A real-time power distribution method for an electric hybrid tractor generator and power battery

Abstract

The application provides a real-time power distribution method for an electric hybrid tractor generator and a power battery, and belongs to the technical field of electric drive control in agricultural machinery and vehicle engineering. The application is used to solve the problems in the prior art that an offline global optimization result is difficult to be directly applied online, a system working mode is not reasonably switched under a complex working condition, and target power generation accuracy of a range extender is insufficient. The steps of the application comprise the following steps: S1, constructing an offline global optimal control database; S2, establishing a working condition characteristic space; S3, online working condition identification and coordinate representation; S4, establishing a system mode selector; S5, establishing a range extender target power prediction model; S6, NARX network offline training and online closed-loop prediction; and S7, outputting a range extender target power instruction: engine-generator set output power, power battery output power, and demand power minus generator output power.

Description

Technical Field

[0001] This invention relates to a real-time power distribution method for a generator and a power battery in an electric hybrid tractor, belonging to the field of electric drive control technology in agricultural machinery and vehicle engineering. Background Technology

[0002] As modern agriculture develops towards low-carbon and intelligent directions, the shortcomings of traditional diesel tractors in terms of fuel economy, emission levels, and adaptability to complex operating conditions are becoming increasingly prominent. Range-extended electric tractors combine the advantages of fast response and high control precision of electric drive systems with the strong continuous power supply capability of engines, making them an important technical solution for medium-to-high load and long-term continuous operation scenarios.

[0003] For range-extended electric tractors, the core task of energy management strategy is to rationally coordinate the power distribution between the range extender and the power battery while meeting the needs of overall machine drive and workload, so that the engine can operate in the highest efficiency range as much as possible, and to balance power, economy, and power supply stability. Existing energy management methods mainly include rule-based methods, optimization-based methods, and learning-based methods. Rule-based methods are simple to implement, but they are highly dependent on experience and parameter calibration, making it difficult to guarantee optimality across the entire operating range. Dynamic programming-based global optimization methods can obtain the optimal control law under a given operating condition, but they rely on complete prior operating condition information, have high computational complexity, and are difficult to use directly for online control.

[0004] While existing learning-based energy management methods have improved adaptability under complex operating conditions to some extent, they still have the following shortcomings in the field of agricultural tractors: First, the ability to perceive operating conditions under complex operating conditions is insufficient, and online control usually remains at the level of responding to instantaneous state variables; Second, there is a lack of effective bridging between offline optimal rules and online control, making it difficult for learning strategies to fully inherit global optimization results; Third, the target power under the combined effects of operating load, drive load, and power battery status has obvious nonlinearity and temporal correlation, and it is difficult to achieve high-precision real-time prediction using only static mapping models. Summary of the Invention

[0005] To address the problems in existing technologies, such as the difficulty in directly applying offline global optimization results online, unreasonable switching of system operating modes under complex working conditions, and insufficient accuracy in generating target power for range extenders, this invention proposes a real-time power allocation method for the generator and power battery of an electric hybrid tractor.

[0006] The technical solution adopted by the present invention to solve the above problems is as follows: The present invention includes the following steps: Step 1: Construct an offline global optimal control database based on the whole vehicle model of the range-extended electric tractor; Step 2: Extract state feature parameters from the offline global optimal control database and establish a work condition feature space; Step 3: Perform online working condition identification and coordinate representation within the working condition feature space, and output the coordinate representation of the current working condition in the clustering feature space. ; Step 4: Based on the coordinate representation and the current vehicle operating status, select the system operating mode and obtain the current optimal system operating mode; Step 5: Based on the obtained current optimal system operating mode, call the corresponding range extender target power prediction model to obtain the current target power prediction value of the range extender. Step 6: Based on the predicted target power value of the range extender, output the target power command of the range extender: the output power of the engine-generator set, and the output power of the power battery equals the required power minus the generator output power.

[0007] Furthermore, step 1 specifically includes: Based on the whole vehicle model of the range-extended electric tractor, and according to different representative working conditions and different initial power battery charge states, the dynamic programming method is used to solve offline to obtain the optimal system working mode sequence and optimal energy flow allocation results under the corresponding working conditions, forming an offline global optimal control database. The offline global optimal control database includes at least the system working mode label, the range extender target power trajectory, the vehicle operating status, the power battery status, and the work load information.

