditional prediction models for the driving range of electric vehicles usually adopt a single-layer structure, processing all factors at one level. This design not only reduces the efficiency of the model, but also limits the flexibility of the model, making it difficult to expand modularly as the demand changes.

[0006] In addition, data-driven models have high data dependence, lack interpretability and extrapolation ability, and perform poorly in data-scarce or new scenarios; while mechanism-based models, although having advantages in interpretability and physical consistency, are complex to model, difficult to estimate parameters, and difficult to handle complex non-linear relationships, and their generalization ability is also poor.

[0007] In view of the above problems, the present invention proposes a hierarchical prediction method for the driving range of electric vehicles driven by the collaboration of knowledge and data. Summary of the Invention

[0008] The purpose of the present invention is to provide a hierarchical prediction method for the driving range of electric vehicles driven by the collaboration of knowledge and data, aiming to solve the problems proposed in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A hierarchical prediction method for the driving range of electric vehicles driven by the collaboration of knowledge and data, comprising the following steps:

[0011] Step S1, collect input features, including historical driving data, vehicle speed, battery temperature, and battery state;

[0012] Step S2, transform the problem of predicting the driving range of an electric vehicle into the problem of predicting the energy consumption rate and the actual energy coefficient of the electric vehicle;

[0013] Step S3, combine the prediction results of the energy consumption rate and the actual energy coefficient to generate a predicted value of the driving range of the electric vehicle.

[0014] Further, the specific process of step S2 is as follows:

[0015] Step 21, use the MLP model to predict the energy consumption rate at the start stage of the journey, and use the LSTM model to predict the energy consumption rate during the driving stage;

[0016] Step 22, first obtain the true actual energy coefficient label by using each charging process data, and adopt the Monte Carlo simulation method to improve the credibility of the label; then use temperature, constant current charging section data, and mileage data as input features to identify fast charging and slow charging, and respectively predict the actual energy coefficient of fast charging and the actual energy coefficient of slow charging; finally, use the LSTM model to learn the time series change of the actual energy coefficient during the vehicle driving process.

[0017] Further, the specific method of step 21 is as follows: first obtain the true energy consumption rate label by using each discharge process data; then use temperature, driving information, and start and end SOC as input features to calculate the energy consumption of each trip segment; then adopt a structure that switches between the MLP model and the LSTM model for prediction; at the start stage of the journey, due to insufficient data, use the MLP model to call historical data for prediction, and when the data volume reaches the preset threshold, the MLP model automatically switches to the LSTM model to learn the time series change of the collected driving data; finally, use hyperparameter tuning and K-fold cross-validation to improve the prediction performance of the model.

[0018] Further, the specific method of step 22 is as follows:

[0019] ;

[0020] Among them, is the input gate; is the forget gate; is the output gate; is the time step t of the candidate memory cell state; is the time step t of the cell state; is the time step t of the hidden state; is the activation function; is the input weight matrix corresponding to the input gate; is the input weight matrix corresponding to the forget gate; is the input weight matrix corresponding to the output gate; is the input weight matrix corresponding to the candidate memory cell state; is the current time step t of the input vector; is the recurrent weight matrix of the input gate; is the recurrent weight matrix of the forget gate; is the recurrent weight matrix of the output gate; is the recurrent weight matrix of the candidate memory cell; is the time step t- 1 of the hidden state; is the bias vector of the input gate; is the bias vector of the forget gate; is the bias vector of the output gate; is the bias vector of the candidate memory cell state; is the time step t of the forget gate activation value; is the time step t- 1 of the cell state; is the time step t of the input gate activation value; is the time step t of the output gate activation value.

[0021] Furthermore, the specific manner of step S3 is as follows:

[0022] Combining the prediction results of the energy consumption rate and the actual energy coefficient, the predicted value of the driving range of the electric vehicle is output through the MLP model, and the MLP model includes an input layer, a hidden layer, and an output layer;

[0023] The input layer receives input features. Assuming the input vector is , X is the entire input feature vector, containing all input features; x 1 , x 2 , …, x nFor each element in the input feature vector, i.e., each feature or variable; n is the number of features; the input layer passes the feature vector to the first hidden layer;

[0024] The neurons in the hidden layer receive the input features from the first input layer and perform a linear transformation through weighted summation:

[0025] ;

[0026] where, represents the output of the l -th neuron in the j -th layer; is the activation function; n represents the number of neurons in the previous layer; represents the l -th input to the i -th neuron in the j -th layer; represents l- the output of the j -th neuron in the -th layer; j represents the bias of the

