New energy automobile intelligent energy management strategy offline optimization and online implementation method and system

Abstract

The invention belongs to the technical field of new energy vehicle control, and discloses a new energy vehicle intelligent energy management strategy offline optimization and online implementation method and system, and the method comprises the steps: carrying out the offline processing of vehicle real-time driving data, and building a driving condition basic database and a driving style basic information database; based on the vehicle daily driving data, the driving condition basic database and the driving style basic information base, performing rolling optimization on the control parameters related to the driving condition characteristics; and estimating the road gradient in real time, and correcting the control parameters after rolling optimization to realize intelligent energy management of the new energy automobile. According to the method, the rolling optimization of the control parameters along with the long-period operation data and the dynamic adjustment of the multi-mode driving data are realized through the process of ''offline processing + rolling optimization + online implementation'', so that the intelligent energy management of the new energy automobile is ensured.

Description

Technical Field

[0001] This invention belongs to the field of new energy vehicle control technology, specifically relating to an offline optimization and online implementation method and system for intelligent energy management strategies of new energy vehicles. Background Technology

[0002] Energy management strategies, as a crucial control method for the efficient operation of new energy vehicle power systems and the conservation of overall vehicle energy consumption, are a core technology and a key challenge in the development of vehicle control systems. Their performance directly impacts the overall performance of the vehicle. Currently, energy management strategies are mainly classified into three categories: rule-based, global optimization-based, and instantaneous optimization-based. Rule-based strategies are simple to design, easy to implement, and have good real-time performance, making them the most widely used. However, the determination of their control rules and switching thresholds relies on engineering experience, resulting in poor adaptability to changes in operating conditions and an inability to achieve optimal energy management. Global optimization-based strategies, on the other hand, can use dynamic programming algorithms to find the decision result that minimizes the vehicle's energy consumption when operating condition information is known, achieving true global optimal control. However, the solution process is too complex, with a large computational load, and the optimization results cannot be directly used for real-time vehicle control. Even if neural network-based or rule-based strategies can be designed based on the optimal results and deployed on the vehicle controller, the design of this type of strategy still has a large dependence on road condition information. Instantaneous optimization-based strategies have better optimization performance than rule-based strategies and stronger real-time performance and dynamic adaptability than global optimization strategies. They can also achieve near-global optimal performance through optimization design, making them potential applications in real-world vehicle controllers.

[0003] Current widely used energy management strategies typically rely on power demand information from standard driving cycles for optimization. However, in real-world scenarios, driving conditions are diverse and random, and vehicle power demand is influenced by driving style and road information. This makes it difficult for energy management strategies based on a single driving cycle to achieve good control performance under complex road conditions. Although some studies have proposed using V2X and technologies such as Geographic Information Systems (GIS), Intelligent Transportation Systems (ITS), and Global Positioning Systems (GPS) to obtain real-time vehicle driving data for predicting short-term driving conditions and designing corresponding control parameter calculation methods, these strategies mostly rely on the limited computing power and data capacity of onboard computing platforms. They fail to fully utilize cloud-based big data platforms and long-term data recording, making it difficult to dynamically optimize vehicle lifecycle control based on the long-term operating habits of different vehicle models. Furthermore, these strategies fail to comprehensively utilize multimodal driving information that significantly affects vehicle energy consumption and power demand characteristics during implementation. Summary of the Invention

[0004] To address the problem that existing energy management strategies do not fully utilize multimodal driving data and cannot achieve optimal control through rolling updates based on long-term operational big data, this invention provides an offline optimization and online implementation method and system for intelligent energy management strategies of new energy vehicles. This method achieves the dynamic adjustment of control parameters based on the rolling optimization of long-term operational data and the dynamic adjustment of multimodal driving data through an "offline processing + rolling optimization + online implementation" process, thereby ensuring intelligent energy management of new energy vehicles.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for offline optimization and online implementation of intelligent energy management strategy for new energy vehicles, the method comprising: Offline processing of real-time vehicle driving data is performed to establish a basic database of driving conditions and a basic information database of driving styles. Based on daily vehicle driving data, a basic database of driving conditions, and a basic information database of driving styles, rolling optimization is performed on control parameters related to driving condition characteristics. Real-time estimation of road gradient and correction of control parameters after rolling optimization enable intelligent energy management of new energy vehicles.

[0006] Preferably, methods for offline processing of real-time vehicle driving data to establish a basic database of driving conditions and a basic database of driving style information include: The raw dataset is obtained by collecting real-time vehicle driving data based on a big data platform; The original dataset is refined and standardized by cropping to obtain segments of various standard working conditions; Extract specific energy consumption characteristic parameters from each clipped standard operating condition segment and perform dimensionality reduction on them; Cluster analysis is performed on standard operating condition segments based on the dimensionality-reduced feature parameters, and representative operating conditions from each type of standard operating condition segment are extracted to form a basic database of driving operating conditions. After designing indicators to mine and cluster driving information, a basic information database of driving styles is established.

