A new energy vehicle power management method and system based on machine learning

Abstract

The present invention discloses a method and system for electric energy management of new energy vehicles based on machine learning, which relates to the technical field of electric energy management of new energy vehicles, including collecting multimodal data through various sensors and external data interfaces; processing multimodal data by eliminating outliers, filling missing values ​​and standardizing; extracting features from the processed multimodal data using a multimodal convolutional neural network, and fusing them into a comprehensive feature vector; inputting the comprehensive feature vector into a reinforcement learning model, and generating an electric energy management strategy through interactive learning with the environment; inputting the electric energy management strategy into an online learning framework of an adaptive learning rate, and optimizing the electric energy management strategy in real time. The present invention significantly improves the efficiency and performance of electric energy management of new energy vehicles by inputting the comprehensive feature vector into the reinforcement learning model, and generating an electric energy management strategy through interactive learning with the environment.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy vehicle power management, and in particular to a new energy vehicle power management method and system based on machine learning. Background Art

[0002] With the transformation of the global energy structure and the enhancement of environmental protection awareness, the development of new energy vehicles (NEVs) has become an important trend in the automotive industry. With the rapid development of machine learning technology, power management methods based on machine learning have gradually become a research hotspot. These methods can adaptively adjust power management strategies by learning a large amount of historical data, thereby improving the energy efficiency and battery life of new energy vehicles.

[0003] Existing data processing methods often ignore the fusion of multimodal data, resulting in the model being unable to fully capture the complex information of vehicle operation. For example, battery status data, driving behavior data, and environmental condition data are usually processed independently, lacking an effective fusion mechanism, which limits the model's ability to comprehensively analyze multi-source information. In addition, most of the existing power management strategy optimization methods are based on offline learning and cannot adapt to changes in driving environment and driving behavior in real time, resulting in insufficient flexibility and real-time performance of power management strategies. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a new energy vehicle power management method and system based on machine learning to solve the problems of insufficient multimodal data fusion and unreal-time optimization of power management strategies in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a new energy vehicle power management method based on machine learning, which includes:

[0008] Collect multimodal data through various sensors and external data interfaces;

[0009] Processing multimodal data by removing outliers, filling missing values ​​and standardizing;

[0010] Use a multimodal convolutional neural network to extract features from the processed multimodal data and fuse them into a comprehensive feature vector;

[0011] The comprehensive feature vector is input into the reinforcement learning model, and the power management strategy is generated by interactive learning with the environment;

[0012] The power management strategy is input into an online learning framework with adaptive learning rate to optimize the power management strategy in real time.

[0013] As a preferred solution of the new energy vehicle power management method based on machine learning described in the present invention, wherein: the multimodal data is collected through various sensors and external data interfaces, and the specific steps are as follows:

[0014] The sensors arranged on the battery of the new energy vehicle collect the battery voltage, current, temperature and the percentage of remaining battery power to obtain the battery status data;

[0015] The on-board computer records the vehicle's acceleration and speed to obtain driving behavior data;

[0016] The temperature and humidity inside and outside the vehicle are collected through the environmental temperature and humidity sensor. The vehicle camera cooperates with the GPS module to obtain the vehicle's location and road conditions information to obtain environmental condition data.

[0017] The battery status data, driving behavior data and environmental condition data are packaged into standard data packets according to timestamps and data categories, and transmitted to the on-board computing unit via the CAN bus.

[0018] As a preferred solution of the new energy vehicle power management method based on machine learning described in the present invention, wherein: the multimodal data is processed by removing outliers, filling missing values ​​and standardizing, and the specific steps are as follows:

[0019] The received data is subjected to outlier elimination and missing value filling in by the on-board computing unit;

[0020] Use the Z-score method to detect outliers and calculate the mean and standard deviation of all data points;

[0021] Based on the mean and standard deviation of the data points, the Z-score value of each data point is calculated, the Z-score values ​​of all data points are normally distributed, and a threshold Z is set.

[0022] When the Z-score value of a data point exceeds Z, the data point is considered an outlier and is removed; otherwise, it is retained.

[0023] Fill the missing values ​​by linear interpolation, mark the position of the missing values, find the previous and next data points, calculate the interpolation of the missing values, and insert the calculated interpolation into the corresponding missing value position;

[0024] Traverse all data points in each type of data to obtain the maximum and minimum values ​​of each type of data, and calculate the Min-Max standardized value for each data point in each type of data.