[0008] Furthermore, step 2 specifically includes: State characteristic parameters are extracted from the work samples. These parameters include vehicle speed, acceleration, drive power, power output shaft power, and overall power demand. Correlation analysis and principal component analysis were performed on the state characteristic parameters to complete feature selection and dimensionality reduction. Cluster analysis was performed on the dimensionality-reduced samples to establish the operational condition feature space and cluster centers.

[0009] Furthermore, step 3 specifically includes: Step 3.1: Call the offline modeling parameters, combine the current sliding window data to extract the current window statistical features, and perform standardization and PCA projection according to the offline parameters to obtain the features. ; Step 3.2: Calculate features Distance features to cluster centers of all samples ; Step 3.3: Determine the current working condition category according to the principle of minimum distance. With operating condition characterization; Step 3.4: Determine if the batch update conditions are met. If not, repeat steps 3.1-3.3 until the batch update conditions are met; if they are met, proceed to step 3.5. Step 3.5: Update the current batch sample set X. Use the MiniBatch K-means algorithm to perform small-batch incremental updates on the cluster centers to obtain temporary cluster centers. ; Step 3.6: Correct the temporary cluster centers according to the weighted update strategy, and output the coordinate representation of the current working condition in the cluster feature space. The coordinate representation of the current working condition in the clustering feature space. This includes updated cluster centers, current operating condition categories, and operating condition representations; The expression for the minimum distance principle is: (1).

[0010] Furthermore, step 4 specifically includes: Determining the required power of the entire vehicle based on a model of a range-extended electric tractor Speed and the state of charge of the power battery The power required for the whole vehicle Speed and the state of charge of the power battery and working condition clustering coordinate representation As input to the supervised classification model ; The optimal system operating mode from the offline global optimal control database is used as the output label to construct a system mode selection sample set. , ,in, M This represents the total number of system operating modes. Based on the system pattern selection sample set, a system pattern selector is constructed using the random forest algorithm, and the optimal system operating mode at the current moment is obtained through ensemble voting of multiple decision trees. .

[0011] Furthermore, step 5 specifically includes: For each type of system operating mode, a corresponding nonlinear autoregressive neural network (NARX) sub-model with external input is constructed; Representing the current working condition in the cluster feature space using coordinates. Drive power Power output shaft power Vehicle power requirements Vehicle speed and state of charge of the power battery As an exogenous input vector; The optimal range extender target power is obtained from the offline global optimal control database. The NARX sub-model is trained using the network regression output labels. Based on the current optimal system operating mode, the corresponding NARX sub-model is invoked to output the predicted target power value of the range extender at the current moment.

[0012] Furthermore, the NARX sub-model employs an open-loop training approach during the training phase, using real historical outputs as feedback inputs; and a closed-loop rolling prediction approach is used during the online deployment phase, taking the network's predicted output from the previous time step as input. As the feedback value at the current moment, it enables real-time recursive prediction; Network Predicted Output The expression is: (2); In formula (2), For the network's predicted output, For exogenous input vectors, and These are the output feedback order and the input delay order, respectively.

[0013] Furthermore, step 6 specifically includes: The predicted target power of the range extender is used as the input command for the range extender energy management controller and sent to the lower-level controller of the range extender to control the working status and output power of the engine-generator set. The output power of the power battery is equal to the demand power minus the generator output power.

[0014] The beneficial effects of this invention are: 1. This invention achieves real-time characterization of complex agricultural operating conditions through sliding time windows, cluster modeling, and online incremental updates, thereby improving the ability to identify online operating conditions.

[0015] 2. This invention constructs a supervised classification database with working condition coordinate representation and uses random forest to achieve system working mode selection, which can improve the accuracy and robustness of system mode switching under complex working conditions. In this embodiment, simulation results show that the recognition accuracy of mode 1, mode 2 and mode 3 reaches 98.62%, 98.81% and 97.78% respectively, and the overall accuracy reaches 98.50%.

[0016] 3. This invention constructs supervised samples based on the global optimization results of offline dynamic programming, and uses a NARX neural network to learn the optimal range extender target power trajectory under different system operating modes, thereby realizing the mapping of offline optimal control law to online executable control strategy.