[0007] Preferably, the method for rolling optimization of control parameters related to driving condition characteristics based on daily vehicle driving data, a basic database of driving conditions, and a basic database of driving style information includes: Newly collected data from daily vehicle operation is selected for inclusion in the database. When the number of newly added data segments reaches a preset threshold, the working condition database and driving style database are updated, and a control parameter retraining instruction is triggered. Using standard fragments of the operating condition database, control parameters are optimized offline and simulated in real time. A training set of control parameters related to driving condition characteristics is obtained to train the neural network, and fuzzy rules are designed based on the clustering results of driving style. Over-the-air (OTA) updates of neural network parameters and fuzzy rules are performed when the vehicle is charging or stationary.

[0008] Preferably, the method for achieving intelligent energy management of new energy vehicles by real-time estimation of road gradient and correction of the control parameters after rolling optimization includes: Design a neural network controller that takes driving condition characteristics as input to estimate the adaptive control parameters for driving conditions; A driving style identification method and a corresponding driving style control parameter compensation algorithm are established for different operating conditions. The control parameters are then corrected by estimating the road slope in real time, thereby realizing intelligent energy management for new energy vehicles.

[0009] Preferred methods for establishing driving style control parameter compensation algorithms include: ; In the formula, , , These are the control parameter compensation coefficients for three driving styles; E Soft , E Norm , E Agg These represent the expected accelerator pedal opening values ​​for drivers with three different driving styles.

[0010] Preferred methods for real-time estimation of road slope to correct control parameters include: ; In the formula, , , These are the control parameter compensation coefficients for three driving styles; E Soft , E Norm , E Agg These represent the expected accelerator pedal opening values ​​for drivers with three different driving styles. The compensation coefficient for the control parameters of driving style; m For vehicle quality; g It is the acceleration due to gravity; C d , f These are the drag coefficients and rolling resistance coefficients; A For windward area; ρ air density; For the current moment k The vehicle speed; The compensation coefficient is the control parameter for road slope. This is the optimal estimate of the road slope.

[0011] The present invention also provides an offline optimization and online implementation system for intelligent energy management strategies of new energy vehicles. The system is used to implement the aforementioned method and includes: an offline processing layer, a rolling optimization layer, and an online implementation layer. The offline processing layer is used to process real-time vehicle driving data offline and establish a basic database of driving conditions and a basic information database of driving style. The rolling optimization layer is used to perform rolling optimization of control parameters related to driving condition characteristics based on daily vehicle driving data, driving condition database, and driving style information database. The online implementation layer is used to estimate the road gradient in real time and correct the control parameters after rolling optimization, so as to realize intelligent energy management of new energy vehicles.

[0012] Preferably, the offline processing layer includes: a data acquisition unit, a processing unit, an extraction unit, and an analysis unit; The acquisition unit is used to collect real-time vehicle driving data based on a big data platform to obtain a raw dataset; The processing unit is used to perform fine processing and standardized cropping on the original dataset to obtain various standard working condition segments; The extraction unit is used to extract specific energy consumption characteristic parameters from each clipped standard working condition segment and perform dimensionality reduction processing on them. The analysis unit is used to perform cluster analysis on standard operating condition segments based on the dimensionality-reduced feature parameters, and extract representative operating conditions from various standard operating condition segments to form a basic database of driving operating conditions. After designing indicators to mine and cluster driving information, a basic information database of driving styles is established.

[0013] Preferably, the rolling optimization layer includes: a data entry and selection unit, a threshold-triggered update unit, an offline simulation training unit, and an OTA remote deployment unit; The data entry selection unit is used to select newly collected data from daily vehicle operation for entry into the database. The threshold-triggered update unit is used to update the working condition basic database and the driving style basic information database when the newly added data fragment reaches a preset threshold, and to trigger the control parameter retraining instruction. The offline simulation training unit is used to perform offline optimization and real-time simulation of control parameters using standard fragments of the working condition database, obtain a training set of control parameters related to driving condition characteristics to train the neural network, and design fuzzy rules based on driving style clustering results. The OTA remote deployment unit is used to update neural network parameters and fuzzy rules via OTA when the vehicle is charging or stationary.

[0014] Preferably, the online implementation layer includes: an estimation unit and a correction unit; The estimation unit is used to design a neural network controller that takes driving condition characteristics as input to estimate the driving condition adaptive control parameters. The correction unit is used to establish driving style identification methods and corresponding driving style control parameter compensation algorithms under different operating conditions, and to estimate the road slope in real time to correct the control parameters, thereby realizing intelligent energy management of new energy vehicles.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Realize rolling optimization of database and control parameters throughout the entire life cycle: After refining, standardizing and cutting, extracting and reducing the specific energy consumption characteristic parameters, mining and clustering driving information data of vehicle historical operation data, a working condition database and a driving style information database are established. A new data entry selection mechanism is established, and differential value analysis is performed on the daily operation data of vehicles. This allows for rolling updates of the database and information database and retraining and OTA of control parameters as they are updated. This fully utilizes vehicle operation data to perform long-term rolling optimization of energy management strategies.

