A method and system for predicting vehicle speed in commercial vehicles based on E-Power architecture

Abstract

This invention provides a method and system for predicting vehicle speed in commercial vehicles using the E-Power architecture, relating to the field of commercial vehicle speed measurement technology. First, vehicle operating data is collected via hardwired and CAN bus lines. Second, the PCA method is used to extract features from the collected vehicle operating data, obtaining variables highly correlated with vehicle speed, which serve as input to the speed prediction model, improving computational efficiency. Finally, a speed prediction model is established based on a recursive fuzzy neural network and a multi-step prediction strategy, achieving accurate prediction of vehicle speed during operation and improving fuel economy and driving comfort.

Description

Technical Field

[0001] This invention relates to the field of commercial vehicle speed measurement technology, and in particular to a method and system for predicting the speed of commercial vehicles based on the E-Power architecture. Background Technology

[0002] Commercial vehicles, as a vital part of the national economy, bear heavy responsibilities in transportation and economic development. Hybrid electric vehicles, with their excellent power performance, have become the most popular new energy vehicles. In particular, the E-Power architecture vehicle's engine is only used to drive the generator to produce electricity and does not directly participate in driving the vehicle. Users do not need to find charging stations, avoiding range anxiety. More importantly, this architecture maximizes the engine's operating point within the range of highest combustion efficiency, achieving energy conservation and emission reduction.

[0003] Existing technology CN201710595427.4 discloses a vehicle speed prediction method based on an adaptive neural fuzzy model. This method acquires a training sample set, including driver intention data and corresponding vehicle speed data; trains the training sample set to establish an adaptive neural fuzzy vehicle speed prediction model; acquires real-time vehicle count data and driver intention data, and inputs them into the adaptive neural fuzzy vehicle speed prediction model to predict the vehicle speed for the next time period. It can be seen that this scheme uses a TS fuzzy inference model to obtain corresponding acceleration or braking intention data from the collected pedal opening and pedal opening change rate data to achieve vehicle speed prediction. This prediction method does not consider vehicle wheel speed, driving mode, gear status, clutch status, remaining battery charge, battery temperature, battery current, battery voltage, engine speed, engine coolant temperature, generator speed, and generator torque, etc., to comprehensively predict vehicle speed, which will lead to low prediction accuracy, failing to meet prediction requirements and affecting vehicle fuel economy and comfort. Summary of the Invention

[0004] This invention provides a method for predicting vehicle speed in E-Power architecture commercial vehicles. The method can combine various vehicle operating data for comprehensive analysis and calculation to obtain accurate vehicle speed prediction, thereby meeting the requirements of vehicle fuel economy and comfort.

[0005] The methods include:

[0006] S101. Collect vehicle operation data;

[0007] S102. Preprocess the collected vehicle operation data and extract features using the PCA method;

[0008] S103. Establish a vehicle speed prediction model and obtain vehicle speed information based on multi-step prediction using a recursive fuzzy neural network.

[0009] Further, it should be noted that in step S102, a PCA method model is established. The 16 obtained influencing factors are used as independent variables x, where x = [x1, x2, …, x 16 , and the vehicle speed is the dependent variable y;

[0010] Among them, x1, x2, …, x 16 respectively represent the throttle pedal opening, brake pedal opening, throttle pedal opening change rate, brake pedal opening change rate, wheel speed, driving mode, gear state, clutch state, remaining battery power, battery temperature, battery current, battery voltage, engine speed, engine water temperature, generator speed, and generator torque.

[0011] Feature extraction is carried out using the following formula of the PCA method:

[0012] F A U = UA

[0013]

[0014] Among them, F A is the covariance matrix, Λ is the value on the diagonal of the relevant eigenvalue matrix, U is the principal component matrix of the β-dimensional eigenvector, x i represents the i-th input variable, μ is the mean of the input variables, and n is the number of variables;

[0015]

[0016] Among them, ρ i is the contribution rate of the i-th variable, and λ i is the eigenvalue of the i-th variable.

