Automobile load prediction method and device and related equipment
The three-layer deep learning model predicts the car load and combines charging information to solve the problem of neglecting travel changes in the existing technology, which improves the accuracy of load prediction.
Patent Information
- Application Number
- CN202410995074.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-26
AI Technical Summary
The vehicle load prediction method in the prior art assumes that users' travel behavior is consistent, ignores changes in working days and holidays, resulting in low prediction accuracy.
A three-layer deep learning model is adopted, including improved LSTM, improved BP neural network and random forest algorithm, and predicts the trip chain type, target time period and battery life distance, and generates charging loads based on charging information.
It improves the accuracy of car load prediction, takes into account the timing and daily changes of users' travel habits, and enhances the accuracy of charging needs.
Smart Images

Figure CN119026936A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of artificial intelligence, and in particular, to a method, device and related equipment for predicting vehicle load. Background Art
[0002] The concentration effect of electric vehicle load is getting bigger and bigger. Electric vehicle charging load has become an important load demand. Accurate prediction of electric vehicle load demand is the key to controlling electric vehicle charging and improving power system operation and performance. The load forecasting method in the prior art is mainly based on trip chain and Monte Carlo simulation (MCS). However, this method assumes that the user's daily travel behavior is consistent, and the travel parameters of the next day are fixed with the historical travel parameters, ignoring the travel changes on weekdays, holidays, etc., resulting in the problem of low accuracy in predicting vehicle load. Summary of the invention
[0003] The embodiments of the present application provide a method, device and related equipment for predicting vehicle charging load to solve the problem of low accuracy in predicting vehicle load in the prior art.
[0004] To solve the above problems, this application is implemented as follows:
[0005] In a first aspect, an embodiment of the present application provides a method for predicting a vehicle charging load, the method comprising:
[0006] During the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction, and the trip chain type of the current trip is obtained, and the trip chain type is used to indicate the starting position and the end position of the current trip;
[0007] Input the trip chain type into the second deep learning model to predict the next stop information of the current trip, and obtain a target time period, wherein the time starting point of the target time period is the start time point of the next stop, and the time ending point of the target time period is the end time point of the next stop. The second deep learning model is a deep learning model generated by improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation;
[0008] Inputting the target time period into a third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and obtaining the cruising distance;
[0009] The charging load of the current trip is generated according to the cruising distance and the charging information.
[0010] Optionally, during the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction to obtain the trip chain type of the current trip, including:
[0011] Get the type of the trip chain of the previous trip of the current trip and the end time of the previous trip;
[0012] generating a cell state of the previous trip according to the type of the trip chain of the previous trip and the end time point of the previous trip, wherein the cell state is used to indicate a driving state parameter of the vehicle to be predicted during the previous trip;
[0013] The type of the trip chain of the previous trip, the start time point of the current trip and the cell state of the previous trip are input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
[0014] Optionally, the inputting the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip into the first deep learning model for prediction to obtain the type of the trip chain of the current trip includes:
[0015] Performing weighted calculation on the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip according to preset weight information to obtain forgetting information, wherein the preset weight information includes a first weight value, a second weight value, and a third weight value, wherein the first weight value corresponds to the type of the trip chain of the previous trip, the second weight value corresponds to the start time point of the current trip, and the third weight value corresponds to the cell state of the previous trip, and the forgetting information is used to indicate the influence of the previous trip on the current trip;
[0016] Updating the forgetting information according to preset bias information to obtain the target forgetting information;
[0017] The cell state of the previous trip is updated according to the target forgetting information to obtain the cell state of the current trip;
[0018] The cell state of the current trip is input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
[0019] Optionally, the second deep learning model is obtained by updating a preset second deep learning model through target parameter information, and the preset second deep learning model is constructed based on initial parameter information, and the target parameter information is obtained by updating the initial parameter information based on an improved BP algorithm and a back propagation algorithm.