[0016] (2) Realize dynamic adjustment of control parameters with multimodal driving: Based on the neural network parameters and fuzzy rules updated by OTA, design a neural network controller with driving condition characteristics as input to estimate control parameters, establish driving style identification methods and corresponding control parameter compensation algorithms under different driving conditions, and further correct control parameters by using real-time estimated road slope, and send the optimal control quantity combination corresponding to the lowest energy consumption cost as control command to each component for execution, so as to complete the multi-information collaborative energy management of new energy vehicles. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the 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.

[0018] Figure 1This is a schematic diagram of an offline optimization and online implementation method for an intelligent energy management strategy for new energy vehicles according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a data refinement processing method to improve data quality according to an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Example 1 This invention proposes an offline optimization and online implementation method for intelligent energy management strategies of new energy vehicles. The core of this method lies in the rolling optimization of control parameters based on long-term operational data and dynamic adjustment of multimodal real-time driving data. First, the original vehicle driving dataset is refined and standardized. Feature parameters are extracted from each segment after cropping, and dimensionality reduction and cluster analysis are performed. Representative segments from each segment are extracted to form a basic operating condition database. Then, driving style mining and cluster analysis are conducted to establish a basic driving style information database. Second, a new data entry selection, database update, and control parameter retraining trigger mechanism is designed. After triggering, standard operating condition segments are used for offline optimization of control parameters. A set of control parameters related to operating condition characteristics is obtained to train the neural network. Fuzzy rules are designed based on the driving style clustering results. Under suitable conditions such as vehicle charging or stationary operation, OTA updates the neural network parameters and fuzzy rules to the controller application layer. Finally, a neural network controller with driving operating condition characteristics as input is designed to estimate control parameters. Driving style identification methods and corresponding control parameter compensation algorithms are established for different operating condition categories. Real-time road gradient estimation is used to further correct the control parameters, thereby achieving intelligent energy management for new energy vehicles. like Figure 1 The diagram shows the flow of this method.

[0022] In this embodiment, autonomous driving data of 60 new energy vehicles over a period of 3 months is collected. The travel tasks cover short-distance commuting, medium-to-long-distance travel, low-speed urban driving, and medium-to-high-speed suburban driving. The collected data includes vehicle speed, acceleration, vehicle position, accelerator pedal travel, brake pedal travel, vehicle operating mode, battery SOC, and instantaneous fuel consumption, with a sampling frequency of 1 Hz. This embodiment collects daily driving data from onboard sensors, processes it after processing by the vehicle controller, and forwards it to the vehicle's CAN network. The onboard terminal receives the data and forwards it to the communication base station via 4G / 5G network. The data is then sent to the backend server via the Internet and stored in a relevant database. After offline processing, the data can be used for vehicle driving data and driver behavior analysis, and for optimizing intelligent energy management strategy control parameters. Finally, the control parameters are updated to the vehicle controller application layer via OTA through the onboard terminal.

[0023] (1) Offline processing layer S101. Refined Data Processing. To address issues such as missing data, data noise, data duplication, data mutations, bad data, and invalid data in the original collected data, this embodiment provides a refined data processing method to improve data quality, such as... Figure 2 As shown. Specifically, the original work condition dataset is first traversed to filter out work condition segments where the starting / ending speed is not 0 within a single day, and work condition segments where the speed is 0 for an extended period. For the former, the zero-speed point closest to the starting (ending) point is found, and the data segment between the zero-speed point and the starting (ending) point is deleted; for the latter, the duration t of the work condition segment is calculated. d Delete duration t d The work segment exceeded 3 minutes. Secondly, the sampling time t between data points... diff Perform difference calculation, if the sampling time difference t diff If the sampling frequency exceeds the set frequency, signal loss is considered to exist at that point. For signal loss durations of one sampling step, linear interpolation is performed by averaging the speeds of the previous and next moments. For longer signal loss durations, the KNN algorithm is used to predict and fill in the missing vehicle speed values. For excessively long signal loss durations, the two nearest zero-speed points before and after the missing data point are directly concatenated, and other data between the two zero-speed points is deleted. Then, the acceleration data 'a' for each moment is calculated based on the speed data, and a maximum acceleration / deceleration threshold 'a' is set for each moment. b,max and a a,maxAcceleration data points exceeding the threshold are corrected to the corresponding threshold, and the vehicle speed data before and after the abnormal data points are corrected according to the corrected acceleration values. Finally, a representation learning method based on an autoencoder is used to decompose and reconstruct the working condition data, mapping the original data to a low-dimensional feature space, and performing parameter reproduction sampling and clustering based on the compressed features. Then, the sampling points in the latent space are mapped according to the data categories to generate a reconstructed working condition with the same features as the original working condition.

