Suspension energy distribution method and device, vehicle, equipment and readable storage medium

Abstract

The invention provides a suspension energy distribution method, device and equipment and a readable storage medium, and the method comprises the steps: obtaining a driving data set collected by a vehicle sensor, converting different types of driving data into feature vectors, and obtaining a driving data set corresponding to each type of driving data based on the feature vector corresponding to each type of driving data; determining a comprehensive feature vector corresponding to the driving data set, inputting the comprehensive feature vector into an embedded network with a preset network structure, obtaining an embedded vector after dimension reduction processing of the embedded network, inputting the embedded vector into a pre-trained strategy network, and obtaining a distribution proportion output by the strategy network. According to the invention, the energy utilization flexibility of the suspension can be improved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of vehicle technology, and in particular, to a suspension energy distribution method, device, vehicle, electronic device, and computer-readable storage medium. Background Art

[0002] With the development of vehicle manufacturing technology, the functions of vehicles are becoming more and more abundant. Current vehicles often need power supply to be used normally, for example, new energy vehicles need power supply to drive. However, the power of vehicles is limited, so the power source can be expanded.

[0003] At present, since the suspension of the vehicle will vibrate when the vehicle is driving, converting the mechanical energy of the suspension vibration into electrical energy can provide a certain amount of additional electrical energy, and the endurance of new energy vehicles can be longer.

[0004] However, the current solution is to only use the electrical energy generated by suspension vibration to improve vehicle endurance, and the use of suspension energy is relatively limited. Summary of the invention

[0005] In view of the above problems, the embodiments of the present disclosure are proposed to provide a suspension energy distribution method, device, vehicle, electronic device and computer-readable storage medium that overcome the above problems or at least partially solve the above problems.

[0006] In a first aspect, an embodiment of the present disclosure discloses a suspension energy distribution method, comprising:

[0007] Acquire a set of driving data collected by vehicle sensors; different vehicle sensors correspond to different types of driving data; wherein the vehicle is used to convert mechanical energy of suspension vibration into electrical energy;

[0008] Convert different types of driving data into feature vectors, and determine a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set;

[0009] Inputting the comprehensive feature vector into an embedding network having a preset network structure, obtaining an embedding vector after dimensionality reduction processing of the embedding network; the dimension of the embedding vector is determined by the network structure; the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector;

[0010] The embedding vector is input into a pre-trained policy network to obtain an allocation ratio output by the policy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

[0011] In a second aspect, the present disclosure discloses a suspension energy distribution device, including:

[0012] A data acquisition module is used to acquire a set of driving data collected by vehicle sensors; different vehicle sensors correspond to different types of driving data; wherein the vehicle is used to convert mechanical energy of suspension vibration into electrical energy;

[0013] A feature vector module, used to convert different types of driving data into feature vectors, and determine a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set;

[0014] A feature dimension module, used for inputting the comprehensive feature vector into an embedding network with a preset network structure to obtain an embedding vector after dimensionality reduction processing of the embedding network; the dimension of the embedding vector is determined by the network structure; the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector;

[0015] The allocation output module is used to input the embedding vector into a pre-trained strategy network to obtain the allocation ratio output by the strategy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

[0016] In a third aspect, the disclosed embodiment also discloses a vehicle, including: an energy recovery component, an energy storage component, and a distribution component;

[0017] The energy recovery component is used to convert the mechanical energy of vibration into electrical energy when the suspension vibrates;

[0018] The energy storage component is used to store the electrical energy for distribution by the distribution component;

[0019] The allocation component is used to obtain a set of driving data collected by vehicle sensors; different vehicle sensors correspond to different types of driving data;

[0020] The allocation component is used to convert different types of driving data into feature vectors, and determine a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set;

[0021] The allocation component is used to input the comprehensive feature vector into an embedding network with a preset network structure to obtain an embedding vector after the embedding network has been processed by dimensionality reduction; the dimension of the embedding vector is determined by the network structure; the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector;

[0022] The allocation component is used to input the embedding vector into a pre-trained policy network to obtain the allocation ratio output by the policy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

[0023] In a fourth aspect, an embodiment of the present disclosure further discloses an electronic device, comprising a processor and a memory, wherein the memory stores a program or instruction that can be executed on the processor, and when the program or instruction is executed by the processor, the steps of the suspension energy distribution method described in the first aspect are implemented.

[0024] In a fifth aspect, an embodiment of the present disclosure further discloses a computer-readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the suspension energy distribution method described in the first aspect are implemented.

[0025] In the disclosed embodiment, by acquiring a driving data set collected by a vehicle sensor, different types of driving data are converted into feature vectors, and based on the feature vectors corresponding to each type of driving data, a comprehensive feature vector corresponding to the driving data set is determined, the comprehensive feature vector is input into an embedding network with a preset network structure, an embedding vector after dimensionality reduction processing of the embedding network is obtained, and the embedding vector is input into a pre-trained strategy network to obtain the allocation ratio output by the strategy network. Different types of driving data can be collected, converted and reduced in dimension to obtain an embedding vector, thereby reducing data complexity; the strategy network outputs an allocation ratio based on the input embedding vector, and can accurately determine the power distribution of different vehicle control systems, thereby improving the diversity and accuracy of the suspension energy utilization process; and the allocation ratio can be flexibly adjusted according to the driving data collected in real time, thereby improving the flexibility and real-time performance of energy distribution; and since the vehicle control system can be allocated additional power, the vehicle control system has more sufficient power and can achieve a better working state, thereby improving the overall driving comfort of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a step diagram of a suspension energy distribution method provided by an embodiment of the present disclosure;

[0027] Figure 2 is an architectural diagram of energy distribution provided by an embodiment of the present disclosure;

[0028] Figure 3 is a step diagram of another suspension energy distribution method provided by an embodiment of the present disclosure;

[0029] Figure 4 is a workflow diagram of a meta-learning optimizer provided in an embodiment of the present disclosure;

[0030] Figure 5is a workflow diagram of an energy management unit provided by an embodiment of the present disclosure;

[0031] Figure 6 is a block diagram of a suspension energy distribution device provided by an embodiment of the present disclosure;

[0032] Figure 7 is a block diagram of a vehicle provided by an embodiment of the present disclosure;

[0033] Figure 8 It is a block diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0034] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0035] Figure 1 A step diagram of a suspension energy distribution method provided by an embodiment of the present disclosure is shown, and the method includes:

[0036] Step 101, obtaining a set of driving data collected by vehicle sensors; different vehicle sensors correspond to different types of driving data; wherein the vehicle is used to convert mechanical energy of suspension vibration into electrical energy.