[0017] Further, it should be noted that in step S103, the vehicle speed is obtained by multi-step prediction of a recurrent fuzzy neural network. The structure of the recurrent fuzzy neural network is divided into four layers: input layer, membership function layer, rule layer, and output layer;

[0018] The structure is a connection method of 7-M-M-1, where M is a positive integer and 2 < M < 30. The connection weight value between the input layer and the membership function layer is 1. The actual output of the recurrent fuzzy neural network is represented as y(t); The vehicle speed prediction method based on the recurrent fuzzy neural network is calculated as:

[0019]

[0020] Where x(t) = [x1(t), x2(t), x3(t), x4(t), x5(t), x6(t), x7(t)] represents the output of the input layer at time t, x1(t) represents the accelerator pedal opening at time t, x2(t) represents the brake pedal opening at time t, x3(t) represents the rate of change of the accelerator pedal opening at time t, x4(t) represents the rate of change of the brake pedal opening at time t, x5(t) represents the remaining battery charge at time t, x6(t) represents the generator speed at time t, x7(t) represents the generator torque at time t, f represents the functional relationship between x(t) and y(t), and w j (t) represents the weight between the j-th neuron and the output neuron in the regular layer at time t, β ij (t) represents the weight between the i-th neuron in the input layer and the j-th neuron in the membership function layer at time t, β ij (t) = 1, m ij (t) represents the i-th element of the center value of the j-th neuron in the membership function layer at time t, σ ij (t) represents the i-th element of the width value of the j-th neuron at time t in the membership function layer. The recursive fuzzy neural network adds a self-feedback connection to the membership function layer. θ ij (t) is the feedback weight of the self-feedback connection of the membership function layer.

[0021] O 2 ij (t-1) is the feedback value of the self-feedback connection of the membership function layer, where:

[0022]

[0023] Where, β ij (t-1) represents the weight between the i-th neuron in the input layer and the j-th neuron in the membership function layer at time t-1, β ij (t-1)=1,m ij (t-1) represents the i-th element of the center value of the j-th neuron in the membership function layer at time t-1, σ ij (t-1) represents the i-th element of the width value of the j-th neuron in the membership function layer at time t-1, θ ij (t-1) is the feedback weight of the self-feedback connection of the membership function layer, O2 ij (t-2) is the feedback value of the self-feedback connection of the membership function layer; the error of the recursive fuzzy neural network is defined as:

[0024]

[0025] Where m is the number of samples, y dy(t) is the expected output of the recurrent fuzzy neural network at time t, and y(t) is the actual output of the recurrent fuzzy neural network at time t.

[0026] It should be further explained that the model calibration process is as follows:

[0027] ① Given that the initial number of hidden layer neurons in a recurrent fuzzy neural network is M, where M is a positive integer greater than 2, and the inputs of the recurrent fuzzy neural network are x(1), x(2), ..., x(t), ..., x(m), the corresponding expected output is y. d (1), y d (2), ..., y d (t), ..., y d (m), where m sets of data are used as training samples for the recursive fuzzy neural network model, and the expected error value is set to E. d E d ∈(0,0.01), initial center value m j In (1), the assignment range for each variable is [-2, 2], m j (1)=(m 1j (1), m 2j (1), ..., m ij (1)), m ij (1) represents the i-th element of the center vector of the j-th neuron in the membership function layer at the initial time;

[0028] Initial center width σ j (1) has an assignment interval of [0,1], σ j (1)=(σ 1j (1), σ 2j (1), …, σ ij (1)), σ ij (1) represents the i-th element of the width value vector of the j-th neuron in the membership function layer at the initial time;

[0029] Initial feedback connection weights θ ij The range of values ​​for (t) is [0,1], j = 1, 2, ..., M;

[0030] The initial weights w(1) are assigned values ​​in the range [-1, 1], w(1) = (w1(1), w2(1), ..., w j (1)), w j (1) represents the connection weight between the j-th neuron of the rule layer and the output layer at the initial time;

[0031] ② Set the number of learning steps s = 1;

[0032] ③When \(t = s\), calculate the output \(y(t)\) of the recurrent fuzzy neural network according to formula (1), and use the gradient descent algorithm to adjust the parameters of the recurrent fuzzy neural network as follows:

[0033]

[0034]

[0035] \(\theta\) ij \((t + 1)=\theta\) ij \((t)-\eta\) θ \((y\) d \((t)-y(t))w\) j \((t)O\) ij \((t)y(t - 1)\)