[0017] 4. Under known operating conditions, the real-time energy management method of the present invention reduces the overall energy consumption by 4.1% compared with the power-following PFCS strategy; under unknown operating conditions, it reduces the overall energy consumption by 3.1% compared with the PFCS rule strategy, indicating that the present invention can improve the economy of energy management while ensuring real-time performance.

[0018] 5. Under combined operating conditions, the fuel consumption of PFCS and K-means-RF-NARX strategies is 508g and 490g, respectively. The method of the present invention reduces fuel consumption by 3.5% compared with PFCS, indicating that the present invention can maintain a better energy management effect than the rule-based strategy under different operating scenarios. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a real-time power distribution method for a hybrid electric tractor generator and power battery; Figure 2 A flowchart for online working condition identification and coordinate representation; Figure 3 Flowchart for offline training of the random forest pattern selector; Figure 4 Flowchart of online invocation of the random forest pattern selector; Figure 5 This is a schematic diagram of the NARX neural network structure; Figure 6 Flowchart for offline training of NARX sub-models; Figure 7 Flowchart of online closed-loop prediction for NARX sub-model; Figure 8 This is a schematic diagram of the energy flow structure of a range-extended electric tractor. Figure 9 A schematic diagram of a forward simulation model of a range-extended electric tractor; Figure 10 A schematic diagram illustrating power allocation decisions for energy management strategies; Figure 11 This is a schematic diagram of the known operating conditions; Figure 12 This is a schematic diagram of an unknown operating condition. Figure 13 This is a comparison chart of the fuel consumption of the present invention with that of the other two strategies under known operating conditions; Figure 14 A comparison chart of fuel consumption of the present invention with two other strategies under unknown operating conditions; Figure 15 A comparison chart showing the fuel consumption of this invention with two other strategies under different operating conditions. Detailed Implementation like Figure 1As shown, the real-time power distribution method for a hybrid tractor generator and power battery described in this embodiment includes the following steps: S1: Construct an offline global optimal control database; This invention is based on a range-extended electric tractor vehicle model. Under different representative working conditions and different initial power battery charge conditions, the dynamic programming method is used for offline solution to obtain the optimal system working mode sequence and optimal energy flow allocation results under the corresponding working conditions, forming an offline optimal control database for subsequent supervised learning. The database includes at least system working mode labels and range extender target power trajectory.

[0020] S2: Establish the characteristic space of the working conditions; State characteristic parameters are extracted from tractor operation samples from the offline optimal control database. Cluster analysis is performed on the operation samples to establish an operation condition feature space and cluster centers. The characteristic parameters include vehicle speed, acceleration, drive power, power output shaft power, and comprehensive demand power. Correlation analysis and principal component analysis are used to complete feature screening and dimensionality reduction.

[0021] S3: Online operating condition identification and coordinate representation; like Figure 2 As shown, during the online operation phase, a sliding time window is used to extract the current operation features, and the operation condition category is identified based on the offline-saved cluster centers. Furthermore, the MiniBatch K-means algorithm is used to perform small-batch incremental updates on the cluster centers to adapt to the dynamic changes in the distribution of unknown operation conditions, outputting the coordinate representation of the current operation condition in the cluster feature space. .

[0022] S4: Establish system mode selector; like Figure 3 As shown, the working condition cluster coordinates are represented. Vehicle power requirements Speed and the state of charge of the power battery As input to the supervised classification model, the optimal system operating mode obtained from offline dynamic programming is used as the output label to construct a system mode selection sample set; its input features are represented as follows: (1); like Figure 4 As shown, the output labels are represented as follows: (2); In formula (2), This represents the total number of system operating modes.

[0023] The system operating modes are defined by multiple operating mode labels derived from offline dynamic programming results. These labels characterize the typical collaborative power supply states of the range extender and power battery under different operating conditions and power demands. Based on the aforementioned sample set, a system mode selector is constructed using a random forest algorithm. The optimal system operating mode for the current moment is obtained through ensemble voting of multiple decision trees. .

[0024] S5: Establish a target power prediction model for the range extender; like Figure 5 As shown, after obtaining the system's operating modes, corresponding NARX sub-models are constructed for each type of system operating mode. For the first... The exogenous input vector of a system is defined as follows: (3); In formula (3), For drive power, For power output shaft power, This refers to the state of charge of the power battery.