[0037] In the disclosed embodiments, vehicles, especially new energy vehicles, have a high demand for electric energy, so the energy-feeding suspension technology can be used to improve the power supply. The energy-feeding suspension technology is an intelligent chassis technology that can convert the mechanical energy of the suspension vibration during vehicle driving into electric energy and reuse it. The energy-feeding suspension technology can not only improve the energy efficiency of the vehicle, but also help reduce emissions and improve driving performance.

[0038] Energy-feeding suspension technology can capture the energy generated during suspension vibration and temporarily store it for subsequent distribution. The process mainly includes: 1. Energy input. When the vehicle is driving, especially when passing through uneven roads, the suspension system will generate mechanical energy due to vibration. The suspension vibration energy includes the up and down movement energy of the shock absorber and the energy of the liquid flow in the hydraulic system; 2. Conversion mechanism. Based on the electromagnetic induction converter, the electromagnetic induction principle can be used to realize the conversion of mechanical energy to electrical energy. For example, when the suspension vibrates, the electromagnetic coil installed in the suspension system cuts the magnetic flux lines to generate current, realizing the conversion of mechanical energy to electrical energy; 3. Energy storage. The converted electrical energy can be temporarily stored in a high-efficiency small energy buffer unit. Supercapacitors can be used as small energy buffer units. Supercapacitors are characterized by fast charging and discharging speeds, which are suitable for fast response energy needs and ensure the immediate availability of energy.

[0039] The energy recovered by the energy-feeding suspension is quite different from the energy of the whole vehicle, mainly including: 1. The energy scale is relatively limited. Compared with the energy demand of the whole vehicle, the energy scale recovered by the energy-feeding suspension alone is relatively small, so it requires precise management and efficient utilization; 2. Volatility and randomness. The energy recovery of the energy-feeding suspension depends on the vibration encountered during vehicle driving. Due to different road conditions, the vibration is discontinuous and irregular in periodic changes, which leads to relatively unstable energy supply; 3. Instantaneity. Since the energy recovered by the energy-feeding suspension is stored in a small energy buffer unit, the capacity of the capacitor is limited compared to the battery of the whole vehicle. Therefore, the energy generated by the vibration often needs to be stored immediately and distributed in a short time, otherwise it may be wasted due to inability to store; 4. Low power and high density. Although it occurs frequently, the energy recovered for each vibration is low in magnitude, which is suitable for low-power or pulsed power supply scenarios.

[0040] Based on the characteristics of suspension energy, in order to reasonably distribute energy, it is necessary to collect driving data through vehicle sensors as the data basis for suspension energy distribution. Vehicle sensors and corresponding types of driving data can be:

[0041] 1. Vibration sensor: The vibration sensor is responsible for collecting the vibration frequency and amplitude of the suspension during vehicle driving. The vibration frequency and amplitude can be used to determine the vibration energy potential;

[0042] 2. Pressure sensor: The pressure sensor is used to monitor the pressure changes inside elastic elements (such as springs and hydraulic systems). The pressure changes can indirectly reflect the energy storage and release status;

[0043] 3. Speed ​​and acceleration sensors provide basic information about the vehicle's motion status, such as driving speed and acceleration, which can be used to determine the timing of energy recovery;

[0044] 4. Road condition recognition sensor, which can identify road conditions (such as smooth, bumpy, slippery) through radar or camera. Different road conditions have different effects on energy recovery;

[0045] 5. Environmental sensors, which can be temperature and humidity sensors. Changes in ambient temperature and humidity conditions can affect battery performance and energy requirements. For example, more energy may need to be reserved for heating in low-temperature environments.

[0046] 6. Other sensors can also detect driver behavior data through other sensors, such as pedal usage (acceleration, braking), steering behavior, etc., which can be used to determine driver habits and predict possible future energy consumption needs. Battery status monitoring data can also be collected, such as battery power, charging and discharging status, etc., to ensure that energy recovery does not exceed the battery's tolerance range, and to optimize battery charging strategies; and control system energy consumption monitoring data can be collected to monitor the real-time energy consumption of each control system (such as air conditioning, lighting, and infotainment systems) to provide a basis for dynamic allocation. The content of the collected driving data is not specifically limited here.

[0047] Step 102, convert different types of driving data into feature vectors, and determine a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set.

[0048] In the disclosed embodiment, different types of driving data are converted into feature vectors to facilitate subsequent analysis and processing of the driving data. Each type of driving data may correspond to its own feature vector. The feature vector may be a classification label corresponding to the driving data or a data combination, which is not limited here.

[0049] Based on the feature vector corresponding to each type of driving data, multiple feature vectors can be combined to obtain a comprehensive feature vector corresponding to the driving data set. For example, feature vector A = [a1, a2], feature vector B = [b1, b2], then the comprehensive feature vector can be F = [A, B]. It can be understood that a single feature vector can represent the current state and environment of the vehicle in a certain dimension, while a comprehensive feature vector represents the current state and environment of the vehicle in multiple dimensions.

[0050] Step 103, inputting the comprehensive feature vector into an embedding network with a preset network structure, and obtaining an embedding vector after dimensionality reduction processing of the embedding network; the dimension of the embedding vector is determined by the network structure; the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector.

[0051] In the disclosed embodiment, an embedding network is used to reduce the dimension of an input feature vector to facilitate subsequent analysis and processing. The embedding network reduces the dimension of the feature vector through an internal preset network structure, and obtains an embedded vector after the dimensionality reduction processing of the comprehensive feature vector. The dimension of the embedded vector is lower than the dimension of the comprehensive feature vector. The specific dimension of the embedded vector is determined by the network structure of the embedding network, and the specific network structure of the embedding network and the specific dimension of the embedded vector are not specifically limited here.