[0025] As a preferred solution of the new energy vehicle power management method based on machine learning described in the present invention, wherein: the multimodal convolutional neural network is used to extract features from the processed multimodal data and fuse them into a comprehensive feature vector. The specific steps are as follows:

[0026] Align the cleaned and standardized battery status data, driving behavior data, and environmental condition data according to timestamps to form a unified time series, with each timestamp corresponding to a set of data points;

[0027] Build a multimodal neural network model and design three input layers to receive battery status data, driving behavior data, and environmental condition data respectively;

[0028] The data received by each input layer is a two-dimensional array consisting of the time series length and the feature dimension at each timestamp;

[0029] Each input layer is followed by an independent convolutional layer, which uses a one-dimensional convolutional layer to extract the time series feature vectors of battery status data, driving behavior data, and environmental condition data;

[0030] Add a pooling layer after each convolutional layer and select the largest eigenvalue in each pooling window as the output;

[0031] The battery status data feature vector, driving behavior data feature vector, and environmental condition data feature vector extracted by the convolution layer and the pooling layer are concatenated along the feature dimension in chronological order to obtain a comprehensive feature vector, the dimension of which is the sum of the dimensions of the three data feature vectors.

[0032] As a preferred solution of the new energy vehicle power management method based on machine learning described in the present invention, wherein: the comprehensive feature vector is input into the reinforcement learning model, and the power management strategy is generated by interactive learning with the environment. The specific steps are as follows:

[0033] Based on the PPO algorithm, a reinforcement learning model consisting of a policy network and a value network is constructed. The policy network outputs the probability distribution of action vectors, and the value network outputs the state value.

[0034] Establish a training environment model, and the reinforcement learning model interacts with it to generate power management strategies;

[0035] The comprehensive feature vector is used as the state vector and divided into a training set and a test set in proportion for training and performance verification of the reinforcement learning model;

[0036] The advantage function of each state is calculated by combining the rewards of multiple time steps using the generalized advantage estimate, which is expressed as follows:

[0037]

[0038] Among them, A(s, a) is the advantage function, which indicates how much higher the expected return of selecting action a in state s is than that of randomly selecting an action, T is the total number of steps from state s to the end state, t is the time step, γ is the discount factor, λ is the smoothing parameter of the generalized advantage estimate, and r t is the immediate reward received at time step t, V(s t ) is the state value at time step t, V(s t+1 ) is the state value at time step t+1;

[0039] The PPO algorithm maximizes the expected return J(θ) by calculating the objective function of the policy network and optimizes the policy network parameter θ. The expression is as follows:

[0040]

[0041] Among them, E s，a represents the average value of all values ​​of state s and action a, π θ (a|s) is the probability that the current policy network selects action a in state s, π′ θ (a|s) is the probability that the old policy network selects action a in state s, ∈ is the clipping range, and clip means to change π θ (a|s) / π′ θ (a|s) is restricted to the interval [1-∈, 1+∈];

[0042] By calculating the loss function of the value network, using the gradient descent method, minimizing the value function, and updating the value network parameters, the expression is as follows:

[0043]

[0044] V target (s j )=r+γV φ (s j+1 );

[0045] Among them, N is the number of state s, j is the index variable of the number of state s, L V is the loss function of the value network, V φ (s j ) is the value predicted by the value network in the jth state, φ is the parameter of the value network, V target (s j ) is the target value of the jth state, r is the immediate reward, V φ (s j+1 ) is the predicted value of the j+1th state;

[0046] After each training round, the performance of the model is evaluated on the validation set. When the performance fluctuation on multiple consecutive validation sets does not exceed 5%, the model training is considered complete.

[0047] Generate power management strategies based on the actions output by the model.

[0048] As a preferred solution of the new energy vehicle power management method based on machine learning described in the present invention, wherein: based on the reinforcement learning model, a training environment model is established, and the specific steps are as follows:

[0049] Based on the training requirements of the reinforcement learning model, a training environment model is constructed by setting the state space, action space and reward function;

[0050] Generate a series of actions based on the initial state vector, and record the corresponding states and rewards. Store the state vector, action, reward of each step, and the next state generated based on the previous action in an experience replay buffer.