[0024] S102. Standardized Trimming. Due to the randomness of vehicle travel tasks in daily life, the duration of individual operating condition sample data after preprocessing is inconsistent. To facilitate subsequent operating condition clustering analysis and database construction, this embodiment provides a composite trip segmentation method to standardize and trim the data processed in the previous stage. Specifically, firstly, timing begins from the start time of the operating condition according to the micro-trip segmentation method and ends when the vehicle stops for the first time after driving. Secondly, the duration of this micro-trip is judged. If the duration is greater than a certain set threshold, the micro-trip is recorded as a qualified short-trip operating condition, the line number is marked, and the current timing is terminated. If the duration is less than a certain set threshold, timing continues until the next micro-trip ends, and the duration is judged again until a qualified short-trip operating condition is obtained. Finally, after obtaining a qualified short trip, the timing and judgment of the next micro-trip begins, and so on until the operating condition ends.

[0025] S103. Feature Parameter Extraction. Further, in order to mine potential operating condition category information in the driving data and establish a mapping relationship between vehicle driving feature parameters and operating condition characteristics, for example, this embodiment initially screens out 42 feature parameters related to vehicle energy consumption as initial feature parameters. Using the operating condition segments trimmed in the previous stage as units, their feature parameter sets are calculated according to the formulas in Table 1, and normalized using the Z-value standardization method. To further quantify and analyze the contribution level of each feature parameter to energy consumption and screen out key feature parameters highly correlated with vehicle energy consumption, this embodiment establishes the vehicle specific energy consumption according to formula (1). With characteristic parameters (in, , , , A multiple linear regression analysis model is used to analyze the relationship between the number of characteristic parameters and the specific energy consumption, and the correlation between each characteristic parameter and the specific energy consumption is calculated using the cosine method of the included angle according to equation (3). For example, this embodiment selects... As a correlation criterion, 18 characteristic parameters that are strongly correlated with specific energy consumption were selected, as shown in Equation (4).

[0026] (1) (2) (3) (4) In the formula, Augment the feature vectors for the feature parameters of each working condition segment; This represents the augmented weight vector for each working condition segment; This is the weight vector; For bias; for n Specific energy consumption for each operating condition segment; For the first j Characteristic parameters and specific energy consumption The correlation coefficient between them; For the first of all sample data j The average value of each characteristic parameter; This represents the average energy consumption across all sample data.

[0027] Table 1 S104. Feature Parameter Dimensionality Reduction. To remove redundant information among the selected specific energy consumption-related feature parameters and enhance their representational role in distance metrics, this embodiment employs principal component analysis (PCA), a common method in unsupervised machine learning, to reduce the dimensionality of the feature parameters. For example, this embodiment sets the cumulative contribution rate of the principal components to be higher than 90%, which can represent most of the operating condition information represented by the original feature parameters. The first four principal components in Table 2 are selected to represent the specific energy consumption feature parameters.

[0028] Table 2 S105. Construction of the Working Condition Database. Further, this embodiment uses a cropped standard working condition segment as the initial dataset and provides a data clustering method combining agglomerative hierarchical clustering and K-means++ cross-validation for working condition clustering analysis. Specifically, the process is as follows: 1) Treat the principal components of each working condition segment as a separate data cluster and calculate the Euclidean distance between each data cluster; 2) Merge the two closest data clusters into one cluster and calculate the Euclidean distance between the merged new data clusters; 3) Repeat step 2) until all data are merged into one cluster; 4) Set different clustering scales according to the hierarchical clustering tree, select the number of working condition data clusters, calculate the DB index for different numbers of clusters, and select the optimal number of clusters. k and the corresponding cluster centers; 5) Randomly select initial cluster centers from the working condition segments. c 1. Calculate the subsequent working condition segments sequentially. c The distance of 1 and the probability of being selected as the next cluster center Select random numbers ,like Then the first i Each working condition segment was identified as a cluster center. c 2, until selected k 6) Calculate the Euclidean distance from each work condition segment to each cluster center and assign it to the data cluster containing the nearest cluster center; 7) Recalculate the average value of the samples in each data cluster and set it as the cluster center of that data cluster; 8) Repeat steps 6)-7) until all work condition segments are classified and the cluster centers no longer change; 9) Verify the Euclidean distance between the cluster centers obtained in 4) and the cluster centers obtained in 8). If it is less than a certain threshold, the clustering is considered complete; otherwise, repeat steps 5)-8). For example, in this embodiment, hierarchical clustering is performed according to steps 1)-3), the number of clusters is set to 2 to 6 according to step 4) and the DB index for different numbers of clusters is calculated. The preferred number of clusters is 4, and K-means++ clustering calculation and cluster center verification are performed according to steps 5)-9). Based on the clustering results, the short-distance segments are sorted according to their distance from the cluster center. The top 50% of the short-distance segments are selected as representative operating conditions for each type of operating condition and form a basic database of driving operating conditions. The retraining flag of the energy management strategy, Flag_opt, is set to 1.