[0027] -th neuron;

[0028] ;

[0029] where, represents the activation value vector of the L- -th layer; is the activation function; represents L- the output value of the

[0030] Finally, the predicted value of the driving range of the electric vehicle is output at the output layer, and the calculation formula is as follows:

[0031] ;

[0032] where, represents the weight matrix of the last layer; is the activation value vector of the L- -th layer; is the bias term of the output layer; is the final output, i.e., the predicted value of the driving range.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. The present invention improves the traditional prediction model for the driving range of electric vehicles. It not only considers the impact of energy consumption on the driving range of electric vehicles, but also takes into account the influence of available energy on the driving range, including factors such as temperature and battery degradation, thereby improving the prediction accuracy of the driving range of electric vehicles.

[0035] 2. Aiming at the complexity problem of the single-layer model, the present invention uses a hierarchical prediction model to transform the problem of predicting the driving range of electric vehicles into the problem of predicting the energy consumption rate and the actual energy coefficient of electric vehicles, enabling each module to focus on the completion of sub-tasks, improving the flexibility and scalability of the model, reducing the development time and cost, and greatly enhancing the accuracy of predicting the driving range of electric vehicles.

[0036] 3. The present invention adopts a driving range prediction model driven by the collaboration of data and mechanism. It not only overcomes the shortcomings of physical models that are difficult to capture complex non-linear relationships and have insufficient prediction real-time performance, but also enhances the interpretability of the model by introducing physical mechanisms, reduces the dependence on large-scale data, effectively avoids the problem of model overfitting, enhances the user's trust level, and is conducive to the popularization of electric vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flowchart of the present invention.

[0038] Figure 2 is a comparison chart of the predicted driving range of the test vehicle.

[0039] Figure 3 is an effect diagram of the real-time prediction of the driving range of the test vehicle. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0041] The following describes in detail the specific implementation of the present invention with reference to specific embodiments.

[0042] As Figure 1As shown in the figure, a hierarchical prediction method for the driving range of an electric vehicle driven by knowledge data collaboration provided by an embodiment of the present invention. The method realizes the prediction of the driving range based on a hierarchical prediction model. The hierarchical prediction model includes an input layer, an upper-layer modular estimation layer, and a lower-layer information fusion layer. The upper-layer modular estimation layer includes an energy consumption rate prediction module and an actual energy coefficient prediction module. The hierarchical prediction model based on real-world data is implemented through a software system, runs on the Ubuntu 18.04 operating system, and uses the PyTorch 2.3.1 package to construct the upper-layer modular estimation layer and the lower-layer information fusion layer.

[0043] The method includes:

[0044] 1) The input layer collects input features, including information such as historical driving data, vehicle speed, battery temperature, and battery state, providing a basis for the operation of the subsequent upper-layer modular estimation layer.

[0045] The key feature variables affecting the energy consumption rate are information such as the temperature during the heat release stage, the idle ratio, the idle energy consumption rate, the vehicle speed, and the starting and ending SOC. These information are used as the input features of the energy consumption rate prediction module.

[0046] The key feature variables affecting the actual energy coefficient are information such as the temperature during the charging stage, the total current, the total voltage, and the mileage data. These information are used as the input features of the actual energy coefficient prediction module.

[0047] 2) Mechanism analysis: According to the mechanism definition, the driving range is the ratio of the available energy to the energy consumption rate, and the available energy is related to the actual energy coefficient. The formula is expressed as follows:

[0048] ;

[0049] ;

[0050] Among them, RDR is the driving range of the electric vehicle, RE battery is the available energy of the electric vehicle, E cos is the energy consumption rate of the electric vehicle, SOC start is the initial state of charge value of the electric vehicle battery, SOC end is the final state of charge value of the electric vehicle battery, C is the actual energy coefficient of the electric vehicle. It can be seen that the driving range of the electric vehicle is closely related to its energy consumption rate and actual energy coefficient.

[0051] Therefore, based on this, the upper-layer modular estimation layer uses the mechanism knowledge (physical knowledge) in mechanism analysis to transform the problem of predicting the driving range of electric vehicles into the problem of predicting the energy consumption rate and the actual energy coefficient of electric vehicles, so as to achieve an accurate prediction of the energy consumption rate of electric vehicles.