[0051] As a preferred solution of the new energy vehicle power management method based on machine learning described in the present invention, wherein: the power management strategy is input into the online learning framework of the adaptive learning rate to optimize the power management strategy in real time. The specific steps are as follows:

[0052] Build an online learning framework based on the Adam optimizer to collect battery status data, driving behavior data, and environmental condition data in real time;

[0053] Run the reinforcement learning model, process the real-time data and integrate the features into the reinforcement learning model, and introduce the Adam optimizer in the strategy update and value update;

[0054] Using the chain rule, calculate the gradient of the policy network objective function with respect to the parameter θ and the gradient of the value network loss function with respect to the parameter φ;

[0055] Updated by the Adam optimizer, the policy network parameters θ and the value network parameters φ are updated.

[0056] In a second aspect, the present invention provides a new energy vehicle power management system based on machine learning, including a data collection module, a data processing module, a feature fusion module, a strategy generation module and a strategy optimization module;

[0057] The data collection module is used to collect multimodal data through various sensors and external data interfaces;

[0058] The data processing module is used to process the multimodal data by removing outliers, filling missing values ​​and standardizing;

[0059] The feature fusion module is used to extract features from the processed multimodal data using a multimodal convolutional neural network and fuse them into a comprehensive feature vector;

[0060] The strategy generation module is used to input the comprehensive feature vector into the reinforcement learning model and generate the power management strategy through interactive learning with the environment;

[0061] The strategy optimization module is used to input the power management strategy into an online learning framework with an adaptive learning rate to optimize the power management strategy in real time.

[0062] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the new energy vehicle power management method based on machine learning as described in the first aspect of the present invention is implemented.

[0063] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the new energy vehicle power management method based on machine learning as described in the first aspect of the present invention is implemented.

[0064] The beneficial effects of the present invention are as follows: the present invention collects multimodal data through various sensors and external data interfaces, comprehensively collects multi-source information of vehicle operation, ensures the diversity and integrity of data, and improves the adaptability of the model to complex driving environments; by using a multimodal convolutional neural network to extract features from the processed multimodal data and fuse them into a comprehensive feature vector, the accuracy and comprehensiveness of feature extraction are improved, and the comprehensive analysis capability of the model for multi-source information is enhanced; by inputting the power management strategy into an online learning framework with an adaptive learning rate, the power management strategy is optimized in real time, the real-time and robustness of the power management strategy are improved, the efficiency and stability of the model in practical applications are ensured, and the energy efficiency and user experience of new energy vehicles are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0066] Figure 1 This is a flow chart of the new energy vehicle power management method based on machine learning in Example 1.

[0067] Figure 2This is a flow chart of the new energy vehicle power management system based on machine learning in Example 1. DETAILED DESCRIPTION

[0068] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0069] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0070] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0071] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a new energy vehicle power management method based on machine learning, comprising the following steps:

[0072] S1. Collect multimodal data through various sensors and external data interfaces.

[0073] Furthermore, the battery voltage, current, temperature and percentage of remaining battery power are collected through sensors arranged on the battery of the new energy vehicle to obtain the battery status data;

[0074] The on-board computer records the vehicle's acceleration and speed to obtain driving behavior data;

[0075] The temperature and humidity inside and outside the vehicle are collected through the environmental temperature and humidity sensor. The vehicle camera cooperates with the GPS module to obtain the vehicle's location and road conditions information to obtain environmental condition data.

[0076] The battery status data, driving behavior data and environmental condition data are packaged into standard data packets according to timestamps and data categories, and transmitted to the on-board computing unit via the CAN bus. During the transmission process, CRC checksum is used for integrity verification to ensure the reliability of data transmission.

[0077] S2. Process the multimodal data by removing outliers, filling missing values ​​and standardizing.