[0029] S106. Construction of Driving Style Information Database. Further, this embodiment, based on the driving style behavior mechanism and causal relationship theory, uses a basic driving condition database as a foundation to mine the driving style information contained therein and classify and label it, thereby completing the establishment of the driving style information database. For example, this embodiment preferably uses three indicators—average vehicle speed, acceleration coefficient, and driving style coefficient—to mine the driving style information contained in the driving condition database. Based on the aforementioned basic driving condition database, for vehicle driving data under different driving condition categories, the average vehicle speed, average acceleration, and average impact are calculated in short-distance segments, as shown in formulas (5) and (6). Considering the sample size, this embodiment preferably divides driving styles into three categories and names them as: aggressive, general, and moderate. After obtaining the driving style index dataset and the number of driving style categories, the indices in the dataset are normalized using Min-Max to bring them to the same order of magnitude. Then, the K-means++ algorithm in unsupervised learning is used to perform cluster analysis on the datasets of different driving conditions. After inverse normalization of the cluster centers, the true cluster centers are obtained, thus completing the construction of the basic information database of driving style.

[0030] (5) (6) In the formula, It is the first i The acceleration at each instant; It is the first i The impact of a moment; It is the acceleration coefficient; It is the driving style coefficient. N For time.

[0031] (2) Rolling optimization layer To achieve efficient utilization of long-term vehicle operation data collected and stored daily on the big data cloud platform, and to solve the data redundancy problem caused by the same vehicle's travel conditions and driving style characteristics, this embodiment provides a method for selecting new driving condition data for database entry and self-updating the database. S201, Data Entry Selection. Specifically, for daily vehicle driving data newly collected after the construction date of the driving condition basic database and driving style basic information database in the big data cloud platform, refined processing, standardized pruning, feature parameter extraction, and feature parameter dimensionality reduction are performed according to the methods exemplified in S101, S102, S103, and S104. The weighted average Euclidean distance between the principal components of each standard driving condition segment and the cluster centers exemplified in S105 is calculated. If this value is greater than a certain set threshold (Bias), the segment is considered to have differential value compared to the driving conditions in the basic database and is stored in the database, and the cumulative count value Sum_1 of the newly entered segments is incremented by 1; otherwise, the segment is deleted, and other newly collected driving data are re-selected according to the method exemplified in S201. S202, Database Update. Specifically, when the cumulative count value Sum_1 of newly added segments exceeds a certain set threshold, all operating condition segments in the database are subjected to operating condition clustering analysis, driving condition database construction, driving information mining, driving style clustering analysis and information database construction again according to the methods shown in S105 and S106, the database is updated, and the retraining flag bit Flag_opt of the energy management strategy is set to 1.

[0032] To achieve rolling optimization of the control algorithm in the energy management strategy based on long-term vehicle operation data, this embodiment provides a method for triggering control parameter retraining and OTA updates. For example, it is determined whether the retraining flag (Flag_opt) is 1. If Flag_opt is 1, control parameter retraining and OTA updates are performed according to the following process; otherwise, no operation is performed. S203, Driving condition synthesis. Specifically, segments are sequentially selected from the updated driving condition database in S202 (or initially from the basic driving condition database in S105) according to the driving condition clustering category using random sampling without replacement. These segments are then stitched together sequentially along the timeline to achieve driving condition synthesis. When the duration of a single synthesized driving condition exceeds a certain set threshold, the synthesis of that driving condition is terminated, and the total number of synthesized driving conditions (Sum_2) is incremented by 1; otherwise, segments are selected for synthesis. It is determined whether the total number of synthesized driving conditions (Sum_2) exceeds a certain set threshold. If so, driving condition synthesis is stopped; otherwise, driving condition synthesis continues. S204, Training set acquisition. Specifically, for the synthesized operating conditions, offline optimization and online simulation of control parameters are performed respectively. For example, using the synthesized operating conditions as input, the control parameters are used as optimization variables, and the energy consumption result of the energy management algorithm is used as the optimization objective function value. The firefly algorithm is used to iteratively optimize the control parameters, and the set of control parameters with the minimum objective function value and their corresponding operating condition data are recorded. This set is then used as the training set for the neural network (S301 specifies the specific neural network). S205, Control Parameter Optimization. Specifically, after completing the offline optimization of all synthesized operating conditions, the training set is divided according to a certain set ratio for neural network training, testing, and verification. For example, a preferred training set division ratio is 7:2:1. Mini-batch stochastic gradient descent is used to improve training efficiency and prevent local optima. The initial learning rate for network training is set to 0.01, and the maximum number of iterations is set to 1000. The neural network is trained, and after training, testing, and verification are completed, the optimal neural network parameters stored in the big data platform are updated. Furthermore, driving style clustering results under each operating condition category are extracted from the driving style information database updated in S202 (or initially from the driving style basic information database in S106). The membership function range and fuzzy rules of the fuzzy controller are designed (the specific process in S302), and the optimal fuzzy control parameters stored on the big data platform are updated. After updating the optimal neural network parameters and optimal fuzzy control parameters, the retraining flag Flag_opt is set to 0. S206, OTA Update. An OTA update request is sent to the vehicle user, and it is determined whether the vehicle is under suitable conditions for OTA update. If both are met, an OTA update is performed. For example, it is determined whether the vehicle hardware platform meets the update conditions; if so, a software OTA update request is sent to the user; otherwise, no operation is performed.Determine if the user accepts OTA software updates. If so, determine if the vehicle is suitable for OTA updates; otherwise, do not perform any operation. If the vehicle is suitable for OTA updates, perform the software OTA update; otherwise, wait for the vehicle's environment to update before making a further determination.