[0025] The optimal target power of the range extender is obtained by solving offline dynamic programming. As the output label of the network regression, that is: (4); This allows us to learn the dynamic nonlinear mapping relationship between the current operating condition, vehicle operating status, and the offline optimal range extender target power.

[0026] S6: Offline training and online loop closure prediction of NARX network; like Figure 6 As shown, the NARX neural network simultaneously incorporates exogenous input variables and historical output feedback information in its current output prediction. Its general form is as follows: (5); In formula (5), For the network's predicted output, For exogenous input vectors, and These are the output feedback order and the input delay order, respectively.

[0027] The training phase adopts an open-loop training method, using real historical outputs as feedback inputs; the online deployment phase adopts a closed-loop rolling prediction method, using the network prediction output of the previous moment as the feedback quantity of the current moment, to achieve real-time recursive prediction.

[0028] S7: Output the target power command for the range extender; like Figure 7 As shown, the system operating mode obtained from S4 The system can call the NARX sub-model trained in the corresponding mode online and output the target power of the range extender at the current moment. The power output of the battery is equal to the required power minus the generator output power, and this target power is used as the input command for the range extender energy management controller for subsequent execution layer control.

[0029] Example like Figure 8 As shown, this embodiment uses a range-extended electric tractor as the controlled object. The range-extended electric tractor includes a vehicle controller, a power battery system, a range extender system, a drive motor system, and a power output shaft motor system. The range extender system includes an engine and a generator, and the power battery system is used to supply energy to the drive motor system and the power output shaft motor system together with the range extender.

[0030] like Figure 9 As shown, this embodiment uses a forward simulation model of the entire range-extended electric tractor to perform offline simulations on multiple representative operating conditions. These operating conditions include at least known samples and can be further extended to unknown samples generated by a stochastic process. For each operating condition, under multiple initial battery state-of-charge conditions, a dynamic programming method is used to solve for the optimal energy management result, obtaining the optimal system operating mode sequence and the optimal range extender target power trajectory for the corresponding operating condition.

[0031] The real-time energy management method is deployed in the vehicle controller and adopts a hierarchical structure, including a condition identification layer, a mode selection layer, and a power prediction layer. The condition identification layer is used to extract and characterize the features of the current operating scenario; the mode selection layer is used to determine the optimal system operating mode based on the condition characterization results and the system status; and the power prediction layer is used to generate the target power command for the range extender under the system operating mode.

[0032] The above offline optimal results are aligned with the operating state variables of the corresponding operating conditions to form an offline global optimal control database. The database includes at least the following: the current operating condition feature vector, the vehicle's required power, vehicle speed, the power battery's state of charge, the operating mode label of the offline dynamic programming optimal system, and the target power of the offline dynamic programming optimal range extender.

[0033] In this embodiment, a sliding time window method is used to extract features from the original job data. Let the length of the sliding window be... The step size is 1 sampling time. Within each time window, statistical features are extracted from variables such as vehicle speed, acceleration, drive power, power output shaft power and comprehensive demand power to form the original working condition feature set.

[0034] Then, correlation analysis and principal component analysis are performed on the original working condition feature set to remove redundant information and complete feature dimensionality reduction. K-means++ algorithm is used to perform clustering modeling on the dimensionality-reduced offline samples to obtain the working condition categories and their initial cluster centers, and the standardized parameters, dimensionality reduction matrix, and initial cluster centers are saved.

[0035] During the online operation phase, the vehicle controller acquires real-time vehicle operating status and workload data, and extracts current-time operating condition features based on a sliding time window. These features are then mapped using standardized parameters and a dimensionality reduction matrix stored in the offline phase, and input into the MiniBatch K-means algorithm to incrementally update the cluster centers, outputting the current operating condition category. Coordinate representation of the current working condition in the cluster feature space The coordinate representation serves as a common input for subsequent mode selection and power prediction.

[0036] In this embodiment, the vehicle controller represents the operating condition cluster coordinates. Vehicle power requirements Speed and the state of charge of the power battery The input features of the mode selector are used to invoke the trained random forest classifier, which outputs the system operating mode at the current sampling time. Then, based on the system's operating mode, the corresponding NARX sub-model is selected to generate the range extender's target power.