[0020] Optionally, before inputting the target time period into the third deep learning model to predict the remaining cruising distance of the vehicle to be predicted and obtaining the cruising distance, the method further includes:
[0021] Acquire a training sample set, the training sample set comprising a plurality of training samples, the plurality of training samples corresponding one to one with a plurality of historical trips, wherein the driving distances corresponding to the plurality of historical trips are different;
[0022] Screening the multiple training samples according to a random forest algorithm to determine multiple target training samples;
[0023] A preset third deep learning model is trained based on the multiple target training samples to obtain the third deep learning model.
[0024] Optionally, generating a charging load for a current trip according to the cruising distance and the charging information includes:
[0025] Generating a state of charge according to the cruising distance, the state of charge being used to indicate remaining power information of a vehicle battery;
[0026] Acquiring error information, wherein the error information includes weather parameters and traffic parameters in the current trip;
[0027] generating a charging power curve based on the state of charge, the error information and the charging information, wherein the charging power curve is used to indicate a charging state of a user's car;
[0028] The charging load of the current trip is generated based on the difference information between the charging power curve and the preset charging power curve, the preset charging power curve is the predicted charging power curve of the current trip, the charging information is used to indicate the charging amount of the vehicle to be predicted in the current trip, and the charging load is used to indicate the charging power of the vehicle to be predicted in the current trip.
[0029] In a second aspect, an embodiment of the present application provides a vehicle load prediction device, the device comprising:
[0030] A first prediction module is used to input the starting time point of the current trip of the vehicle to be predicted into the first deep learning model for prediction during the driving process of the vehicle to be predicted, and obtain the trip chain type of the current trip, wherein the trip chain type is used to indicate the starting position and the end position of the current trip;
[0031] A second prediction module is used to predict the information to obtain a target time period, wherein the starting time of the target time period is the starting time point of the next parking, and the ending time point of the target time period is the ending time point of the next parking, and the second deep learning model is a deep learning model generated by improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation;
[0032] A third prediction module, used for inputting the target time period into a third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and obtaining the cruising distance;
[0033] A generation module is used to generate the charging load of the current trip according to the cruising range and charging information.
[0034] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; the processor is used to read the program in the memory to implement the steps of the method described in the first aspect above.
[0035] In a fourth aspect, an embodiment of the present application further provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method described in the first aspect above.
[0036] In a fifth aspect, an embodiment of the present application further provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the steps in the method described in the first aspect.
[0037] The present application provides a method, device and related equipment for predicting automobile load, the method comprising: during the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction, and the trip chain type of the current trip is obtained, and the trip chain type is used to indicate the starting position and the end position of the current trip; the trip chain type is input into the second deep learning model to predict the next parking information of the current trip, and the target time period is obtained, the time starting point of the target time period is the starting time point of the next parking, and the time end point of the target time period is the end time point of the next parking, the second deep learning model is a deep learning model generated after improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation; the target time period is input into the third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and the cruising distance is obtained; the charging load of the current trip is generated according to the cruising distance and the charging information. The present application predicts the trip chain type, target time period and cruising distance of the current trip through three deep learning models, and determines the charging load of the vehicle to be predicted by combining the charging information, thereby improving the accuracy of predicting the automobile load. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0039] Figure 1 It is a flow chart of a method for predicting vehicle load provided in an embodiment of the present application;
[0040] Figure 2 It is a schematic diagram of the structure of a three-layer deep learning model provided in an embodiment of the present application;
[0041] Figure 3 It is a schematic diagram of the principle of the second deep learning model provided in an embodiment of the present application;
[0042] Figure 4 is a flow chart of a second deep learning model provided in an embodiment of the present application;
[0043] Figure 5 is a schematic diagram of the attribution function of the road condition and temperature provided in an embodiment of the present application;
[0044] Figure 6 It is a schematic diagram of the membership function of power consumption per kilometer provided in an embodiment of the present application;
[0045] Figure 7It is a structural schematic diagram of a vehicle load prediction device provided in an embodiment of the present application;
[0046] Figure 8 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0048] The terms "first", "second" etc. in the embodiments of the present application are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. In addition, the terms "include" and "have" and any variation thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment. In addition, "and / or" is used in the present application to represent at least one of connected objects, such as A and / or B and / or C, representing the inclusion of separate A, separate B, separate C, and A and B all exist, B and C all exist, A and C all exist, and 7 situations that A, B and C all exist.