[0052] a. The energy consumption rate refers to the electric energy consumed per kilometer during the driving of an electric vehicle. The energy consumption rate prediction module uses a Multi-Layer Perceptron (MLP) model to predict the energy consumption rate at the beginning stage of the journey and a Long Short-Term Memory (LSTM) model to predict the energy consumption rate during the driving stage. The specific method is as follows:

[0053] First, use the data of each discharge process to obtain the true energy consumption rate label; then use the temperature (the highest temperature in the heat release stage, the lowest temperature in the heat release stage), driving information (idle ratio, idle energy consumption rate, vehicle speed) and start and end SOC (start SOC, end SOC) as input features, and use the mechanism knowledge (physical knowledge) used in the above mechanism analysis to calculate the energy consumption of each journey segment; then adopt a structure that switches between the MLP model and the LSTM model for prediction; at the beginning stage of the journey, due to insufficient data, use the MLP model to call historical data for prediction. When enough data is collected (driving time ≥ 25 min), the MLP model automatically switches to the LSTM model to learn the temporal changes of the collected driving data, so as to improve the accuracy of the energy consumption rate prediction model; finally, use hyperparameter tuning and K-fold cross-validation to improve the prediction performance of the model.

[0054] b. The actual energy coefficient reflects the effectiveness of energy utilization in the actual use of electric vehicles. The actual energy coefficient prediction module first uses the data of each charging process to obtain the true actual energy coefficient label. In order to enhance the prediction accuracy of the model, the Monte Carlo simulation method is adopted to improve the credibility of the training label, ensuring that the model can more accurately reflect the actual energy utilization situation; then use the temperature (the highest temperature in the charging stage, the lowest temperature in the charging stage), constant current charging section data (total current, total voltage) and mileage data as input features to identify fast charging and slow charging, and predict the actual energy coefficient of fast charging and slow charging respectively; finally, use the LSTM model to learn the temporal changes of the actual energy coefficient during the vehicle driving process. The specific method is as follows:

[0055] ;

[0056] Among them, is the input gate, which determines how much new information is introduced into the memory unit; is the forget gate, which controls how much old information is discarded; is the output gate, which determines how much information is output from the memory cell to the next time step or the output layer of the network; is the time step t is the candidate memory cell state at time step is the time step t is the cell state at time step is multiplied by the input gate and the result; is the time step t is the hidden state at time step is the activation function; is the input weight matrix corresponding to the input gate; is the input weight matrix corresponding to the forget gate; is the input weight matrix corresponding to the output gate; is the input weight matrix corresponding to the candidate memory cell state; is the current time step t input vector; is the recurrent weight matrix of the input gate; is the recurrent weight matrix of the forget gate; is the recurrent weight matrix of the output gate; is the recurrent weight matrix of the candidate memory cell; is the time step t- is the hidden state at time step 1; is the bias vector of the input gate; is the bias vector of the forget gate; is the bias vector of the output gate; is the bias vector of the candidate memory cell state; is the time step t is the activation value of the forget gate at time step is the time step t- is the cell state at time step 1; is the time step t is the activation value of the input gate at time step is the time step t is the activation value of the output gate at time step

[0057] 3) According to the mechanism analysis, it can be seen that the driving range of electric vehicles is strongly correlated with the energy consumption rate and the actual energy coefficient. Therefore, the lower-layer information fusion layer takes the energy consumption rate and the actual energy coefficient as the input features of this layer, and outputs the driving range of electric vehicles through the MLP model. The MLP model includes an input layer, a hidden layer, and an output layer; the specific method is as follows:

[0058] The input layer receives input features, which are the outputs from the upper-layer modular estimation layers, namely the energy consumption rate prediction module and the actual energy coefficient prediction module; assume the input vector from the upper-layer modular estimation layer is X = [x 1 , x 2 , …, x n , X is the entire input feature vector, containing all input features; x 1 , x 2 , …, x n are each element in the input feature vector, that is, each feature or variable; n is the number of features; the input layer passes the feature vector to the first hidden layer;

[0059] The neurons in the hidden layer receive the input features from the first input layer and perform a linear transformation through weighted summation:

[0060] ;

[0061] Among them, represents the output of the l th neuron in the j th layer; is the activation function; n represents the number of neurons in the upper layer (the l- 1st layer); represents the weight of the l th input to the i th neuron in the j th layer; represents l- the output of the j th neuron in the represents the bias of the j th neuron;

[0062] Subsequently, the output value of the linear transformation is non-linearly transformed through a non-linear activation function as follows:

[0063] ;

[0064] Among them, represents the activation value vector of the L- 1st layer; represents the activation function; represents L- the output value of the

[0065] Finally, the predicted value of the electric vehicle's driving range is output at the output layer, and the calculation formula is as follows:

[0066] ;

[0067] Among them, Represents the weight matrix of the last layer; is the activation value vector of the L- first layer; is the bias term of the output layer; is the final output, i.e., the predicted value of the cruising range.