[0078] Furthermore, the received data is subjected to outlier elimination and missing value filling in the on-board computing unit;

[0079] Use the Z-score method to detect outliers and calculate the mean and standard deviation of all data points. The expression is as follows:

[0080]

[0081]

[0082] Where μ is the mean of the data points, n is the number of data points, and x i is the i-th data point, i is the index variable of the number of data points, and σ is the standard deviation of the data points;

[0083] Based on the mean and standard deviation of the data points, the Z-score value of each data point is calculated as follows:

[0084]

[0085] Among them, Z i is the Z-score value of the i-th data point;

[0086] Based on the principle of three times the standard deviation, the Z-score values ​​of all data points are normally distributed and a threshold Z is set, Z = 3σ

[0087] When the Z-score value of the data point | Z i When |>Z, the data point is considered as an outlier and is removed, otherwise it is retained;

[0088] Furthermore, the missing values ​​are filled by linear interpolation, the position of the missing values ​​is marked, and the previous and next data points are found, the interpolation of the missing values ​​is calculated, and the calculated interpolation is inserted into the corresponding missing value position. The expression is as follows:

[0089]

[0090] Among them, t i is the timestamp;

[0091] Traverse all data points in each type of data to obtain the maximum and minimum values ​​of each type of data, and calculate the Min-Max standardized value for each data point in each type of data. The standardized data value will fall within [0, 1], and the expression is as follows:

[0092]

[0093] Among them, x′ i is the standardized value of the data point, min(x) is the maximum value of the data point in each data, and min(x) is the minimum value of the data point in each data;

[0094] If max(x)=min(x) appears in a certain type of data, in order to avoid the error of using zero as the divisor, the normalized value of this type of data is directly defaulted to 0.5.

[0095] S3. Use a multimodal convolutional neural network to extract features from the processed multimodal data and fuse them into a comprehensive feature vector.

[0096] Furthermore, the cleaned and standardized battery status data, driving behavior data, and environmental condition data are aligned according to timestamps to form a unified time series. Each timestamp corresponds to a set of data points, including battery voltage, current, temperature, and battery remaining power percentage, new energy vehicle acceleration and speed, and ambient temperature and humidity;

[0097] Build a multimodal neural network model and design three input layers to receive battery status data, driving behavior data, and environmental condition data respectively;

[0098] The data received by each input layer is a two-dimensional array consisting of the time series length and the feature dimension at each timestamp;

[0099] Each input layer is followed by an independent convolutional layer, which uses a one-dimensional convolutional layer to extract the time series feature vectors of battery status data, driving behavior data, and environmental condition data;

[0100] Add a pooling layer after each convolutional layer, select the largest eigenvalue in each pooling window as the output, reduce the feature dimension, and retain important features;

[0101] Furthermore, the battery status data feature vector, driving behavior data feature vector, and environmental condition data feature vector extracted by the convolution layer and the pooling layer are concatenated along the feature dimension in chronological order to obtain a comprehensive feature vector, the dimension of which is the sum of the dimensions of the three data feature vectors;

[0102] It should be noted that the feature concatenation operation can effectively fuse the feature vectors extracted from different data types to form a comprehensive feature vector, providing high-quality data support for the subsequent generation of power management strategies, ensuring the rationality and effectiveness of the feature vector, and improving the level of intelligence of power management of new energy vehicles.

[0103] S4. Input the comprehensive feature vector into the reinforcement learning model and generate the power management strategy through interactive learning with the environment.

[0104] Furthermore, based on the PPO algorithm, a reinforcement learning model consisting of a policy network and a value network is constructed. The policy network outputs the probability distribution of the action vector, and the value network outputs the state value.

[0105] Based on the training requirements of the reinforcement learning model, a training environment model is constructed by setting the state space, action space and reward function, and the reinforcement learning model interacts with it to generate an energy management strategy;

[0106] The state space is the state vector, which is the same as the comprehensive feature vector and represents the vehicle’s current battery state, driving behavior, and environmental conditions;

[0107] The actions in the action space are adjustable parameters in the power management process, including battery charging and discharging power, motor power distribution, and regenerative braking energy recovery rate;

[0108] The reward function refers to giving positive and negative rewards based on battery life, mileage, and charging time. The expression is as follows:

[0109] R(s,a)=w1×M+w2×S-w3×Z

[0110] Among them, R is the value of the reward function, s is the state vector, a is the action, M is the battery life, that is, maximizing the number of battery cycles, S is the cruising range, maximizing the driving distance after a single charge, Z is the charging time, minimizing the time required for charging, w1, w2 and w3 are the weight coefficients of M, S and Z respectively.