[0049] See also Figure 1 , Figure 1 FIG. 1 is a flow chart of a method for predicting vehicle load provided in an embodiment of the present application. Figure 1 As shown, the vehicle load prediction method may include the following steps:
[0050] Step 101: During the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction to obtain the trip chain type of the current trip, and the trip chain type is used to indicate the starting position and the end position of the current trip.
[0051] In this embodiment, load prediction is performed on the vehicle to be predicted driven by the user during driving, wherein, Figure 2 As shown, the present application predicts vehicle load through a three-layer deep learning model, wherein the first deep learning model is used to predict the trip chain type of the current trip according to the start time point of the current trip of the vehicle to be predicted, wherein the trip chain type can be used to represent the starting point and end point of the trip.
[0052] Specifically, there is a strong correlation between user itinerary chains and travel activities, and the previous itinerary will affect subsequent travel. Therefore, the first deep learning model can use the improved LSTM algorithm to predict the type of itinerary chain.
[0053] Step 102: Input the trip chain type into the second deep learning model to predict the next stop information of the current trip and obtain a target time period. The start time of the target time period is the start time of the next stop, and the end time of the target time period is the end time of the next stop. The second deep learning model is a deep learning model generated by improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation.
[0054] In this embodiment, the second deep learning model uses a BP neural network to predict the start and end time of the trip and generate a target time period. Considering that the traditional BP neural network model is prone to fall into local optimality and lead to training failure when solving nonlinear problems, an improved BP algorithm is introduced to help BP optimize the global search and speed up the solution. In addition, this application does not use the Monte Carlo algorithm in the prior art. Although the traditional Monte Carlo algorithm-based prediction simulates the travel behavior of the car owner, it assumes that the user's daily travel behavior is consistent, and the travel parameters of the second day are fixed with the historical travel parameters, ignoring the travel changes on weekdays, holidays, etc., which will lead to errors in the prediction results.
[0055] Step 103: Input the target time period into a third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and obtain the cruising distance.
[0056] In this embodiment, taking into account the differences in users' driving habits, the Random Forest (RF) algorithm is used as the third deep learning model of the model. By constructing multiple decision trees to predict the user's driving distance, the spatiotemporal characteristics of the trip chain are obtained.
[0057] Step 104: Generate a charging load for the current trip based on the cruising distance and charging information.
[0058] In this embodiment, the charging load of the current trip is generated based on the determined cruising range and the charging information of whether the car is charged, wherein the user can charge during the current trip, and the charging load of the car is predicted by the charging information to determine the charging load of the user's current trip. Specifically, the charging load is used to represent the predicted charging power of the vehicle in the current trip.
[0059] This application is based on the strong temporal sequence between users' daily travel activities, uses the concept of travel chain to perform feature engineering on travel data, extracts the spatiotemporal characteristics of users' travel, describes users' daily travel habits, and lays the foundation for analyzing charging needs. Since battery power consumption is difficult to accurately predict, the power consumption per kilometer is obtained through fuzzy calculation methods.
[0060] The present application provides a method for predicting automobile load, the method comprising: during the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction, and the trip chain type of the current trip is obtained, and the trip chain type is used to indicate the starting position and the end position of the current trip; the trip chain type is input into the second deep learning model to predict the next parking information of the current trip, and the target time period is obtained, the time starting point of the target time period is the starting time point of the next parking, and the time end point of the target time period is the end time point of the next parking, the second deep learning model is a deep learning model generated after improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation; the target time period is input into the third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and the cruising distance is obtained; the charging load of the current trip is generated according to the cruising distance and the charging information. The present application predicts the trip chain type, target time period and cruising distance of the current trip through three deep learning models, and determines the charging load of the vehicle to be predicted by combining the charging information, thereby improving the accuracy of predicting the automobile load.