[0033] (3) Online implementation layer To fully utilize multimodal driving data during vehicle operation and achieve dynamic adjustment of control parameters based on this data, this embodiment provides an online implementation method for a multi-information collaborative intelligent energy management algorithm. S301, Adaptive Parameter Estimation. For example, firstly, a neural network structure is designed with five input features: demand power, vehicle speed, SOC, deviation between expected and actual SOC, pure electric operating time, and optimal control parameters, and adaptive control parameters as the output feature. The preferred neural network layer is 3 layers, with 5 neurons in the input layer corresponding to the 5 input features and 2 neurons in the output layer corresponding to the adaptive control parameters. The hidden layer neurons are preferably 10, representing the number of neurons with the smallest error on the training set. Secondly, a ReLU function with leakage is preferred as the activation function. Then, a squared loss function is preferred to calculate the error between the output value and the true value. Finally, the optimal neural network parameters are set according to the method exemplified in S205, thereby enabling online estimation of the optimal adaptive control parameters based on parameters such as vehicle demand power, vehicle speed, SOC, SOC deviation, and pure electric operating time during vehicle operation. S302, Driving Style Control Parameter Compensation. For example, firstly, a fuzzy controller is designed according to different operating condition categories, using the absolute values ​​of the average accelerator pedal opening and the average rate of change within the recognition period as inputs and driving style as the output. The average accelerator pedal opening and the average rate of change are defined as input linguistic variables, and driving style as the output linguistic variable. Secondly, the universe of discourse for the average accelerator pedal opening is defined as {0, 0.1, 0.2, ..., 0.9, 1}, the universe of discourse for the absolute value of the average rate of change of the accelerator pedal opening is defined as {0, 0.1, 0.2, ..., 0.7, 0.8}, and the universe of discourse for driving style is defined as {0, 0.1, 0}. Given the input variables {0.2, ..., 0.9, 1}, define the fuzzy linguistic values ​​as {small (S), medium (M), large (B)} and the output variables as {soft, normal, aggressive (Agg)}. Then, select the Z-type, Gaussian, and S-type joint membership functions as the membership functions of the input variables, and select the Gaussian membership function as the membership function of the output variables. Set the optimal fuzzy control parameters for different operating conditions according to the method in the example in S205. At any time during vehicle operation, with a preset recognition period t... cObtain specific energy consumption-related characteristic parameters and derive principal component values. Then, perform distance analysis with cluster centers in the operating condition database to identify the current operating condition, using this as the preset update period t. u The internal working condition types are determined, and the preset update cycle t is calculated. u The average accelerator pedal opening and average rate of change of opening are used to identify the fuzzy controller under the corresponding working condition, thereby realizing the identification of driving style under different working condition categories in the current time period; finally, the expected value of driving pedal opening for each driving style under different working condition categories is calculated according to the driving style information database in S202, and the compensation coefficient of the control parameter is calculated according to the expected value of driving pedal opening according to formula (7). S303, road slope control parameter compensation. For example, firstly, the state space equation for road slope estimation is constructed using a strong tracking Kalman filter algorithm. A covariance matrix is ​​defined, and prior estimates and error covariance are calculated. The error covariance matrix is ​​corrected with a suboptimal fading factor. Then, the posterior error covariance and estimate are calculated according to the measurement update equation, and the optimal road slope estimate is calculated. Secondly, road slope information is calculated in real-time based on vehicle location and altitude information provided by the intelligent transportation system. When the signal quality is normal and k=nT (T is a preset update period), the optimal slope estimate is corrected based on the current road slope information. Otherwise, the optimal slope estimate from the strong tracking Kalman filter is output. Finally, the optimal road slope estimate is calculated based on the current optimal road slope estimate. Calculate the vehicle's power demand at the current slope and the vehicle's power demand when the slope is 0, respectively, and calculate the control parameter compensation coefficient based on their ratio, as shown in equation (8). S304, Multi-information Collaborative Intelligent Energy Management. Specifically, the adaptive optimal control parameters calculated by the method in example S301 are multiplied sequentially with the compensation coefficients of the driving style and road slope control parameters calculated by the methods in examples S302 and S303 to obtain multi-information collaborative control parameters. Based on the current working state and output capacity limitations of the power components, the upper and lower limits of the optimization variables are set according to the maximum and minimum output limits of the power components represented by the optimization variables. Combining the controller's computing power and the actual operating state of the vehicle, optimization freezing logic is set to prevent frequent changes in the control commands of the power components caused by frequent optimization. When the changes in the vehicle's required torque, vehicle speed, etc., are small compared to the previous moment, the optimization freezing flag is activated. At this time, optimization calculation is no longer performed. Instead, the optimal control command from the previous moment is used to perform iterative optimization calculation of the objective function. The optimal control quantity combination corresponding to the lowest energy consumption cost is sent as the control command to each component for execution, completing the intelligent energy management of the new energy vehicle. The real-time driving data of the vehicle is uploaded to the big data platform in example S201.