[0052] Step 104, input the embedding vector into a pre-trained policy network to obtain the allocation ratio output by the policy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

[0053] In the embodiment of the present disclosure, the embedding network is used to perform dimensionality reduction processing to obtain an embedding vector, and the embedding vector is input into a pre-trained strategy network to obtain a distribution ratio. The strategy network may include a strategy function π θ (a|e), policy function π θ It can be a conditional probability distribution function, a strategy function π θ The probability distribution of taking action a in a given state e is defined. The output layer of the policy network can be a Softmax layer so that the final output is a probability distribution and the sum of the distribution ratios is 1.

[0054] The given state e can be an embedding vector. Since the embedding vector is a vector obtained by reducing the dimensionality of the comprehensive feature vector that describes the vehicle state in multiple dimensions, the embedding vector can be used to describe the current vehicle state. Action a is the allocation ratio. Based on the embedding vector, the policy network can output the probability distribution corresponding to different vehicle control systems, that is, the allocation ratio. The allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

[0055] The policy network can calculate the unnormalized score of each vehicle control system and obtain the final allocation ratio through the Softmax layer. The formula is as follows:

[0056]

[0057] Where i represents the i-th vehicle control system, a i represents the proportion of electric energy corresponding to the i-th vehicle control system, j represents the total number of vehicle control systems, e is the embedding vector, θ is the parameter of the policy network, fi(e; θ) is the unnormalized score of the i-th vehicle control system, exp is the exponential function symbol, which represents the power operation of the natural exponent e, and N represents the number of vehicle control systems.

[0058] The vehicle control system may be a subsystem in the vehicle control system. The vehicle control system (subsystem) may include a power system, an electric auxiliary system, a cockpit system, a chassis system, and an intelligent assisted driving system. Among them, the power system may include a motor, the electric auxiliary system may include an air conditioning system, a seat heating / ventilation system, and a lighting system, the cockpit system may include an audio-visual entertainment system and a navigation system, the chassis system may include a start / stop system and an electric power steering system, and the intelligent assisted driving system may include a camera and a radar. The type and number of vehicle control systems are not specifically limited here.

[0059] According to the distribution ratio, the electric energy generated by the suspension vibration is distributed to the vehicle control system, which can enable the vehicle control system to work based on more sufficient electric energy, thereby improving the vehicle comfort. Different vehicle control systems have the following effects when receiving electric energy distribution:

[0060] Power system, which increases the driving range of new energy vehicles; air conditioning system, which can temporarily support the operation of the air conditioning system fan in summer / winter to assist cooling, especially when idling or driving at low speeds, reducing direct consumption of the battery; seat heating / ventilation system, which provides partial power for the seat heating or ventilation system under appropriate conditions to improve passenger comfort; lighting system, which provides power for interior lighting or lights, especially when parking at night or driving at low speeds, reducing dependence on the main battery; audio and video entertainment system, which provides power for the car audio and entertainment screen, extending the use time of the entertainment system without consuming the main energy of the vehicle; navigation system, which provides power for the navigation system's car computer, display screen, and voice; start / stop system, which provides additional power for the vehicle's automatic start / stop system, helping to start the engine more smoothly and efficiently and reduce fuel consumption; electric power steering system, which replenishes energy for the electric power steering system when driving at low speeds or parking, reducing the burden on the main battery; camera and radar system, which provides auxiliary power for low-power sensors and cameras in the vehicle's driving system to ensure that the functions are online and improve safety.

[0061] Figure 2 is an architectural diagram of energy distribution provided by an embodiment of the present disclosure; Figure 2 It includes G1 suspension, which generates mechanical energy through vibration; G2, energy recovery layer, which converts mechanical energy into electrical energy; G3, energy buffer unit, which stores electrical energy; G4 data acquisition layer; G5, energy management unit, which determines the distribution ratio based on the collected driving data; G6 energy distribution layer, which distributes electrical energy according to the distribution ratio.

[0062] In summary, in the embodiment of the present disclosure, by acquiring a set of driving data collected by a vehicle sensor, different types of driving data are converted into feature vectors, and based on the feature vectors corresponding to each type of driving data, a comprehensive feature vector corresponding to the driving data set is determined, the comprehensive feature vector is input into an embedding network with a preset network structure, an embedding vector after dimensionality reduction processing of the embedding network is obtained, and the embedding vector is input into a pre-trained strategy network to obtain the allocation ratio output by the strategy network. Different types of driving data can be collected, converted and reduced in dimension to obtain an embedding vector, thereby reducing data complexity; the strategy network outputs an allocation ratio based on the input embedding vector, and can accurately determine the power distribution of different vehicle control systems, thereby improving the diversity and accuracy of the suspension energy utilization process; and the allocation ratio can be flexibly adjusted according to the driving data collected in real time, thereby improving the flexibility and real-time performance of energy distribution; and since the vehicle control system can be allocated additional power, the vehicle control system has more sufficient power and can achieve a better working state, thereby improving the overall comfort of vehicle driving.

[0063] refer to Figure 3 , which shows a flow chart of the steps of a suspension energy distribution method provided by an embodiment of the present disclosure, the method comprising:

[0064] Step 301, obtaining a set of driving data collected by a vehicle sensor; different vehicle sensors correspond to different types of driving data; wherein the vehicle is used to convert mechanical energy of suspension vibration into electrical energy;

[0065] Step 302, converting different types of driving data into feature vectors, and determining a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set;

[0066] Step 303, inputting the comprehensive feature vector into an embedding network with a preset network structure, and obtaining an embedding vector after dimensionality reduction processing of the embedding network; the dimension of the embedding vector is determined by the network structure; and the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector;

[0067] Step 304, input the embedding vector into a pre-trained policy network to obtain the allocation ratio output by the policy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

[0068] The above steps 301-304 can refer to the above Figure 1 The contents of the embodiments will not be described in detail here.

[0069] Optionally, the step of converting different types of driving data into feature vectors, and determining a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data, comprises:

[0070] For different types of driving data, based on the data type of the driving data, determining a preset data processing operation corresponding to the data type;

[0071] Executing the preset data processing operation to convert the driving data to obtain a feature vector corresponding to the driving data;

[0072] Based on the feature vector corresponding to each type of driving data, a comprehensive feature vector corresponding to the driving data set is constructed.