[0068] Example 1, training and validation;

[0069] The collected actual driving data of electric vehicles is divided into a training set, a validation set, and a test set in chronological order. Specifically, the first 90% of the data is used for the training set and the validation set to train the model parameters and adjust the hyperparameters, and the last 10% of the data is used for the test set to evaluate the model performance. The construction and training of the model are based on the PyTorch framework, and all experiments are carried out on a PC server equipped with an Intel(R) Core(TM) i7-13700KF 3.40 GHz processor and a 12 GB NVIDIA GeForce RTX 4070Ti GPU.

[0070] To prevent the deep learning model from overfitting, the Dropout method with a probability of 0.3 is adopted and applied between the convolutional layer, the recurrent layer, and the fully connected layer. The Adam optimizer is selected, and the mean squared error (MSE) is used as the loss function. After adjusting the hyperparameters on the validation set, the time step is finally determined to be 25, the batch size is 64, and the bidirectional LSTM is configured with 64 hidden neurons to fully learn the time features. The entire model training is carried out for 30 epochs, and the model evaluation metrics are the mean absolute percentage error (MAPE) and the coefficient of determination ( R 2 ), to conduct a comprehensive and quantitative analysis of the experimental results. MAPE (mean absolute percentage error) is a statistical metric used to measure the accuracy of the model, mainly used to evaluate the deviation between the predicted value and the actual value of the model. MAPE represents the average percentage of the prediction error and can intuitively reflect the accuracy of the prediction result relative to the actual result. The smaller its value, the higher the prediction accuracy of the model. R 2 Measures the matching degree between the variability of the model prediction and the actual variability of the data. R 2 The value of R 2 usually ranges from 0 to 1. The closer the value is to 1, the better the model explains the variation of the data. The specific calculation formulas of MAPE and

[0071] ;

[0072] ;

[0073] Among them, is an index variable used to iterate through each observation; n is the total number of observations; is the actual cruising range of the electric vehicle, is the cruising range predicted by the hierarchical prediction model.

[0074] The trained model is used to infer the test set. The comparison chart of the predicted cruising range of the test vehicle is as shown in Figure 2 As shown. It can be seen from the figure that the R 2 of the hierarchical prediction model is closest to 1 and the MAPE is the smallest. The real-time prediction effect diagram of the cruising range of the test vehicle is as shown in Figure 3 As shown. It can be seen from the figure that the hierarchical prediction model shows excellent prediction results both before and after switching. The above results demonstrate the effectiveness and generality of the hierarchical prediction model. Vehicles No. 1-4 are predicted respectively using the hierarchical prediction model, the End-to-End model, and the CLTC cycle. The relevant evaluation indexes of the model test set are shown in Tables 1, 2, and 3.

[0075] Table 1. Prediction indexes of different test vehicles R 2

[0076]

[0077] Table 2. MAPE of prediction indexes of different test vehicles

[0078]

[0079] Table 3. Different indexes of real-time prediction of the cruising range of the test vehicle

[0080]

[0081] It can be seen from Table 1 that the prediction indexes of the hierarchical prediction model for vehicles No. 1-4 R 2 are closer to 1, indicating that the hierarchical prediction model better explains the variation of the data. It can be seen from Table 2 that the values of the prediction index MAPE of the hierarchical prediction model for vehicles No. 1-4 are smaller, indicating that the hierarchical prediction model has higher prediction accuracy. It can be seen from Table 3 that in the real-time prediction result indexes of the hierarchical prediction model, R 2 are closer to 1 and the values of MAPE are smaller, indicating that the real-time prediction of the hierarchical prediction model better explains the variability of the data and has higher prediction accuracy.

[0082] In summary, the hierarchical prediction model is mainly based on data-driven and the mechanism knowledge (physical knowledge) used in the above mechanism analysis, and adopts a hierarchical architecture, which can perform rapid training and real-time prediction with limited computing resources. In addition, the model uses K-fold cross-validation to improve the robustness of the model and enhances the prediction effect through hyperparameter tuning, making it have higher accuracy and better generalization ability compared with other single-layer or single models. Therefore, it can accurately estimate the driving range of actual electric vehicles within the full operating conditions. It can be seen that the hierarchical prediction method for the driving range of electric vehicles driven by knowledge and data collaboration designed by the present invention can accurately predict the driving range of electric vehicles in real time.