[0036] \(w\) j \((t + 1)=w\) j \((t)-\eta\) w \((y\) d \((t)-y(t))O\) ij \((t)\)

[0037] where \(\eta\) m is the learning rate of the center vector \(m\) ij \(\eta\) m \(\in(0, 0.01]\); \(\eta\) σ is the learning rate of the width \(\sigma\) j \(\eta\) σ \(\in(0, 0.01]\), \(\eta\) θ is the learning rate of the feedback connection weight \(\theta\) ij \(\eta\) θ \(\in(0, 0.02]\), \(\eta\) w is the learning rate of the connection weight \(w\) j \(\eta\) w \(\in(0, 0.01]\), \(m\) ij \((t + 1)\) is the \(i\)-th element of the center vector of the \(j\)-th neuron in the membership function layer at time \(t + 1\), \(\sigma\) ij \((t + 1)\) is the \(i\)-th element of the width vector of the \(j\)-th neuron in the membership function layer at time \(t + 1\), \(\theta\) ij \((t + 1)\) is the feedback connection weight of the \(i\)-th element of the \(j\)-th neuron in the rule layer at time \(t + 1\), \(w\) j \((t + 1)\) is the connection weight between the \(j\)-th neuron in the rule layer and the output neuron at time \(t + 1\);

[0038] ④Increase the learning step \(s\) by 1. If the step \(s\lt N\), go back to step ③ for continued training. If \(s = N\), go to step ⑤;

[0039] ⑤Calculate the performance of the recurrent fuzzy neural network according to the error formula. Let \(E\) dFor ideal error, if E(t) ≥ E d If so, proceed to step ③ to continue training, if E(t) <E d If so, then stop adjusting;

[0040] The multi-step formula for predicting vehicle speed is as follows:

[0041]

[0042] Where h∈(1,…,H) is the number of prediction steps in multi-step prediction, and d is the dimension of the input variable, also called the number of time delay steps. It is the output of the output neuron after h steps, where h is 5.

[0043] The present invention also provides an E-Power architecture commercial vehicle speed prediction system, the system comprising: a data acquisition module, a data preprocessing module, and a speed prediction module;

[0044] The data acquisition module is connected to the vehicle's controller via signal hardwire and CAN bus to acquire vehicle operating data;

[0045] The data preprocessing module is used to preprocess the collected vehicle operation data and to extract features using the PCA method;

[0046] The vehicle speed prediction module is used to establish a vehicle speed prediction model and obtain vehicle speed information based on multi-step prediction using a recursive fuzzy neural network.

[0047] As can be seen from the above technical solutions, the present invention has the following advantages:

[0048] The E-Power architecture commercial vehicle speed prediction method provided by this invention utilizes PCA technology. By establishing a prediction model and employing sensor monitoring and data transmission technologies, it obtains information such as accelerator pedal opening, brake pedal opening, accelerator pedal opening rate of change, brake pedal opening rate of change, wheel speed, driving mode, gear status, clutch status, remaining battery charge, battery temperature, battery current, battery voltage, engine speed, engine coolant temperature, generator speed, and generator torque. It then establishes a principal component analysis (PCA) model to predict the commercial vehicle's speed. This effectively solves the problem that traditional speed prediction data considers only a single factor, failing to meet prediction requirements and thus affecting vehicle fuel economy and comfort.

[0049] This invention can collect and summarize vehicle operation data, and allows users to set collection parameters according to prediction needs, effectively improving prediction accuracy. It can also efficiently collect, store, and process prediction process and result data, and achieve full-process monitoring based on vehicle driving status, using a multi-dimensional space to describe the entire prediction process. This improves prediction accuracy and efficiency, controls vehicle fuel efficiency, and thus achieves accurate and scientific vehicle speed prediction. Attached Figure Description

[0050] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 Flowchart of E-Power architecture commercial vehicle speed prediction method;

[0052] Figure 2 This is a diagram of the recursive fuzzy neural network structure.

[0053] Figure 3 This is a schematic diagram of the E-Power architecture commercial vehicle speed prediction system. Detailed Implementation

[0054] The E-Power architecture commercial vehicle speed prediction method provided by this invention aims to solve the problem that existing technologies have too few speed prediction indicators and do not fully cover the vehicle's operating status, resulting in inaccurate predictions that affect the vehicle's fuel economy and comfort.