[0034] (7) (8) In the formula, , , These are the control parameter compensation coefficients for three driving styles; E Soft , E Norm , E Agg These represent the expected accelerator pedal opening values ​​for drivers with three different driving styles. The compensation coefficient for the control parameters of driving style; m For vehicle quality; g It is the acceleration due to gravity; C d , f These are the drag coefficients and rolling resistance coefficients; A For windward area; ρ air density; For the current moment k The vehicle speed; The compensation coefficient is the control parameter for road slope.

[0035] Implementation 2 The present invention also provides an offline optimization and online implementation system for intelligent energy management strategies of new energy vehicles. The system is used to implement the method described in Embodiment 1. The system includes: an offline processing layer, a rolling optimization layer, and an online implementation layer. The offline processing layer is used to process real-time vehicle driving data offline and establish a basic database of driving conditions and a basic information database of driving style. The rolling optimization layer is used to perform rolling optimization of control parameters related to driving condition characteristics based on daily vehicle driving data, driving condition database, and driving style information database. The online implementation layer is used to estimate road gradient in real time and correct the control parameters after rolling optimization to realize intelligent energy management of new energy vehicles.

[0036] In this embodiment, the offline processing layer includes: a data acquisition unit, a processing unit, an extraction unit, and an analysis unit; The data acquisition unit is used to collect real-time vehicle driving data based on a big data platform to obtain the raw dataset. The processing unit is used to perform fine processing and standardized cropping on the original dataset to obtain various standard working condition segments; The extraction unit is used to extract specific energy consumption characteristic parameters from each clipped standard working condition segment and perform dimensionality reduction on them. The analysis unit is used to perform cluster analysis on standard operating condition segments based on the dimensionality-reduced feature parameters, and extract representative operating conditions from various standard operating condition segments to form a basic database of driving operating conditions. After designing indicators to mine and cluster driving information, a basic information database of driving styles is established.

[0037] In this embodiment, the rolling optimization layer includes: a data entry and selection unit, a threshold-triggered update unit, an offline simulation training unit, and an OTA remote deployment unit; The data entry selection unit is used to select newly collected data from daily vehicle operation in the online implementation layer for entry into the database. The threshold-triggered update unit is used to update the working condition basic database and driving style basic information database according to the offline processing layer method when the newly added data fragment reaches a certain threshold, and trigger the control parameter retraining instruction. The offline simulation training unit is used to optimize control parameters offline and simulate them in real time using standard fragments of the working condition database. It obtains a training set of control parameters related to driving condition characteristics to train the neural network and designs fuzzy rules based on the clustering results of driving style. The OTA remote deployment unit is used to update neural network parameters and fuzzy rules to the online implementation layer under suitable conditions such as vehicle charging or stationary conditions.

[0038] In this embodiment, the online implementation layer includes: an estimation unit and a correction unit; The estimation unit is used to design a neural network controller that takes driving condition characteristics as input to estimate the parameters of the adaptive control system. The correction unit is used to establish driving style identification methods and corresponding driving style control parameter compensation algorithms under different operating conditions, and to estimate the road slope in real time to correct the control parameters, thereby realizing intelligent energy management of new energy vehicles.

[0039] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for offline optimization and online implementation of intelligent energy management strategy for new energy vehicles, characterized in that, The method includes: Offline processing of real-time vehicle driving data is performed to establish a basic database of driving conditions and a basic information database of driving styles. Based on daily vehicle driving data, a basic database of driving conditions, and a basic information database of driving styles, rolling optimization is performed on control parameters related to driving condition characteristics. Real-time estimation of road gradient and correction of control parameters after rolling optimization enable intelligent energy management of new energy vehicles.

2. The method according to claim 1, characterized in that, Methods for offline processing of real-time vehicle driving data to establish a basic database of driving conditions and a basic database of driving style information include: The raw dataset is obtained by collecting real-time vehicle driving data based on a big data platform; The original dataset is refined and standardized by cropping to obtain segments of various standard working conditions; Extract specific energy consumption characteristic parameters from each clipped standard operating condition segment and perform dimensionality reduction on them; Cluster analysis is performed on standard operating condition segments based on the dimensionality-reduced feature parameters, and representative operating conditions from each type of standard operating condition segment are extracted to form a basic database of driving operating conditions. After designing indicators to mine and cluster driving information, a basic information database of driving styles is established.