[0061] In some feasible implementations, optionally, during the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction to obtain the trip chain type of the current trip, including:
[0062] Get the type of the trip chain of the previous trip of the current trip and the end time of the previous trip;
[0063] generating a cell state of the previous trip according to the type of the trip chain of the previous trip and the end time point of the previous trip, wherein the cell state is used to indicate a driving state parameter of the vehicle to be predicted during the previous trip;
[0064] The type of the trip chain of the previous trip, the start time point of the current trip and the cell state of the previous trip are input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
[0065] In this embodiment, the first deep learning model predicts the type of itinerary chain by using Bi+LSTM. The itinerary chain type prediction in the prior art is based on the Markov chain model. The LSTM model can reflect the impact of historical travel activities on the future. The various itineraries of a user in a day are not independent of each other.
[0066] Specifically, the cell state of the previous trip is generated according to the type of the trip chain of the previous trip and the end time point of the previous trip, wherein the cell state can store relevant information carried in the sequence processing process.
[0067] Therefore, the type of the trip chain of the previous trip, the start time point of the current trip and the cell state of the previous trip are input into the first deep learning model for prediction, so as to obtain the type of the trip chain of the current trip.
[0068] Optionally, the inputting the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip into the first deep learning model for prediction to obtain the type of the trip chain of the current trip includes:
[0069] Performing weighted calculation on the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip according to preset weight information to obtain forgetting information, wherein the preset weight information includes a first weight value, a second weight value, and a third weight value, wherein the first weight value corresponds to the type of the trip chain of the previous trip, the second weight value corresponds to the start time point of the current trip, and the third weight value corresponds to the cell state of the previous trip, and the forgetting information is used to indicate the influence of the previous trip on the current trip;
[0070] Updating the forgetting information according to preset bias information to obtain the target forgetting information;
[0071] The cell state of the previous trip is updated according to the target forgetting information to obtain the cell state of the current trip;
[0072] The cell state of the current trip is input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
[0073] In this embodiment, the core concept of LSTM is the cell state, which can store relevant information carried during sequence processing. As the cell state develops, information will continue to be retained or forgotten, which is controlled by three state gates.
[0074] Among them, the forget gate will control the hidden state information of the previous moment through learning, so as to forget and discard irrelevant information, and pass it through the sigmoid function. If the value is larger, it means that it is more related to the current moment, then choose to retain the information, otherwise choose to forget the information. The judgment formula is as follows:
[0075] f t =σ(W xf ·x t-1 +W sf ·T s_t +W cf ·c t-1 +b f )
[0076] Among them, W xf , W sf and W cf Respectively represent the weights of the corresponding inputs, b f Indicates bias. c t-1 is the cell state of the previous period, and σ is the sigmoid function. t-1 It is the type of the previous travel chain, T s-t is the start time of the current trip chain.
[0077] The input gate controls the data input at the current moment, passing the previous hidden state and the current features through the sigmoid function. The larger the value, the more important it is. The tanh function is used to compress the value to adjust the network. The formula is as follows:
[0078] i t =σ(W xi ·x t-1 +W si ·T s_t +W ci ·c t-1 +b i )
[0079] c t =f t ·c t-1 +i t tanh(W xc ·x t +W sc ·T s_t +b c )
[0080] Among them, i t is the new information to be retained, c t It is the newly formed cell state after renewal.
[0081] The output gate controls the output of each cell unit and determines the next hidden state information.
[0082] o t =σ(W xo ·x t-1 +W so ·T s_t +W co ·c t-1 +b o )
[0083] x t =W x o t ·tanh(c t )
[0084] Among them, t Determines which states in the unit will be used for output, x t Is the type of the resulting itinerary chain.