[0073] In the disclosed embodiment, different types of driving data may be converted into vectors through different data processing operations. First, it is necessary to determine the preset data processing operation corresponding to the data type based on the data type of the driving data. It is understandable that the collection and reading of driving data can be based on a preset time interval Δt. Among them, if the time interval Δt is small, the reading is frequent, the computing power is consumed, but the timeliness of energy recovery adjustment is high; if the time interval Δt is large, the computing power is consumed, and the timeliness of energy recovery adjustment is low. The appropriate reading time interval Δt can be selected according to demand to balance cost and effect.

[0074] For example, for the suspension vibration data collected by the vibration sensor, the preset data processing operation can be: select several frequency components with the largest amplitude, and then perform fast Fourier transform. The data processing operation can refer to the following formula:

[0075] f vib,i =FFT(V sensor ) i

[0076] [N top ],i=1,2,...,N top

[0077] Among them, f vib,i is the vibration frequency eigenvector of the ith component among several frequency components with the largest amplitude, V sensor is the original data sequence of the vibration sensor within the time interval Δt, FFT stands for fast Fourier transform, [N top ] represents several frequency components with the largest amplitude, which can be set as needed [N top ] value.

[0078] For suspension vibration data, the amplitude characteristics can also be processed. The data processing operation can refer to the following formula:

[0079] A vib,i =RMS(V sensor ,f i ),i=1,2,...,N top

[0080] Among them, A vib,i represents the amplitude eigenvector, RMS(V sensor , f i ) represents the original data sequence V sensor At frequency f i The root mean square amplitude under top Indicates several frequencies.

[0081] For the pressure change data collected by the pressure sensor, the data processing operation can refer to the following formula:

[0082] ΔP presp =P max -P min

[0083]

[0084] Where Pmax and Pmin are the maximum and minimum pressure values ​​recorded by the pressure sensor, ΔP press is the difference between the maximum and minimum pressure values, Δt is the time interval, and Rate press is the pressure eigenvector.

[0085] For velocity and acceleration data, data processing operations can refer to the following formula:

[0086] V ppeed_acc =[v,a]

[0087] Where v is the average velocity within Δt, a is the average acceleration within Δt, V speed_acc are the velocity and acceleration eigenvectors.

[0088] For vehicle posture data, the data processing operation can refer to the following formula:

[0089] Θ attitude =[θ p ,θ r ]

[0090] Among them, θ p and θ r are the average pitch angle and average roll angle of the vehicle within the time interval Δt, Θ attitude is the posture feature vector.

[0091] For traffic data, the data processing operation can refer to the following formula:

[0092] c road =classify(I radar_cam )

[0093] Among them, classify is based on I radar_cam (radar or camera data) road condition feature extraction classification function, c road Represents the road condition feature vector after identification. It is understandable that when designing the feature extraction classification function, categories can be preset, such as smooth, potholes, roughness, etc. The finer the classification, the higher the training cost and duration, and the higher the degree of scene refinement of energy distribution management; the coarser the classification, the lower the training cost and duration, and the lower the degree of scene refinement of energy distribution management. Appropriate categories can be preset according to project requirements to balance training costs and effects.

[0094] For environmental condition data, data processing operations can refer to the following formula:

[0095] E env =[T,H]

[0096] Among them, Eenv is the environmental feature vector, T and H represent the current environmental temperature and humidity respectively.

[0097] For driver behavior data, two feature vectors can be included, refer to the following formula:

[0098] P pedal =[n accel ,n brake ]

[0099] Among them, n accel and n brake are the number of acceleration and braking times within the time interval Δt. pedal is the pedal usage eigenvector.

[0100]

[0101] in, is the average value of each steering angle in the time interval, σ φ is the standard deviation of the steering angles within the time interval, Φ steering is the steering eigenvector.

[0102] For system status and energy consumption data, two feature vectors can be included, refer to the following formula:

[0103] B battery =[b%,Discharge Rate]

[0104] Where b% is the remaining battery power percentage, and Discharge Rate is the current discharge rate.battery is the battery state feature vector.

[0105] E subsys =[E subsys1 ,E subsys2 ,...,E subsysn ]

[0106] Among them, Esubsysi is the instantaneous energy consumption of the i-th vehicle control system, E subsys is the system energy consumption characteristic vector.

[0107] All the above feature vectors are combined to form a comprehensive feature vector F, which is used for subsequent task embedding learning. The definition of the comprehensive feature vector F refers to the following formula:

[0108]

[0109] c road ,E env ,P pedal ,Φ steering ,B battery ,E subsya ]

[0110] Among them, the feature vectors of each dimension in the comprehensive feature vector F are not described in detail here.

[0111] In the implementation of the embodiments of the present disclosure, for each type of driving data, based on the data type of the driving data, a preset data processing operation corresponding to the data type is determined, the preset data processing operation is executed to convert the driving data, and a feature vector corresponding to the driving data is obtained. Based on the feature vector corresponding to each type of driving data, a comprehensive feature vector corresponding to the driving data set is constructed. The feature vectors corresponding to different types of driving data can be merged, which is convenient for simplifying subsequent data processing and improving the accuracy and efficiency of energy allocation.

[0112] Optionally, the network structure of the embedding network includes an input layer, at least one fully connected layer and an output layer, each layer having a plurality of nodes;

[0113] The step 303 of inputting the comprehensive feature vector into an embedding network having a preset network structure to obtain an embedding vector after the embedding network has been processed by dimensionality reduction includes:

[0114] Sub-step 3031, inputting the comprehensive feature vector into the input layer of the embedding network; the number of nodes in the input layer corresponds to the dimension of the comprehensive feature vector;

[0115] Sub-step 3032, obtaining a first output of the input layer, and inputting the first output into at least one fully connected layer, so as to reduce the dimension of the comprehensive feature vector through the nodes of at least one fully connected layer; the number of nodes in each layer of at least one fully connected layer is in a decreasing relationship;

[0116] Sub-step 3033, obtaining a second output of at least one fully connected layer, and inputting the second output into the output layer to obtain an embedding vector output by the output layer; the number of nodes in the output layer corresponds to the dimension of the embedding vector.