[0083] The above are only the preferred embodiments of the present invention. It should be noted that for those skilled in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent.

Claims

1. A knowledge data collaboratively driven electric vehicle driving range hierarchical prediction method, characterized in that: The following steps are involved: Step S1, collecting input features, including historical driving data, vehicle speed, battery temperature and battery status; Step S2, converting the problem of predicting the driving range of the electric vehicle into the problem of predicting the energy consumption rate and actual energy coefficient of the electric vehicle; Step S3, combining the prediction results of the energy consumption rate and the actual energy coefficient to generate a predicted value of the driving range of the electric vehicle; The specific process of step S2 is as follows: Step 21: Use the MLP model to predict the energy consumption rate at the beginning of the trip, and use the LSTM model to predict the energy consumption rate during the driving phase. Step 22: First, use each charging process data to obtain the real actual energy coefficient label, and use the Monte Carlo simulation method to improve the credibility of the label; then use the temperature, constant current charging segment data and mileage data as input features to perform fast and slow charging identification, and perform fast charging actual energy coefficient prediction and slow charging actual energy coefficient prediction respectively; Finally, the LSTM model is used to learn the time series changes of the actual energy coefficient during vehicle driving.

2. The knowledge data collaboratively driven electric vehicle driving range hierarchical prediction method according to claim 1 is characterized in that: The specific method of step 21 is: first, use each discharge process data to obtain the real energy consumption rate label; then use temperature, driving information and start and end SOC as input features to calculate the energy consumption of each trip segment; then use the structure of switching between MLP model and LSTM model to make predictions; At the beginning of the journey, due to insufficient data, the MLP model is used to call historical data for prediction. When the amount of data reaches the preset threshold, the MLP model autonomously switches to the LSTM model to learn the time series changes of the collected driving data. Finally, hyperparameter tuning and K-fold cross-validation are used to improve the prediction performance of the model.

3. The knowledge data collaboratively driven electric vehicle driving range hierarchical prediction method according to claim 1 is characterized in that: The specific method of step 22 is as follows: ； in, is the input gate; For the Gate of Oblivion; is the output gate; is the time step t The candidate memory cell states; is the time step t The unit status; is the time step t The hidden state of is the activation function; is the input weight matrix corresponding to the input gate; is the input weight matrix corresponding to the forget gate; Is the input weight matrix corresponding to the output gate; is the input weight matrix corresponding to the candidate memory unit state; is the current time step t The input vector of is the recursive weight matrix of the input gate; is the recursive weight matrix of the forget gate; is the recursive weight matrix of the output gate; is the recursive weight matrix of the candidate memory unit; is the time step t- The hidden state of 1; is the bias vector of the input gate; is the bias vector of the forget gate; is the bias vector of the output gate; is the bias vector of the candidate memory cell state; is the time step t The forget gate activation value of is the time step t- 1's unit status; is the time step t The input gate activation value of is the time step t The output gate activation value of .

4. The knowledge data collaboratively driven electric vehicle driving range hierarchical prediction method according to claim 1 is characterized in that: The specific method of step S3 is as follows: Combining the prediction results of the energy consumption rate and the actual energy coefficient, the predicted value of the driving range of the electric vehicle is output through the MLP model, and the MLP model includes an input layer, a hidden layer and an output layer; The input layer receives input features. Assume that the input vector is , X is the entire input feature vector, including all input features; x1,x2,…,x n For each element in the input feature vector, that is, each feature or variable; n is the number of features; the input layer passes the feature vector to the first hidden layer; The neurons in the hidden layer receive the input features of the first input layer and perform linear transformation through weighted summation: ； in, Indicates l Tier j The output of a neuron; is the activation function; n Indicates the number of neurons in the previous layer; Indicates l Tier i Input to j The weights of the neurons; express l- 1st floor j The output of a neuron; Indicates j The bias of each neuron; Subsequently, the output value of the linear transformation is transformed nonlinearly through a nonlinear activation function as follows: ； in, Indicates L- The activation value vector of layer 1; represents the activation function; express L- Output value of layer 1; Finally, the predicted value of the electric vehicle's driving range is output in the output layer, and the calculation formula is as follows: ； in, Represents the weight matrix of the last layer; It is L- The activation value vector of layer 1; is the bias term of the output layer; is the final output, i.e. the predicted driving range.

Images