[0055] This invention relates to a vehicle speed prediction method for E-Power architecture commercial vehicles, which can be applied to hybrid vehicles to improve vehicle safety, fuel economy, and emissions performance. The speed prediction method involves: first, collecting vehicle operating data via hardwired data and CAN messages; second, extracting relevant variables using feature analysis methods to obtain the input for the prediction model; and finally, establishing a speed prediction model based on a recursive fuzzy neural network.

[0056] The E-Power architecture commercial vehicle speed prediction method can also acquire and process related data based on artificial intelligence technology. Among them, the digital twin-driven intelligent diagnostic method for CNC machine tools utilizes digital computers or digitally controlled machines to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. This includes theories, methods, technologies, and application devices.

[0057] The E-Power architecture's commercial vehicle speed prediction method utilizes Principal Component Analysis (PCA) technology. By establishing a prediction model and employing sensor monitoring and data transmission technologies, it obtains information on accelerator pedal opening, brake pedal opening, accelerator pedal opening rate of change, brake pedal opening rate of change, wheel speed, driving mode, gear status, clutch status, remaining battery charge, battery temperature, battery current, battery voltage, engine speed, engine coolant temperature, generator speed, and generator torque. This information is then used to build a PCA model to predict the commercial vehicle's speed. This effectively addresses the problem of traditional speed prediction methods using limited data, which fails to meet prediction requirements and negatively impacts fuel economy and comfort.

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Please see Figures 1 to 2 The diagram shows a flowchart of a vehicle speed prediction method for commercial vehicles based on the E-Power architecture, in a specific embodiment. The method includes:

[0060] S101. Collect vehicle operation data;

[0061] The prediction process of this invention involves multiple parameters, which may include, but are not limited to, accelerator pedal opening, brake pedal opening, accelerator pedal opening change rate, brake pedal opening change rate, wheel speed, driving mode, gear status, clutch status, remaining battery charge, battery temperature, battery current, battery voltage, engine speed, engine coolant temperature, generator speed, and generator torque.

[0062] S102. Preprocess the collected vehicle operation data and extract features using the PCA method;

[0063] Using vehicle operation data as the research object, a principal component analysis (PCA) model was established, with 16 influencing factors obtained as independent variables x, x = [x1, x2, ..., x...]. 16 [The vehicle speed is the dependent variable y;]

[0064] Where x1, x2, ..., x 16 These are respectively represented as accelerator pedal opening, brake pedal opening, accelerator pedal opening change rate, brake pedal opening change rate, wheel speed, driving mode, gear status, clutch status, remaining battery charge, battery temperature, battery current, battery voltage, engine speed, engine coolant temperature, generator speed, and generator torque.

[0065] The accelerator pedal opening range is [0, 100].

[0066] The brake pedal opening range is [0, 100].

[0067] The rate of change of accelerator pedal opening is in the range of [0,1].

[0068] The rate of change of brake pedal opening ranges from [0,1].

[0069] The driving modes are divided into pure electric mode, hybrid mode, parking mode, and energy recovery mode;

[0070] The gear positions are divided into 1st gear, 2nd gear, and reverse gear;

[0071] The clutch states are divided into disengagement, engagement, and partial engagement;

[0072] The remaining battery charge range is [0, 100]. The battery temperature is in degrees Celsius, the battery current is in amperes (A), the battery voltage is in volts (V), the engine speed is in rpm / min, the engine coolant temperature is in degrees Celsius, the engine speed is in rpm / min, and the generator torque is in nits (Nm).

[0073] Feature extraction is performed using the following formula based on the PCA method:

[0074] F A U = UA

[0075]

[0076] Among them, F A It is the covariance matrix, Λ is the diagonal value of the correlation eigenvalue matrix, U is the principal component matrix of the β-dimensional eigenvectors, and x i Let μ represent the i-th input variable, μ be the mean of the input variables, and n be the number of variables.

[0077]

[0078] Where, ρ i λ is the contribution rate of the i-th variable. i It is the eigenvalue of the i-th variable.