[0117] In the disclosed embodiment, the network structure of the embedding network includes an input layer, a fully connected layer, and an output layer, each layer includes multiple nodes, i.e., neurons, and the embedding network can learn nonlinear relationships between features. The number of nodes in the input layer can correspond to the dimension of the comprehensive feature vector, the number of nodes in the last fully connected layer can correspond to the embedding vector of the required specific dimension, and the last fully connected layer can be used as a bottleneck layer, which can limit the dimension of the feature vector.

[0118] In addition, each fully connected layer may also include an activation function, which may be a hyperbolic tangent function (Tanh), a rectified linear unit function (RELU), an exponential linear unit function (ELU), etc. The activation function may introduce nonlinearity, thereby being able to fit more complex functions.

[0119] For example, if the comprehensive feature vector F has N features and there are 3 fully connected layers, the input layer will have N nodes. The number of nodes in the first fully connected layer can be set to half of the number of nodes in the input layer, that is, N / 2, which can initially reduce the feature dimension. The number of nodes in the second fully connected layer can be further halved, that is, N / 4. The third fully connected layer, the bottleneck layer, is the last layer of the embedding network. The number of nodes in the bottleneck layer determines the final dimension of the embedding vector. The number of nodes in the bottleneck layer can be N / 8, and the number of nodes can be set to N / 8 or less according to the expected dimension of the final task embedding. For example, when N is 50, the bottleneck layer can have 5 to 10 nodes, and a 5 to 10-dimensional embedding vector can be generated in the end.

[0120] The output layer finally outputs a low-dimensional embedding vector. The embedding vector generated by the output layer is a K-dimensional vector, and the formula is as follows:

[0121] e=[e1,e2,...,e K ]

[0122] Among them, each e represents the projection of the input comprehensive feature vector on the corresponding dimension.

[0123] In the implementation of the embodiment of the present disclosure, the comprehensive feature vector is input into the input layer of the embedding network, and the dimension of the comprehensive feature vector is reduced through at least one fully connected layer and the nodes in the fully connected layer, and the embedding vector output by the output layer is obtained. The dimension of the comprehensive feature vector can be reduced, and the dimension of the obtained embedding vector is much smaller than the dimension of the comprehensive feature vector, so that redundant information can be reduced while capturing key features, thereby improving the simplicity and efficiency of data processing.

[0124] Optionally, the method further includes:

[0125] Step A1, constructing an expected reward function and an immediate reward function; the expected reward function is used to determine the expected value of the allocation ratio output by the initial strategy network based on the embedding vector; the immediate reward function is used to determine the effectiveness of the allocation ratio; the initial strategy network has initial parameters;

[0126] Step A2, constructing a gradient description function based on the expected reward function and the immediate reward function; the gradient description function is used to determine the update gradient of the expected reward function with respect to the initial parameters;

[0127] Step A3: determining an update gradient according to the gradient description function, and updating the initial parameters in the initial policy network based on the update gradient to obtain the pre-trained policy network.

[0128] In the disclosed embodiment, a policy network may be pre-constructed, and the policy network may be a parameterized policy function learned using deep learning technology, and the policy function may map an embedding vector to an energy allocation scheme, i.e., an allocation ratio.

[0129] We need to find a set of parameters θ that maximizes the expected return J(θ) given the embedding vector e. The expected return function is defined as:

[0130] J(θ)=E e～p(e)，a～πθ(·∣e)

[0131] Among them, J(θ) represents the mathematical expectation, E e～p(e)，a～πθ(·∣e) represents the expected value of all possible actions a given the embedding vector e for all possible task embedding vectors e and policies πθ. p(e) is the distribution of the embedding vector.

[0132] The immediate reward function is defined as:

[0133] [r(a,e)]

[0134] Among them, r(a,e) is the immediate reward function, which evaluates the effectiveness or effect of executing action a (allocation ratio) under the state corresponding to the embedding vector e.

[0135] Furthermore, policy gradient update is performed. In order to optimize J(θ), the policy gradient method can be used to update the parameters θ of the policy network. The definition of the gradient description function is as follows:

[0136]

[0137] in, represents the updated gradient of the expected reward function J(θ) with respect to the parameter θ, Represents strategy π θ (a|e) is the gradient of the log-likelihood with respect to the parameter θ, r(a,e) is the immediate reward function, where the log-likelihood is used to make the gradient calculation simpler.

[0138] Sampling e and a means obtaining the existing embedding vector e and the corresponding allocation ratio a. Based on e, a and the above gradient description function, the updated gradient is calculated. The gradient ascent method can be used to update θ. The formula is as follows:

[0139]

[0140] Among them, θ t is the parameter at time t, is the corresponding update gradient, α is the learning rate, and θ t+1 It is the updated parameter at time t+1, and finally the pre-trained strategy network is obtained.

[0141] In implementing the embodiments of the present disclosure, an expected reward function and an immediate reward function are constructed, a gradient description function is constructed based on the expected reward function and the immediate reward function, an update gradient is determined according to the gradient description function, and initial parameters in the initial policy network are updated based on the updated gradient to obtain a pre-trained policy network. By updating the parameters, a more accurate policy network can be obtained, thereby improving the accuracy of the allocation ratio.

[0142] Optionally, the method further includes:

[0143] Acquire current parameters of the policy network, and acquire a first data set; the first data set includes multiple groups of update data, each group of update data includes an embedding vector and a distribution ratio;

[0144] Selecting a preset number of groups of target update data from the first data set, and constructing a second data set based on the target update data;

[0145] For each update data in the first data set, performing a preset number of gradient descent updates on the policy network based on the second data set to obtain a fast adaptation parameter of the update data;

[0146] The current parameter is updated based on the fast adaptation parameter of each updated data in the first data set.

[0147] In the disclosed embodiment, in order to enhance the rapid adaptability of the policy network when facing unfamiliar scenarios, a meta-learning optimizer may be added to the policy network to adjust the parameters of the policy network, and the current parameters θ of the policy network may be used as the starting point for adjustment.

[0148] A large number of virtual tasks can be extracted from historical vehicle data and used for meta-learning training. Each task includes an embedding vector e and a corresponding real or simulated distribution ratio a. A first data set is constructed based on these virtual tasks, and each task is a set of updated data.