[0085] The improved LSTM algorithm can consider the influence of the past and the future at the same time. Figure 3 As shown, among which, Figure 3 Structurally, it can be seen as a combination of forward LSTM and backward LSTM. Information flow can flow both forward and backward in the hidden layer. The output of the hidden layer is determined by the positive and negative hidden layer vectors. Usually, a concatenation operation is performed, that is, the hidden layer state of the forward LSTM is concatenated with the hidden layer state of the backward LSTM at the corresponding position. The improved LSTM algorithm can simultaneously obtain the bidirectional dependency information of the input sequence. For the output at time t, Represents the hidden layer state of the forward LSTM layer at time t, which contains the information of time t and before in the input sequence. Represents the hidden layer state of the backward LSTM layer at time t, which contains the information of time t and later in the input sequence. and h t Use addition, averaging, and concatenation to process.
[0086] Time series can be divided into single-step prediction and multi-step prediction according to the step size. Single-step prediction means that each time a prediction is made, the input window value predicts a future value. Multi-step prediction means that each time a prediction is made, the input window predicts n values in the future. This application uses recursive multi-step prediction based on a sliding window for prediction. Compared with direct prediction, recursive multi-step prediction inputs the time series into the prediction model in the form of a sliding window, continuously updates the training sample data, and dynamically controls the data from the current moment to the future period through weights.
[0087] Optionally, the second deep learning model is obtained by updating a preset second deep learning model through target parameter information, and the preset second deep learning model is constructed based on initial parameter information, and the target parameter information is obtained by updating the initial parameter information based on an improved BP algorithm and a back propagation algorithm.
[0088] In this embodiment, if Figure 4 As shown, Figure 4 : is a flow chart of the second deep learning model in this embodiment. The second deep learning model is a BP neural network used to predict the time characteristic of the travel chain, that is, the start time and end time of the trip. The preset second deep learning model is constructed by initial parameter information, and the initial parameter information is updated by the improved BP algorithm and the back propagation algorithm to obtain the second deep learning model.
[0089] In recent studies on travel behavior analysis, BP neural network is widely used due to its excellent nonlinear mapping and generalization, but it also has problems of slow convergence and local optimality. Therefore, considering the length and difficulty of training time, this layer uses the improved BP algorithm and the combination of improved BP algorithm and BP neural network for prediction. The input is the start time of the trip chain and the type of the current trip chain. The output is the end time of the trip chain and the start time of the next trip chain.
[0090] It should be noted that the beetle whisker search algorithm is an optimization algorithm inspired by nature, which simulates the detection and search behavior of beetles. The beetle's large antennae are composed of a variety of olfactory receptor cells, which are mainly used to detect the surrounding environment and perceive the smell of prey. The algorithm simulates the mechanism of the beetle's antennae, and determines the search direction and moves forward by the strength of the left and right antennae's perception of the prey's smell until the prey is found. The beetle whisker search algorithm, like the wolf pack algorithm and the cuckoo search algorithm, is a heuristic algorithm inspired by animal foraging. It mainly has two steps: exploration and detection. However, for single-objective problem solving, the improved BP algorithm is superior in solution speed.
[0091] First, in order to simulate the search behavior, the following formula is used to simulate the search direction
[0092]
[0093] Among them, rnd represents the random function and k is the dimension of the position.
[0094] Secondly, the search position on the left and right sides of the center of mass is obtained according to the direction
[0095]
[0096]
[0097] Among them, x rt The location of the area where the longicorn searches on the right side for the tth time; x lt is the area where the longicorn searches on the left for the tth time; x t is the location of the beetle in the tth exploration; d t is the sensing antenna length of the longicorn, and its value should be large enough to cover an appropriate search area so as to jump out of the local minimum point at the beginning.
[0098] Then, in order to clarify the detection behavior, the following iterative model is further generated to update the spatial coordinates by considering the search behavior associated with odor detection.