[0079] If the contribution rate exceeds 85%, it can be considered to be able to explain the output variable. By selecting an appropriate contribution rate, the relevant variables can then be selected.

[0080] S103. Establish a vehicle speed prediction model and obtain vehicle speed information based on multi-step prediction using a recursive fuzzy neural network.

[0081] The vehicle speed is obtained by multi-step prediction of a recurrent fuzzy neural network. The structure of the recurrent fuzzy neural network is divided into four layers: an input layer, a membership function layer, a rule layer, and an output layer.

[0082] The connection mode is 7-M-M-1, where M is a positive integer and 2 < M < 30. The connection weight between the input layer and the membership function layer is 1. The actual output of the recurrent fuzzy neural network is denoted as y(t). The vehicle speed prediction method based on the recurrent fuzzy neural network is calculated as follows:

[0083]

[0084] where x(t) = [x1(t), x2(t), x3(t), x4(t), x5(t), x6(t), x7(t)] represents the output of the input layer at time t. x1(t) represents the value of the throttle pedal opening at time t, x2(t) represents the value of the brake pedal opening at time t, x3(t) represents the value of the change rate of the throttle pedal opening at time t, x4(t) represents the value of the change rate of the brake pedal opening at time t, x5(t) represents the value of the remaining battery power at time t, x6(t) represents the value of the generator speed at time t, and x7(t) represents the value of the generator torque at time t. f represents the corresponding functional relationship between x(t) and y(t), w j (t) represents the weight between the j-th neuron in the rule layer and the output neuron at time t, and β ij (t) represents the weight between the i-th neuron in the input layer and the j-th neuron in the membership function layer at time t, and β ij (t) = 1, m ij (t) represents the i-th element of the central value of the j-th neuron in the membership function layer at time t, and σ ij (t) represents the i-th element of the width value of the j-th neuron in the membership function layer at time t. The recurrent fuzzy neural network adds a self-feedback connection in the membership function layer, and θ ij (t) is the feedback weight of the self-feedback connection in the membership function layer.

[0085] O 2 ij (t - 1) is the feedback value of the self-feedback connection in the membership function layer, where:

[0086]

[0087] where, β ij (t - 1) represents the weight between the i-th neuron in the input layer and the j-th neuron in the membership function layer at time t - 1, and β ij (t - 1) = 1, m ij (t - 1) represents the i-th element of the central value of the j-th neuron in the membership function layer at time t - 1, and σ ij(t-1) represents the i-th element of the width value of the j-th neuron in the membership function layer at time t-1, θ ij (t-1) is the feedback weight of the self-feedback connection of the membership function layer, O2 ij (t-2) is the feedback value of the self-feedback connection of the membership function layer; the error of the recursive fuzzy neural network is defined as:

[0088]

[0089] Where m is the number of samples, y d y(t) is the expected output of the recurrent fuzzy neural network at time t, and y(t) is the actual output of the recurrent fuzzy neural network at time t.

[0090] It should be further explained that the model calibration process is as follows:

[0091] ① Given that the initial number of hidden layer neurons in a recurrent fuzzy neural network is M, where M is a positive integer greater than 2, and the inputs of the recurrent fuzzy neural network are x(1), x(2), ..., x(t), ..., x(m), the corresponding expected output is y. d (1), y d (2), ..., y d (t), ..., y d (m), where m sets of data are used as training samples for the recursive fuzzy neural network model, and the expected error value is set to E. d E d ∈(0,0.01), initial center value m j In (1), the assignment range for each variable is [-2, 2], m j (1)=(m 1j (1), m 2j (1), ..., m ij (1)), m ij (1) represents the i-th element of the center vector of the j-th neuron in the membership function layer at the initial time;

[0092] Initial center width σ j (1) has an assignment interval of [0,1], σ j (1)=(σ 1j (1), σ 2j (1), …, σ ij (1)), σ ij (1) represents the i-th element of the width value vector of the j-th neuron in the membership function layer at the initial time;

[0093] Initial feedback connection weights θ ij The range of values ​​for (t) is [0,1], j = 1, 2, ..., M;

[0094] The initial weights w(1) are assigned values ​​in the range [-1, 1], w(1) = (w1(1), w2(1), ..., w j (1)), w j (1) represents the connection weight between the j-th neuron of the rule layer and the output layer at the initial time;