[0149] A preset number of target update data sets are selected from the first data set, and the target update data forms a second data set. For each extracted task Ti, use Perform several gradient descent updates on the policy network to obtain the fast adaptation parameters to complete the inner loop update. The formula is as follows:

[0150]

[0151] Among them, θi′ is the fast adaptation parameter, θ is the current parameter, α is the learning rate, Denotes the expected return function J(θ) relative to The gradient of , k is the number of gradient descent updates.

[0152] For each set of update data in the first data set, that is, each task, the fast adaptation parameter is calculated, and the current parameter θ of the strategy network is updated through the fast adaptation parameter of each update data to complete the parameter movement. The formula is as follows:

[0153]

[0154] The above formula indicates that the current parameter θ is updated by the average value of the fast adaptation parameter θi′ corresponding to each set of update data in the first data set, that is, parameter shift is performed, β is the learning rate, and m is the number of update data in the first data set.

[0155] The above inner loop update and parameter movement steps can be repeated until the policy network shows good fast adaptation performance in the new scene.

[0156] Figure 4is a workflow diagram of a meta-learning optimizer provided in an embodiment of the present disclosure;

[0157] Step 401, task preparation, extracting a virtual task, i.e. updating data to obtain a first data set;

[0158] Step 402, initialization, obtaining the current parameter θ of the policy network as the starting point for adjustment;

[0159] Step 403, task extraction, extracting a small amount of data from the first data set as a support set

[0160] Step 404, inner loop update: For each extracted task, use Perform several gradient descent updates on the network to obtain fast adaptation parameters;

[0161] Step 405, parameter movement: calculate the average value of the fast adaptation parameters on all tasks and update the global parameters.

[0162] The embodiment of the present disclosure is implemented by acquiring a first data set, selecting a preset number of target update data in the first data set, and constructing a second data set based on the target update data. For each update data in the first data set, the policy network is updated with a preset number of gradient descents based on the second data set to obtain the fast adaptation parameters of the update data; based on the fast adaptation parameters of each update data, the current parameters are updated. The meta-learning optimizer is implemented through the above steps, and the parameters are adjusted through the meta-learning optimizer, so that the policy network can quickly adjust its energy allocation strategy when facing new driving scenarios, which is suitable for handling dynamic and uncertain environments, and enhances the adaptability and generalization ability of the policy network.

[0163] Optionally, the allocation ratio includes different proportions of electric energy and corresponding system information; the method further includes:

[0164] Step 305, determining a target vehicle control system corresponding to the system information according to the system information corresponding to each electric energy proportion in the allocation ratio;

[0165] Step 306: Allocate electric energy corresponding to the electric energy proportion to the target vehicle control system according to the electric energy proportion.

[0166] In the embodiment of the present disclosure, the allocation ratio includes different proportions of electric energy and corresponding system information, and the system information may be an identification, index, and other information of the vehicle control system.

[0167] Different vehicle control systems (subsystems) can be numbered to identify different subsystems of the vehicle and their subdivided functions. For example, the major subsystem is represented by: "P" + number, such as P1, P2, etc., representing the highest level of system classification, such as power system, power auxiliary system, etc. Under each major subsystem, it can be further subdivided into specific secondary classifications, which are indicated by adding a hyphen and another number after the major subsystem number, such as "P2-1" represents the air conditioning system in the power auxiliary system, "P2-2" represents the seat heating / ventilation system, and so on, as shown in the following table:

[0168]

[0169]

[0170] Based on the numbering process shown in the table above, space can be reserved for adding new subsystems or modifying the existing system structure. Whether adding a new major subsystem or adding new functions under an existing major category, you only need to follow the same numbering process, without the need to reconstruct the database or code structure on a large scale, which has high flexibility and scalability. In the database, the above numbering process helps to create indexes, speed up queries, facilitate database operations when processing large data sets, and simplify the classification storage and retrieval process of data.

[0171] After obtaining the allocation ratio, the target vehicle control system corresponding to the system information is determined according to the system information corresponding to each power proportion in the allocation ratio, such as the number, and then the power corresponding to its own power proportion is allocated to the target vehicle control system.

[0172] By implementing the embodiments of the present disclosure, a target vehicle control system corresponding to the system information is determined based on the system information corresponding to each electric energy proportion in the allocation ratio; and electric energy corresponding to the electric energy proportion is allocated to the target vehicle control system based on the electric energy proportion. This enables fast and accurate electric energy distribution according to the allocation ratio, thereby improving the efficiency and accuracy of electric energy distribution.

[0173] Under actual deployment conditions, the policy network can have the generalization capabilities of pre-training, and can also be fine-tuned online to adapt to driving conditions and environmental changes that were not fully foreseen during the training phase. The fine-tuning process usually involves real-time data collection, analysis, and immediate adjustment of the policy to ensure that the vehicle can still operate efficiently and safely in various unknown or unexpected situations. The following methods can be adopted during the implementation process:

[0174] 1. Performance monitoring: Continuously monitor key performance indicators such as energy recovery efficiency, driving comfort and safety, and use these indicators as feedback signals to determine the effectiveness of the current strategy;

[0175] 2. Feedback mechanism: Allow users to provide feedback on their satisfaction with the energy allocation strategy through the interface. The system will fine-tune the algorithm based on the feedback to achieve more personalized services.

[0176] 3. Adaptive learning rate: Adaptive learning rate strategies such as AdaGrad, RMSProp or Adam can be used in the feedback loop. These optimization algorithms can dynamically adjust the learning rate according to the historical gradient, so as to maintain efficient updates when facing learning tasks of different difficulty.

[0177] Separate hardware or software can be designed to implement the method steps of the present invention, for example, an energy management unit can be designed, and the energy management unit can also include a data acquisition layer, a data preprocessing and feature extraction layer, a task embedding generator, and a strategy network (including a meta-learning optimizer).

[0178] Figure 5 is a workflow diagram of an energy management unit provided by an embodiment of the present disclosure;

[0179] Step 501, data collection layer, collecting driving data from vehicle sensors;

[0180] Step 502, data preprocessing and feature extraction layer, processing the driving data into a comprehensive feature vector;

[0181] Step 503, the task embedding generator performs dimensionality reduction processing and outputs an embedding vector;

[0182] Step 504: the strategy network outputs the energy allocation result, ie, the allocation ratio.