[0037] For different system operating modes, corresponding NARX sub-models are established. For the first... The exogenous input vector of a system is defined as follows: (6); In formula (6), For drive power, The output shaft power is the target power of the range extender, obtained by offline dynamic programming. As labels for network training output.

[0038] The general expression for the output of the NARX neural network is: (7); In formula (7), For the network's predicted output, For exogenous input vectors, To output the feedback order, The input delay order is used. During the training phase, an open-loop training method is employed, using real historical outputs as feedback inputs. During the online deployment phase, a closed-loop rolling prediction method is used, employing the network's prediction output from the previous time step as the feedback quantity for the current time step, thereby achieving real-time recursive prediction.

[0039] During online control, the vehicle controller first outputs the current system operating mode based on the random forest mode selector. Then, the NARX sub-model in the corresponding mode is called to output the target power of the range extender at the current moment. .

[0040] like Figure 10 As shown, the online execution process of the energy management strategy for the range-extended electric tractor in this embodiment is as follows: Step 1: Collect information on the vehicle's operating status, power battery status, and operating load at the current moment; Step 2: Extract working condition features using a sliding time window, and output the current working condition category and working condition cluster coordinate representation. ; Step 3: Characterize the operating conditions Vehicle power requirements Speed and the state of charge of the power battery Input the random forest mode selector to obtain the current system operating mode. ; Step 4: Based on the system operating mode, call the corresponding NARX sub-model and output the target power of the range extender. ; Step 5: Set the target power of the range extender The signal is sent to the lower-level controller of the range extender to control the working status and output power of the engine-generator set. The output power of the power battery is equal to the required power minus the generator output power.

[0041] Through the above process, the optimal control law of offline dynamic programming can be approximated online, thereby improving the energy management economy of the whole vehicle while ensuring real-time performance.

[0042] In addition, to verify the technical effects of the present invention, the following simulation verification was conducted: To verify the effectiveness of the method described in this embodiment, a simulation model of a range-extended electric tractor was built on the MATLAB / Simulink platform, and tests were conducted using both known and unknown operating condition samples. For example... Figure 11 As shown. The known working condition sample is a set of working conditions on hilly and undulating terrain over a period of 3600 seconds, such as... Figure 12 As shown, the unknown working condition sample is a set of 4750s working conditions generated using a Markov chain based on the known working conditions.

[0043] like Figure 13 and Figure 14 As shown, the K-means-RF-NARX real-time energy management method described in this embodiment is compared with the PFCS rule strategy. Figure 13 and Figure 14 These are samples with known operating conditions and samples with unknown operating conditions. Simulation results show that: under known operating conditions, the method in this embodiment reduces overall energy consumption by 4.1% compared to the PFCS rule-based strategy; under unknown operating conditions, it reduces overall energy consumption by 3.1% compared to the PFCS rule-based strategy; under combined operating conditions, the fuel consumption of the PFCS and K-means-RF-NARX strategies are 508g and 490g, respectively, and the method in this embodiment reduces overall energy consumption by 3.5% compared to the PFCS rule-based strategy.

[0044] like Figure 15 As shown above, the results demonstrate that the method described in this embodiment can maintain energy management performance superior to rule-based strategies and close to offline global optimal strategies under different operating scenarios, thereby verifying the real-time performance, economy, and adaptability of the present invention under complex agricultural operating conditions.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A method for real-time power distribution between a generator and a power battery in an electric hybrid tractor, characterized in that, include: Step 1: Construct an offline global optimal control database based on the whole vehicle model of the range-extended electric tractor; Step 2: Extract state feature parameters from the offline global optimal control database and establish a work condition feature space; Step 3: Perform online working condition identification and coordinate representation within the working condition feature space, and output the coordinate representation of the current working condition in the clustering feature space. ; Step 4: Based on the coordinate representation and the current vehicle operating status, select the system operating mode and obtain the current optimal system operating mode; Step 5: Based on the current optimal system operating mode, call the corresponding range extender target power prediction model to obtain the current target power prediction value of the range extender; Step 6: Based on the predicted target power value of the range extender, output the target power command of the range extender: the output power of the engine-generator set, and the output power of the power battery equals the required power minus the generator output power.