[0095] ② Set the number of learning steps s = 1;

[0096] ③t=s, calculate the output y(t) of the recurrent fuzzy neural network according to formula (1), and adjust the parameters of the recurrent fuzzy neural network using the gradient descent algorithm:

[0097]

[0098]

[0099] θ ij (t+1)=θ ij (t)-η θ (y d (t)-y(t))w j (t)O ij (t)y(t-1)

[0100] w j (t+1)=w j (t)-η w (y d (t)-y(t))O ij (t)

[0101] Where, η m Center vector m ij The learning rate, η m ∈(0, 0.01];η σ For width σ j The learning rate, η σ ∈(0, 0.01], η θ For the feedback connection weight θ ij The learning rate, η θ ∈(0, 0.02], η w For the connection weight w j The learning rate, η w ∈(0, 0.01], m ij (t+1) is the i-th element of the center vector of the j-th neuron in the membership function layer at time t+1, σ ij (t+1) is the i-th element of the width vector of the j-th membership layer neuron at time t+1, θ ij (t+1) represents the feedback connection weight of the i-th element of the j-th regular layer neuron at time t+1, w j(t + 1) is the connection weight between the j-th rule layer neuron and the output neuron at the (t + 1) moment;

[0102] ④ Increase the learning step s by 1. If the step s < N, go back to step ③ for continued training. If s = N, go to step ⑤;

[0103] ⑤ Calculate the performance of the recursive fuzzy neural network according to the error formula. Let E d be the ideal error. If E(t) ≥ E d , go back to step ③ for continued training. If E(t) < E d , stop the adjustment;

[0104] The multi-step prediction formula for vehicle speed is as follows:

[0105]

[0106] Among them, h ∈ (1,..., H) is the prediction step number of multi-step prediction, d is the dimension of the input variable, also called the step number of time delay, is the output of the output neuron h steps in the future, and the value of h is 5.

[0107] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0108] The present invention can collect vehicle operation data and summarize it. It can also set collection parameters by the user according to prediction needs to meet the prediction requirements and effectively improve prediction accuracy. The present invention can also efficiently collect, store, and process the data of the prediction process and result data, and can also implement full-process monitoring based on the vehicle driving state, and describe the entire prediction process using a multi-dimensional space. Improve the accuracy and efficiency of prediction, control the fuel performance of the vehicle, so as to achieve the accuracy and scientificity of vehicle speed prediction.

[0109] The following is an embodiment of the E-Power architecture commercial vehicle speed prediction system provided by the present disclosure. This system and the E-Power architecture commercial vehicle speed prediction method in the above embodiments belong to the same inventive concept. For the details not described in detail in the embodiment of the E-Power architecture commercial vehicle speed prediction system, reference can be made to the embodiment of the E-Power architecture commercial vehicle speed prediction method.

[0110] The system includes: a data acquisition module, a data preprocessing module, and a vehicle speed prediction module;

[0111] The data acquisition module is connected to the vehicle's vehicle controller through signal hard wires and CAN lines to obtain vehicle operation data;

[0112] The data preprocessing module is used to preprocess the collected vehicle operation data and to extract features using the PCA method;

[0113] The vehicle speed prediction module is used to establish a vehicle speed prediction model and obtain vehicle speed information based on multi-step prediction using a recursive fuzzy neural network.

[0114] The system of this invention collects driver intention data and vehicle operation data through hard wiring and CAN messages; it uses principal component analysis to extract relevant variables to obtain the input of the prediction model, thereby improving computational efficiency; finally, it establishes a vehicle speed prediction model based on a recursive fuzzy neural network, which can accurately predict vehicle speed during operation, thereby improving fuel economy and driving comfort.

[0115] The units and algorithm steps of the various examples described in the embodiments of the E-Power architecture commercial vehicle speed prediction method and system provided by this invention can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0116] The flowcharts and block diagrams of the E-Power architecture commercial vehicle speed prediction method illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in the flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the figures. For example, two consecutively represented blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0117] In embodiments of the present invention, computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or power server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (exemplarily using an Internet service provider for Internet connection).