3. The method according to claim 1, characterized in that, Methods for rolling optimization of control parameters related to driving condition characteristics based on daily vehicle driving data, a basic database of driving conditions, and a basic information database of driving styles include: Newly collected data from daily vehicle operation is selected for inclusion in the database. When the number of newly added data segments reaches a preset threshold, the working condition database and driving style database are updated, and a control parameter retraining instruction is triggered. Using standard fragments of the operating condition database, offline optimization and real-time simulation of control parameters are performed. A training set of control parameters related to driving condition characteristics is obtained to train the neural network, and fuzzy rules are designed based on the clustering results of driving style. Over-the-air (OTA) updates of neural network parameters and fuzzy rules are performed when the vehicle is charging or stationary.

4. The method according to claim 1, characterized in that, Methods for achieving intelligent energy management of new energy vehicles by real-time estimation of road gradient and correction of control parameters after rolling optimization include: Design a neural network controller that takes driving condition characteristics as input to estimate the adaptive control parameters for driving conditions; A driving style identification method and a corresponding driving style control parameter compensation algorithm are established for different operating conditions. The control parameters are then corrected by estimating the road slope in real time, thereby realizing intelligent energy management for new energy vehicles.

5. The method according to claim 4, characterized in that, Methods for establishing driving style control parameter compensation algorithms include: ； In the formula, , , These are the control parameter compensation coefficients for three driving styles; E Soft , E Norm , E Agg These represent the expected accelerator pedal opening values ​​for drivers with three different driving styles.

6. The method according to claim 4, characterized in that, Methods for real-time estimation of road slope to correct control parameters include: ； In the formula, , , These are the control parameter compensation coefficients for three driving styles; E Soft , E Norm , E Agg These represent the expected accelerator pedal opening values ​​for drivers with three different driving styles. The compensation coefficient for the control parameters of driving style; m For vehicle quality; g It is the acceleration due to gravity; C d , f These are the drag coefficients and rolling resistance coefficients; A For windward area; ρ air density; For the current moment k The vehicle speed; The compensation coefficient is the control parameter for road slope. This is the optimal estimate of the road slope.

7. A system for offline optimization and online implementation of intelligent energy management strategies for new energy vehicles, the system being used to implement the method described in any one of claims 1-6, characterized in that, The system includes: an offline processing layer, a rolling optimization layer, and an online implementation layer; The offline processing layer is used to process real-time vehicle driving data offline and establish a basic database of driving conditions and a basic information database of driving style. The rolling optimization layer is used to perform rolling optimization of control parameters related to driving condition characteristics based on daily vehicle driving data, driving condition database, and driving style information database. The online implementation layer is used to estimate the road gradient in real time and correct the control parameters after rolling optimization, so as to realize intelligent energy management of new energy vehicles.

8. The system according to claim 7, characterized in that, The offline processing layer includes: a data acquisition unit, a processing unit, an extraction unit, and an analysis unit; The acquisition unit is used to collect real-time vehicle driving data based on a big data platform to obtain a raw dataset; The processing unit is used to perform fine processing and standardized cropping on the original dataset to obtain various standard working condition segments; The extraction unit is used to extract specific energy consumption characteristic parameters from each clipped standard working condition segment and perform dimensionality reduction processing on them. The analysis unit is used to perform cluster analysis on standard operating condition segments based on the dimensionality-reduced feature parameters, and extract representative operating conditions from various standard operating condition segments to form a basic database of driving operating conditions. After designing indicators to mine and cluster driving information, a basic information database of driving styles is established.

9. The system according to claim 7, characterized in that, The rolling optimization layer includes: a data entry and selection unit, a threshold-triggered update unit, an offline simulation training unit, and an OTA remote deployment unit; The data entry selection unit is used to select newly collected data from daily vehicle operation for entry into the database. The threshold-triggered update unit is used to update the working condition basic database and the driving style basic information database when the newly added data fragment reaches a preset threshold, and to trigger the control parameter retraining instruction. The offline simulation training unit is used to optimize control parameters offline and simulate them in real time using standard fragments of the working condition database, obtain a control parameter training set related to driving condition characteristics to train the neural network, and design fuzzy rules based on driving style clustering results. The OTA remote deployment unit is used to update neural network parameters and fuzzy rules via OTA when the vehicle is charging or stationary.

10. The system according to claim 7, characterized in that, The online implementation layer includes: an estimation unit and a correction unit; The estimation unit is used to design a neural network controller that takes driving condition characteristics as input to estimate the driving condition adaptive control parameters. The correction unit is used to establish driving style identification methods and corresponding driving style control parameter compensation algorithms under different operating conditions, and to estimate the road slope in real time to correct the control parameters, thereby realizing intelligent energy management of new energy vehicles.

Images