[0111] Generate a series of actions based on the initial state vector, record the corresponding states and rewards, and store the state vector, action, reward, and next state generated based on the previous action in an experience replay buffer.

[0112] Furthermore, the state vector is divided into 80% training set and 20% test set in proportion for training and performance verification of the reinforcement learning model;

[0113] The advantage function of each state is calculated by combining the rewards of multiple time steps using the generalized advantage estimate, which is expressed as follows:

[0114]

[0115] Among them, A(s, a) is the advantage function, which indicates how much higher the expected return of selecting action a in state s is than that of randomly selecting an action, T is the total number of steps from state s to the end state, t is the time step, γ is the discount factor used to reduce the importance of future rewards, usually set to 0.99, λ is the smoothing parameter of the generalized advantage estimate, which is used to balance the impact of short-term and long-term rewards, usually set to 0.95, and r t is the immediate reward received at time step t, V(s t ) is the state value at time step t, V(S t+1 ) is the state value at time step t+1;

[0116] Furthermore, the PPO algorithm is used to maximize the expected return J(θ) by calculating the objective function of the policy network and optimize the policy network parameter θ. The expression is as follows:

[0117]

[0118] Among them, E s，a represents the average value of all values ​​of state s and action a, π θ (a|s) is the probability that the current policy network selects action a in state s, π′ θ (a|s) is the probability that the old policy network selects action a in state s, ∈ is the clipping range, usually set to 0.2, which is used to limit the amplitude of policy updates, and clip means to change π θ (a|s) / π′ θ (a|s) is restricted to the interval [1-∈, 1+∈];

[0119] By calculating the loss function of the value network, using the gradient descent method, minimizing the value function, and updating the value network parameters, the expression is as follows:

[0120]

[0121] V target (s j )=r+γV φ (s j+1 );

[0122] Where N is the number of states s, j is the index variable of the number of states s, and L V is the loss function of the value network, V φ (s j ) is the value predicted by the value network in the jth state, φ is the parameter of the value network, V target (s j ) is the target value of the jth state, r is the immediate reward, V φ (s j+1 ) is the predicted value of the j+1th state;

[0123] Furthermore, after each training round, the performance of the model is evaluated on the validation set, with battery life, driving range and charging time as the performance evaluation indicators of the reinforcement learning model. When the performance fluctuation on multiple consecutive validation sets does not exceed 5%, the model training is considered complete;

[0124] Generate an energy management strategy based on the actions output by the model, including battery charge and discharge power, motor power distribution, and regenerative braking energy recovery rate;

[0125] It should be noted that the reinforcement learning model can effectively learn and generate the optimal power management strategy to improve the battery life and overall energy efficiency of new energy vehicles.

[0126] S5. Input the power management strategy into the online learning framework with adaptive learning rate to optimize the power management strategy in real time.

[0127] Furthermore, an online learning framework is built based on the Adam optimizer to collect battery status data, driving behavior data, and environmental condition data in real time;

[0128] Run the reinforcement learning model, process the real-time data and integrate the features into the reinforcement learning model, and introduce the Adam optimizer in the strategy update and value update;

[0129] Using the chain rule, calculate the gradient of the policy network objective function with respect to the parameter θ and the gradient of the value network loss function with respect to the parameter φ;

[0130] The performance of the power management strategy is monitored in real time. Based on the real-time feedback, the Adam optimizer dynamically adjusts the strategy network parameters θ and the value network parameters φ to ensure the optimal performance of the power management strategy.

[0131] This embodiment also provides a new energy vehicle power management system based on machine learning, including: a data collection module, a data processing module, a feature fusion module, a strategy generation module and a strategy optimization module;

[0132] A data collection module for collecting multimodal data through various sensors and external data interfaces;

[0133] Data processing module, used to process multimodal data by removing outliers, filling missing values ​​and standardizing;

[0134] A feature fusion module is used to extract features from the processed multimodal data using a multimodal convolutional neural network and fuse them into a comprehensive feature vector;

[0135] A strategy generation module is used to input the comprehensive feature vector into the reinforcement learning model and generate the power management strategy through interactive learning with the environment;

[0136] The strategy optimization module is used to input the power management strategy into the online learning framework with adaptive learning rate and optimize the power management strategy in real time.