[0118] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting vehicle speed in commercial vehicles using an E-Power architecture, characterized in that the method... include: S101. Collect vehicle operation data; S102. Preprocess the collected vehicle operation data and extract features using the PCA method; S103. Establish a vehicle speed prediction model and obtain vehicle speed information based on multi-step prediction using a recursive fuzzy neural network; In step S101, the vehicle operating data includes accelerator pedal opening, brake pedal opening, accelerator pedal opening change rate, brake pedal opening change rate, wheel speed, driving mode, gear status, clutch status, remaining battery charge, battery temperature, battery current, battery voltage, engine speed, engine coolant temperature, generator speed, and generator torque. In step S102, a PCA method model is established, and the obtained 16 influence factors are taken as independent variables x, x=[ x 1 ,x 2 ,…， x 16 ] and the vehicle speed is taken as dependent variable y; wherein, x 1 ,x 2 ,…，x 16 respectively represent an accelerator pedal opening degree, a brake pedal opening degree, an accelerator pedal opening degree change rate, a brake pedal opening degree change rate, a wheel speed, a driving mode, a gear state, a clutch state, a battery remaining capacity, a battery temperature, a battery current, a battery voltage, an engine speed, an engine water temperature, a generator speed, and a generator torque; In step S103, the vehicle speed is obtained by multi-step prediction from a recurrent fuzzy neural network. The recurrent fuzzy neural network has a structure of four layers: an input layer, a membership function layer, a rule layer, and an output layer. The structure is 7- M - M -1 connection method, where M It is a positive integer, and 2 < M< 30, the connection weight between the input layer and the membership function layer is 1, and the actual output of the recurrent fuzzy neural network is represented as: y ( t The vehicle speed prediction method based on recursive fuzzy neural networks calculates as follows: Where x( t )=[ x 1( t ), x 2( t ), x 3( t ), x 4( t ), x 5( t ), x 6( t ), x 7( t )]express t The output of the input layer at any time, x 1( t )express t The value of the accelerator pedal opening at any time x 2( t )express t The value of the brake pedal opening at any time x 3( t )express t The value of the rate of change of accelerator pedal opening at any given time. x 4( t )express t The value of the rate of change of brake pedal opening at any given time. x 5( t )express t The value of the remaining battery power at any time, x 6( t )express t The value of generator speed at any time x 7( t )express t The value of the generator torque at any given time. f Indicate x( t )and y ( t The corresponding functional relationship between ) w j ( t )express t Time Rule Layer j The weights between each neuron and the output neuron ß ij ( t )express t Time input layer i The nth neuron and its membership function layer j The weights between neurons ß ij ( t )=1, m ij ( t ) indicates the membership function layer t Time of the first j The first neuron center value of the i elements, σ ij ( t ) indicates the membership function layer t Time of the first j The first neuron width value i The recurrent fuzzy neural network adds self-feedback connections to the membership function layer. θ ij ( t ) is the feedback weight of the self-feedback connection of the membership function layer, O 2 ij ( t -1) is the square of the feedback value of the self-feedback connection of the membership function layer.

2. The E-Power architecture commercial vehicle speed prediction method according to claim 1, characterized in that, Feature extraction is performed using the following formula based on the PCA method: Among them, F A It is the covariance matrix, Λ is the value on the diagonal of the correlation eigenvalue matrix, and U is... β The principal component matrix of the eigenvectors, x i Representing the i One input variable, μ It is the mean of the input variables. n It is the number of variables; in, ρ i It is the first i The contribution rate of each variable λ i It is the first i The characteristic values ​​of each variable.

3. The E-Power architecture commercial vehicle speed prediction method according to claim 1, characterized in that, in, ß ij ( t -1) indicates t -1 time input layer i The nth neuron and its membership function layer j The weights between neurons ß ij ( t -1)=1, m ij ( t -1) indicates the membership function layer. t -1 moment j The first neuron center value of the i elements, σ ij ( t -1) indicates the membership function layer. t -1 moment j The first neuron width value i One element, θ ij ( t -1) represents the feedback weight of the self-feedback connection of the membership function layer. It is the square of the feedback value of the self-feedback connection of the membership function layer; the error of the recursive fuzzy neural network is defined as: in, m It is the number of samples. y d ( t )yes t The expected output of the time-varying recursive fuzzy neural network. y ( t )yes t The actual output of the time-recursive fuzzy neural network.