2. The real-time power distribution method for a hybrid electric tractor generator and power battery according to claim 1, characterized in that, Step 1 specifically includes: Based on the whole vehicle model of the range-extended electric tractor, and according to different representative working conditions and different initial power battery charge states, the dynamic programming method is used to solve offline to obtain the optimal system working mode sequence and optimal energy flow allocation results under the corresponding working conditions, forming an offline global optimal control database. The offline global optimal control database includes at least the system working mode label, the range extender target power trajectory, the vehicle operating status, the power battery status, and the work load information.

3. The real-time power distribution method for a hybrid electric tractor generator and power battery according to claim 1, characterized in that, Step 2 specifically includes: State feature parameters are extracted from the work samples. These state feature parameters include vehicle speed, acceleration, drive power, power output shaft power, and comprehensive power demand. Correlation analysis and principal component analysis were performed on the state characteristic parameters to complete feature selection and dimensionality reduction. Cluster analysis was performed on the dimensionality-reduced samples to establish the operational condition feature space and cluster centers.

4. The real-time power distribution method for a hybrid electric tractor generator and power battery according to claim 3, characterized in that, Step 3 specifically includes: Step 3.1: Call the offline modeling parameters, combine the current sliding window data to extract the current window statistical features, and perform standardization and PCA projection according to the offline parameters to obtain the features. ; Step 3.2: Calculate features Distance features to cluster centers of all samples ; Step 3.3: Determine the current working condition category according to the principle of minimum distance. With operating condition characterization; Step 3.4: Determine if the batch update conditions are met. If not, repeat steps 3.1-3.3 until the batch update conditions are met; if they are met, proceed to step 3.

5. Step 3.5: Update the current batch sample set X. Use the MiniBatch K-means algorithm to perform small-batch incremental updates on the cluster centers to obtain temporary cluster centers. ; Step 3.6: Correct the temporary cluster centers according to the weighted update strategy, and output the coordinate representation of the current working condition in the cluster feature space. The coordinate representation of the current working condition in the clustering feature space. This includes updated cluster centers, current operating condition categories, and operating condition representations; The expression for the minimum distance principle is: （1）。 5. The real-time power distribution method for a hybrid tractor generator and power battery according to claim 1, characterized in that, Step 4 specifically includes: Determining the required power of the entire vehicle based on a model of a range-extended electric tractor Speed and the state of charge of the power battery The power required for the whole vehicle Speed and the state of charge of the power battery and working condition clustering coordinate representation As input to the supervised classification model ; The optimal system operating mode in the offline global optimal control database is used as the output label to construct a system mode selection sample set. , ,in, M This represents the total number of system operating modes. Based on the system mode selection sample set, a system mode selector is constructed using the random forest algorithm, and the optimal system operating mode at the current moment is obtained through ensemble voting of multiple decision trees. .

6. The real-time power distribution method for a hybrid electric tractor generator and power battery according to claim 1, characterized in that, Step 5 specifically includes: For each type of system operating mode, a corresponding nonlinear autoregressive neural network (NARX) sub-model with external input is constructed; Representing the current working condition in the cluster feature space using coordinates. Drive power Power output shaft power Vehicle power requirements Vehicle speed and state of charge of the power battery As an exogenous input vector; The optimal range extender target power is obtained from the offline global optimal control database. The NARX sub-model is trained using the network regression output labels. Based on the current optimal system operating mode, the corresponding NARX sub-model is invoked to output the predicted target power value of the range extender at the current moment.

7. The real-time power distribution method for a hybrid electric tractor generator and power battery according to claim 6, characterized in that, The training phase of the NARX sub-model adopts an open-loop training method, using real historical outputs as feedback inputs; During the online deployment phase, a closed-loop rolling prediction method is adopted, which takes the network prediction output from the previous moment as an example. As the feedback value at the current moment, it enables real-time recursive prediction; Network Predicted Output The expression is: （2）； In formula (2), For the network's predicted output, For exogenous input vectors, and These are the output feedback order and the input delay order, respectively.

8. The real-time power distribution method for a hybrid electric tractor generator and power battery according to claim 1, characterized in that, Step 6 specifically includes: The predicted target power of the range extender is used as the input command for the range extender energy management controller and sent to the lower-level controller of the range extender to control the working status and output power of the engine-generator set. The output power of the power battery is equal to the demand power minus the generator output power.

Images