[0137] This embodiment also provides a computer device, which is suitable for the case of a new energy vehicle power management method based on machine learning, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the new energy vehicle power management method based on machine learning as proposed in the above embodiment.

[0138] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0139] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for implementing the electric energy management of new energy vehicles based on machine learning as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, SRAM for short), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM for short), erasable programmable read-only memory (Erasable Programmable Read Only Memory, EPROM for short), programmable read-only memory (Programmable Red-Only Memory, PROM for short), read-only memory (Read-Only Memory, ROM for short), magnetic storage, flash memory, disk or optical disk.

[0140] In summary, the present invention collects multimodal data through various sensors and external data interfaces, comprehensively collects multi-source information of vehicle operation, ensures the diversity and integrity of data, and improves the adaptability of the model to complex driving environments; extracts features from the processed multimodal data by using a multimodal convolutional neural network, and fuses them into a comprehensive feature vector, thereby improving the accuracy and comprehensiveness of feature extraction and enhancing the comprehensive analysis capability of the model for multi-source information; and optimizes the power management strategy in real time by inputting the power management strategy into an online learning framework with an adaptive learning rate, thereby improving the real-time and robustness of the power management strategy, ensuring the efficiency and stability of the model in practical applications, and improving the energy efficiency and user experience of new energy vehicles.

[0141] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A new energy vehicle power management method based on machine learning, characterized in that: include, Collect multimodal data through various sensors and external data interfaces; Processing multimodal data by removing outliers, filling missing values ​​and standardizing; Use a multimodal convolutional neural network to extract features from the processed multimodal data and fuse them into a comprehensive feature vector. The specific steps include: Align the cleaned and standardized battery status data, driving behavior data, and environmental condition data according to timestamps to form a unified time series, with each timestamp corresponding to a set of data points; Construct a multimodal neural network model and design three input layers to receive battery status data, driving behavior data, and environmental condition data respectively; The data received by each input layer is a two-dimensional array consisting of the time series length and the feature dimension at each timestamp; The comprehensive feature vector is input into the reinforcement learning model, and the power management strategy is generated through interactive learning with the environment. The specific steps are as follows: Construct a reinforcement learning model consisting of a policy network and a value network based on the PPO algorithm; Establish a training environment model, and use the reinforcement learning model to interact with it to generate power management strategies; The comprehensive feature vector is used as the state vector and divided into a training set and a test set; The advantage function of each state is calculated by combining the rewards of multiple time steps using the generalized advantage estimate, which is expressed as follows: Where A(s,a) is the advantage function, T is the total number of steps from state s to the end state, t is the time step, γ is the discount factor, λ is the smoothing parameter of the generalized advantage estimate, and r t is the immediate reward received at time step t, V(s t ) is the state value at time step t, V(s t+1 ) is the state value at time step t+1; The objective function of the policy network is calculated by the PPO algorithm to maximize the expected return J(θ) and optimize the policy network parameter θ. The expression is as follows: Among them, E s,a represents the average value of all values ​​of state s and action a, π θ (a|s) is the probability that the current policy network selects action a in state s, π ′ θ (a|s) is the probability that the old policy network selects action a in state s, ∈ is the clipping range, and clip means to change π θ (a|s) / π ′ θ (a|s) is restricted to the interval [1-∈, 1+∈]; Using the gradient descent method, by calculating the loss function of the value network, minimizing the value function, and updating the value network parameters, the expression is as follows: 5 target (s j )=r+γV φ (s j+1 ); Among them, N is the number of state s, j is the index variable of the number of state s, L V is the loss function of the value network, V φ (s j ) is the value predicted by the value network in the jth state, φ is the parameter of the value network, V target (s j ) is the target value of the jth state, r is the immediate reward, V φ (s j+1 ) is the predicted value of the j+1th state; After each training round, the performance of the model is evaluated on the validation set; Generate power management strategies based on the actions output by the model; The power management strategy is input into an online learning framework with adaptive learning rate to optimize the power management strategy in real time.