4. The E-Power architecture commercial vehicle speed prediction method according to claim 1, characterized in that, The model calibration process is as follows: ① Given the initial number of hidden layer neurons in a recursive fuzzy neural network as... M , M For positive integers greater than 2, the inputs to the recurrent fuzzy neural network are x(1), x(2), ..., x( t ), ..., x( m The corresponding expected output is y d (1), y d (2), …, y d ( t ), ..., y d ( m ), m The dataset is used as training samples for the recursive fuzzy neural network model, and the expected error value is set to... E d , E d ∈(0, 0.01), initial center value m j In (1), the assignment range for each variable is [-2, 2], m j (1)=( m 1j (1), m 2j (1), … , m ij (1)), m ij (1) indicates the membership function layer at the initial time. j The first neuron center vector of the _ ... i One element; Initial center width σ j (1) has an assignment interval of [0,1], σ j (1)=( σ 1 j (1), σ 2 j (1), … , σij (1)), σ ij (1) indicates the membership function layer at the initial time. j The first neuron width value vector i One element; Initial feedback connection weights θ ij ( t The assignment range for ) is [0,1]. j =1, 2, ..., M ; The initial weights w(1) are assigned values ​​in the range [-1, 1], w(1) = ( w 1(1), w 2(1), ..., w j (1)), w j (1) Represents the initial time of the rule layer. j The connection weights between each neuron and the output layer; ② Set the number of learning steps s =1; ③ t = s The output of the recursive fuzzy neural network is calculated according to the formula. y ( t The parameters of the recurrent fuzzy neural network are adjusted using the gradient descent algorithm as follows: in, η m Center vector m ij learning rate, η m ∈(0, 0.01]; η σ Width σ j learning rate, η σ ∈(0, 0.01], η θ For feedback connection weights θ ij The learning rate η θ ∈(0, 0.02], η w For connection weights w j learning rate, η w ∈(0, 0.01], m ij ( t +1) is t +1 time belongs to the membership function layer j The first neuron center vector of the _ ... i One element, σ ij ( t +1) is t +1 moment j The width vector of the nth member function layer neuron i One element, θ ij ( t +1) is t +1 moment j The first regular layer neuron i The feedback connection weights of each element. w j ( t +1) is t +1 moment j The connection weights between each rule layer neuron and the output neuron; y d ( t )yes t The expected output of the time-varying recursive fuzzy neural network. y ( t )yes t The actual output of the time-varying recursive fuzzy neural network; y ( t- 1) is t The actual output of the recurrent fuzzy neural network at time -1; Indicates the membership function layer t Time of the first j The first neuron width value i The square of each element; Indicates the membership function layer t Time of the first j The first neuron width value i The cube of the elements; It is a membership function layer t Feedback values ​​from the constantly self-feedback connection; ④ Learning Steps s Increase by 1 if the number of steps s < N Then proceed to step ③ for further training. s = N Turning step ⑤; ⑤ Calculate the performance of the recurrent fuzzy neural network according to the error formula, assuming... E d For ideal error, if E ( t )≥ E d Then proceed to step ③ for further training. E ( t )< E d If so, then stop adjusting; The multi-step vehicle speed prediction formula is as follows: in, h ∈(1,…, H () represents the number of prediction steps in multi-step prediction. d It refers to the dimension of the input variable, also known as the number of delay steps. It is the future h The output of the output neuron after the step. h The value is 5.

5. A vehicle speed prediction system for commercial vehicles based on the E-Power architecture, characterized in that, The system adopts the E-Power architecture commercial vehicle speed prediction method as described in any one of claims 1 to 4; The system includes: a data acquisition module, a data preprocessing module, and a vehicle speed prediction module; The data acquisition module is connected to the vehicle's controller via signal hardwire and CAN bus to acquire vehicle operating data; The data preprocessing module is used to preprocess the collected vehicle operation data and to extract features using the PCA method; The vehicle speed prediction module is used to establish a vehicle speed prediction model and obtain vehicle speed information based on multi-step prediction using a recursive fuzzy neural network.

Images