[0183] In summary, the embodiments of the present invention have the following effects:

[0184] 1. Significantly improve energy utilization efficiency: By introducing task embedding and meta-learning optimization strategies, this technical solution can achieve precise control and dynamic allocation of energy recovery by the energy-feeding suspension. It avoids the limitations of fixed allocation methods and enables energy to be diversified according to the actual driving status and needs of the vehicle, thereby significantly improving energy utilization efficiency and extending the driving range of new energy vehicles.

[0185] 2. Expanding the application of energy: The recovered energy is not limited to increasing the driving range, but can also be intelligently distributed according to the actual needs of various systems inside the vehicle, such as headlight lighting, air conditioning adjustment, entertainment system power supply, etc. The diversified energy application methods further improve the overall energy utilization efficiency and driving comfort of the vehicle.

[0186] 3. Enhance system adaptability and flexibility: The adaptive mechanism enables real-time perception and adaptation to different driving scenarios and user preferences, and dynamically adjusts the energy allocation strategy. Whether it is urban roads, highways or complex and changeable off-road environments, the system can make accurate energy management decisions to ensure that the vehicle can maintain optimal performance in different scenarios.

[0187] 4. Improve user experience: Through intelligent and efficient energy distribution, the energy consumption of new energy vehicles is effectively reduced, the charging frequency is reduced, and a more convenient user experience is brought to users. At the same time, energy applications in multiple fields also improve the comfort and intelligence of vehicles.

[0188] Figure 6 A suspension energy distribution device is provided in an embodiment of the present disclosure. The device 60 includes:

[0189] The data acquisition module 601 is used to acquire a set of driving data collected by vehicle sensors; different vehicle sensors correspond to different types of driving data; wherein the vehicle is used to convert mechanical energy of suspension vibration into electrical energy;

[0190] A feature vector module 602 is used to convert different types of driving data into feature vectors, and determine a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set;

[0191] A feature dimension module 603 is used to input the comprehensive feature vector into an embedding network with a preset network structure to obtain an embedding vector after the embedding network has been processed by dimensionality reduction; the dimension of the embedding vector is determined by the network structure; and the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector;

[0192] The allocation output module 604 is used to input the embedding vector into the pre-trained strategy network to obtain the allocation ratio output by the strategy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

[0193] Optionally, the device further comprises:

[0194] The first function construction module is used to construct an expected return function and an immediate reward function; the expected return function is used to determine the expected value of the allocation ratio output by the initial strategy network based on the embedding vector; the immediate reward function is used to determine the effectiveness of the allocation ratio; the initial strategy network has initial parameters;

[0195] A second function construction module is used to construct a gradient description function based on the expected reward function and the immediate reward function; the gradient description function is used to determine the update gradient of the expected reward function with respect to the initial parameters;

[0196] A gradient update module is used to determine an update gradient according to the gradient description function, and to update the initial parameters in the initial policy network based on the update gradient to obtain the pre-trained policy network.

[0197] Optionally, the device further comprises:

[0198] A first set module, used to obtain current parameters of the policy network and obtain a first data set; the first data set includes multiple groups of update data, each group of update data includes an embedding vector and a distribution ratio;

[0199] A second data set module, configured to select a preset number of groups of target update data from the first data set, and construct a second data set based on the target update data;

[0200] An adaptation parameter module, configured to perform a preset number of gradient descent updates on the policy network based on the second data set for each update data in the first data set, to obtain a fast adaptation parameter of the update data;

[0201] A parameter updating module is used to update the current parameter based on the fast adaptation parameter of each updated data in the first data set.

[0202] Optionally, the network structure of the embedding network includes an input layer, at least one fully connected layer and an output layer, each layer having a plurality of nodes; and a feature dimension module, including:

[0203] A vector input submodule, used to input the comprehensive feature vector into the input layer of the embedding network; the number of nodes in the input layer corresponds to the dimension of the comprehensive feature vector;

[0204] a dimensionality reduction submodule, configured to obtain a first output of the input layer, and input the first output to at least one fully connected layer, so as to reduce the dimension of the comprehensive feature vector through nodes of at least one fully connected layer; the number of nodes in each layer of at least one fully connected layer is in a decreasing relationship;

[0205] A vector output submodule is used to obtain a second output of at least one fully connected layer and input the second output to the output layer to obtain an embedded vector output by the output layer; the number of nodes in the output layer corresponds to the dimension of the embedded vector.

[0206] Optional, feature vector modules include:

[0207] A type operation submodule, for determining, for different types of driving data, a preset data processing operation corresponding to the data type based on the data type of the driving data;

[0208] A data processing submodule, configured to execute the preset data processing operation to convert the driving data to obtain a feature vector corresponding to the driving data;

[0209] The vector merging submodule is used to construct a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data.

[0210] Optionally, the allocation ratio includes different electric energy proportions and corresponding system information, and the device further includes:

[0211] A system information module, used to determine a target vehicle control system corresponding to the system information according to the system information corresponding to each electric energy proportion in the allocation ratio;

[0212] An energy distribution module is used to distribute electric energy corresponding to the electric energy proportion to the target vehicle control system according to the electric energy proportion.

[0213] In summary, in the embodiment of the present disclosure, by acquiring a set of driving data collected by a vehicle sensor, different types of driving data are converted into feature vectors, and based on the feature vectors corresponding to each type of driving data, a comprehensive feature vector corresponding to the driving data set is determined, the comprehensive feature vector is input into an embedding network with a preset network structure, an embedding vector after dimensionality reduction processing of the embedding network is obtained, and the embedding vector is input into a pre-trained strategy network to obtain the allocation ratio output by the strategy network. Different types of driving data can be collected, converted and reduced in dimension to obtain an embedding vector, thereby reducing data complexity; the strategy network outputs an allocation ratio based on the input embedding vector, and can accurately determine the power distribution of different vehicle control systems, thereby improving the diversity and accuracy of the suspension energy utilization process; and the allocation ratio can be flexibly adjusted according to the driving data collected in real time, thereby improving the flexibility and real-time performance of energy distribution; and since the vehicle control system can be allocated additional power, the vehicle control system has more sufficient power and can achieve a better working state, thereby improving the overall comfort of vehicle driving.