2. The method for managing electric energy of new energy vehicles based on machine learning as claimed in claim 1, characterized in that: The specific steps of collecting multimodal data through various sensors and external data interfaces are as follows: The sensors arranged on the battery of the new energy vehicle collect the battery voltage, current, temperature and the percentage of remaining battery power to obtain the battery status data; The on-board computer records the vehicle's acceleration and speed to obtain driving behavior data; The temperature and humidity inside and outside the vehicle are collected through the environmental temperature and humidity sensor. The vehicle camera cooperates with the GPS module to obtain the vehicle's location and road conditions information to obtain environmental condition data. The battery status data, driving behavior data and environmental condition data are packaged into standard data packets according to timestamps and data categories, and transmitted to the on-board computing unit via the CAN bus.

3. The new energy vehicle power management method based on machine learning as claimed in claim 2, characterized in that: The multimodal data is processed by removing outliers, filling missing values ​​and standardizing. The specific steps are as follows: The received data is subjected to outlier elimination and missing value filling in by the on-board computing unit; Use the Z-score method to detect outliers and calculate the mean and standard deviation of all data points; Based on the mean and standard deviation of the data points, calculate the Z-score value of each data point, perform normal distribution on the Z-score values ​​of all data points, and set the threshold Z; When the Z-score value of a data point exceeds Z, the data point is considered to be an outlier and is removed; otherwise, the data point is retained; Fill the missing values ​​by linear interpolation, mark the position of the missing values, find the previous and next data points, calculate the interpolation of the missing values, and insert the calculated interpolation into the corresponding missing value position; Traverse all data points in each type of data to obtain the maximum and minimum values ​​of each type of data, and calculate the Min-Max standardized value for each data point in each type of data.

4. The method for managing electric energy of new energy vehicles based on machine learning as claimed in claim 3, characterized in that: The multimodal convolutional neural network is used to extract features from the processed multimodal data and fuse them into a comprehensive feature vector. The specific steps also include: Each input layer is followed by an independent convolutional layer, which uses a one-dimensional convolutional layer to extract the time series feature vectors of battery status data, driving behavior data, and environmental condition data; Add a pooling layer after each convolutional layer and select the largest eigenvalue in each pooling window as the output; The battery status data feature vector, driving behavior data feature vector, and environmental condition data feature vector extracted by the convolution layer and the pooling layer are concatenated along the feature dimension in chronological order to obtain a comprehensive feature vector.

5. The method for managing electric energy of new energy vehicles based on machine learning as claimed in claim 4, characterized in that: The specific steps of establishing the training environment model are as follows: Based on the training requirements of the reinforcement learning model, the training environment model is constructed by setting the state space, action space and reward function; Generate a series of actions based on the initial state vector, and record the corresponding states and rewards. Store the state vector, action, reward of each step, and the next state generated based on the previous action in an experience replay buffer.

6. The new energy vehicle power management method based on machine learning as claimed in claim 5, characterized in that: The power management strategy is input into the online learning framework of the adaptive learning rate to optimize the power management strategy in real time. The specific steps are as follows: Build an online learning framework based on the Adam optimizer to collect battery status data, driving behavior data, and environmental condition data in real time; Run the reinforcement learning model, process the real-time data and integrate the features into the reinforcement learning model, and introduce the Adam optimizer in the strategy update and value update; Using the chain rule, calculate the gradient of the policy network objective function with respect to the parameter θ and the gradient of the value network loss function with respect to the parameter φ; Updated by the Adam optimizer, the policy network parameters θ and the value network parameters φ are updated.

7. A new energy vehicle power management system based on machine learning, based on the new energy vehicle power management method based on machine learning according to any one of claims 1 to 6, characterized in that: Including data collection module, data processing module, feature fusion module, strategy generation module and strategy optimization module; The data collection module is used to collect multimodal data through various sensors and external data interfaces; The data processing module is used to process the multimodal data by removing outliers, filling missing values ​​and standardizing; The feature fusion module is used to extract features from the processed multimodal data using a multimodal convolutional neural network and fuse them into a comprehensive feature vector; The strategy generation module is used to input the comprehensive feature vector into the reinforcement learning model and generate the power management strategy through interactive learning with the environment; The strategy optimization module is used to input the power management strategy into an online learning framework with an adaptive learning rate to optimize the power management strategy in real time.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the new energy vehicle power management method based on machine learning described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the new energy vehicle power management method based on machine learning described in any one of claims 1 to 6 are implemented.

Images