[0214] The present application also provides a vehicle, such as Figure 7 As shown, the vehicle 70 includes: an energy recovery component 701, an energy storage component 702, and a distribution component 703;

[0215] The energy recovery component 701 is used to convert vibration mechanical energy into electrical energy when the suspension vibrates;

[0216] The energy storage component 702 is used to store the electrical energy for distribution by the distribution component;

[0217] The distribution component 703 is used to execute the steps of the suspension energy distribution method in the method embodiment.

[0218] The present application also provides an electronic device, such as Figure 8 As shown, it includes a processor 1001 , a communication interface 1002 , a memory 1003 and a communication bus 1004 , wherein the processor 1001 , the communication interface 1002 , and the memory 1003 communicate with each other via the communication bus 1004 .

[0219] The memory 1003 is used to store computer programs.

[0220] When the processor 1001 is used to execute the program stored in the memory 1003, the steps in the above-mentioned suspension energy distribution method are implemented, which will not be repeated here.

[0221] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0222] The communication interface is used for communication between the above electronic device and other devices.

[0223] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0224] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0225] In another embodiment provided in the present application, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the suspension energy distribution method described in the above embodiment.

[0226] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute the suspension energy distribution method described in the above embodiment.

[0227] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0228] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0229] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. For the embodiments of the device, electronic device, computer-readable storage medium and computer program products containing instructions, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0230] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the protection scope of the present application.

Claims

1. A suspension energy distribution method, characterized in that: The method comprises: Acquire a set of driving data collected by vehicle sensors; different vehicle sensors correspond to different types of driving data; wherein the vehicle is used to convert mechanical energy of suspension vibration into electrical energy; Convert different types of driving data into feature vectors, and determine a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set; Inputting the comprehensive feature vector into an embedding network having a preset network structure, obtaining an embedding vector after dimensionality reduction processing of the embedding network; the dimension of the embedding vector is determined by the network structure; the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector; The embedding vector is input into a pre-trained policy network to obtain an allocation ratio output by the policy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

2. The method according to claim 1, characterized in that The method further comprises: Constructing an expected return function and an immediate reward function; the expected return function is used to determine the expected value of the allocation ratio output by the initial strategy network based on the embedding vector; the immediate reward function is used to determine the effectiveness of the allocation ratio; the initial strategy network has initial parameters; Based on the expected reward function and the immediate reward function, construct a gradient description function; the gradient description function is used to determine the update gradient of the expected reward function with respect to the initial parameters; An update gradient is determined according to the gradient description function, and the initial parameters in the initial policy network are updated based on the update gradient to obtain the pre-trained policy network.

3. The method according to claim 1, characterized in that The method further comprises: Acquire current parameters of the policy network, and acquire a first data set; the first data set includes multiple groups of update data, each group of update data includes an embedding vector and a distribution ratio; Selecting a preset number of groups of target update data from the first data set, and constructing a second data set based on the target update data; For each update data in the first data set, performing a preset number of gradient descent updates on the policy network based on the second data set to obtain a fast adaptation parameter of the update data; The current parameter is updated based on the fast adaptation parameter of each updated data in the first data set.

4. The method according to claim 1, characterized in that: The network structure of the embedding network includes an input layer, at least one fully connected layer and an output layer, each layer having a plurality of nodes; The step of inputting the comprehensive feature vector into an embedding network having a preset network structure to obtain an embedding vector after dimensionality reduction processing of the embedding network comprises: Inputting the comprehensive feature vector into the input layer of the embedding network; the number of nodes in the input layer corresponds to the dimension of the comprehensive feature vector; Obtaining a first output of the input layer, and inputting the first output into at least one fully connected layer, so as to reduce the dimension of the comprehensive feature vector through nodes of at least one fully connected layer; the number of nodes in each layer of at least one fully connected layer has a decreasing relationship; Obtain a second output of at least one fully connected layer, and input the second output into the output layer to obtain an embedding vector output by the output layer; the number of nodes in the output layer corresponds to the dimension of the embedding vector.

5. The method according to claim 1, characterized in that: The step of converting different types of driving data into feature vectors and determining a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data comprises: For different types of driving data, based on the data type of the driving data, determining a preset data processing operation corresponding to the data type; Executing the preset data processing operation to convert the driving data to obtain a feature vector corresponding to the driving data; Based on the feature vector corresponding to each type of driving data, a comprehensive feature vector corresponding to the driving data set is constructed.

6. The method according to claim 1, characterized in that The allocation ratio includes different proportions of electric energy and corresponding system information; the method also includes: Determining a target vehicle control system corresponding to the system information according to the system information corresponding to each electric energy proportion in the allocation ratio; According to the electric energy proportion, electric energy corresponding to the electric energy proportion is allocated to the target vehicle control system.

7. A suspension energy distribution device, characterized in that: The device comprises: A data acquisition module is used to acquire a set of driving data collected by vehicle sensors; different vehicle sensors correspond to different types of driving data; wherein the vehicle is used to convert mechanical energy of suspension vibration into electrical energy; A feature vector module, used to convert different types of driving data into feature vectors, and determine a comprehensive feature vector corresponding to the driving data set based on the feature vector corresponding to each type of driving data; the dimension of the comprehensive feature vector is determined by the data type of the driving data set; A feature dimension module, used for inputting the comprehensive feature vector into an embedding network with a preset network structure to obtain an embedding vector after dimensionality reduction processing of the embedding network; the dimension of the embedding vector is determined by the network structure; the dimension of the embedding vector is smaller than the dimension of the comprehensive feature vector; The allocation output module is used to input the embedding vector into a pre-trained strategy network to obtain the allocation ratio output by the strategy network; the allocation ratio represents the proportion of electric energy corresponding to different vehicle control systems.

8. A vehicle, characterized in that: The vehicle comprises: an energy recovery component, an energy storage component, and a distribution component; The energy recovery component is used to convert the mechanical energy of vibration into electrical energy when the suspension vibrates; The energy storage component is used to store the electrical energy for distribution by the distribution component; The distribution component is used to execute the steps in the suspension energy distribution method described in any one of claims 1 to 6.

9. An electronic device, characterized in that: include: A processor, a communication interface, a memory and a communication bus; wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor is used to implement the steps in the suspension energy distribution method as described in any one of claims 1 to 6 when executing the program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the suspension energy distribution method according to any one of claims 1 to 6 are implemented.

Images