[0099]
[0100] Among them, sign is the sign function, f is the smell function, and δ is the update step size. Usually, the step size needs to be continuously reduced during the update. The algorithm flow of the improved BP algorithm-BP is as follows. For this purpose, the improved BP algorithm-BP is organically combined. The central idea is: set the initial optimization space dimension of the longhorn beetle whisker search algorithm, and the fitness value uses the mean square error predicted by the BP neural network. Use the collected data as training samples, and at the initial position, perform the next iteration according to the perception of the left and right whiskers, and continuously approach the target problem. This corresponds to the process of iterating and correcting the weights and thresholds in the neural network. After the preliminary optimization of the improved BP algorithm, the initial weights and thresholds of the BP network are optimized, so that it is not easy to fall into local problems and can output the network more accurately.
[0101] Optionally, before inputting the target time period into the third deep learning model to predict the remaining cruising distance of the vehicle to be predicted and obtaining the cruising distance, the method further includes:
[0102] Acquire a training sample set, the training sample set comprising a plurality of training samples, the plurality of training samples corresponding one to one with a plurality of historical trips, wherein the driving distances corresponding to the plurality of historical trips are different;
[0103] Screening the multiple training samples according to a random forest algorithm to determine multiple target training samples;
[0104] A preset third deep learning model is trained based on the multiple target training samples to obtain the third deep learning model.
[0105] In this embodiment, the third deep learning model determines the driving distance of the trip chain. A random forest algorithm can be used to screen multiple training samples to determine multiple target training samples, thereby training the preset third deep learning model through the multiple target training samples to obtain the third deep learning model.
[0106] It should be noted that the random forest is an optimized version of bagging based on the decision tree model, which is proposed to solve the shortcomings of weak generalization ability of decision trees. The same batch of data can only generate one tree with the same algorithm. The bagging strategy can generate different data sets, and then generate multiple decision trees. Each tree is generated according to the following rules: for each tree, after confirming the sample set size, a certain number of training samples are randomly and with replacement extracted from the entire data set as the training set of the tree; a certain number of feature dimensions is selected that does not exceed the sample feature dimension, and a feature subset is randomly selected based on the selected feature dimension. Each time a tree splits, the best feature in its own feature subset is used as the basis for splitting; each tree grows to the maximum extent, and there is no pruning process.
[0107] Specifically, a portion of samples are randomly selected from the total sample for training. Because bad samples are equivalent to noise, adding them to training will lead to a decrease in model accuracy. The average value of the results of multiple experiments is obtained by voting as the final output, which greatly avoids bad sample data and thus improves accuracy.
[0108] Optionally, generating a charging load for a current trip according to the cruising distance and the charging information includes:
[0109] Generating a state of charge according to the cruising distance, the state of charge being used to indicate remaining power information of a vehicle battery;
[0110] Acquiring error information, wherein the error information includes weather parameters and traffic parameters in the current trip;
[0111] generating a charging power curve based on the state of charge, the error information and the charging information, wherein the charging power curve is used to indicate a charging state of a user's car;
[0112] The charging load of the current trip is generated based on the difference information between the charging power curve and the preset charging power curve, the preset charging power curve is the predicted charging power curve of the current trip, the charging information is used to indicate the charging amount of the vehicle to be predicted in the current trip, and the charging load is used to indicate the charging power of the vehicle to be predicted in the current trip.
[0113] In this embodiment, after obtaining the cruising range, the state of charge (SOC), that is, the remaining power, is generated. Specifically, SOC is the ratio of the remaining dischargeable power of the battery after it has been used for a period of time or has been left unused for a long time to the power in its fully charged state, usually expressed as a percentage. When SOC is equal to 1, it means that the battery is fully charged; when SOC is equal to 0, it means that the battery has been completely discharged.
[0114] In addition, one of the shortcomings of the time series-based prediction in the prior art is that it does not take into account the randomness of the user's charging behavior. The user's charging behavior mainly depends on personal preference. Some users are accustomed to returning home to charge after a day's journey, while some users prefer to charge when they go out. Different psychological tolerances for the state of charge (SOC) will also cause differences in charging behavior. If the daily mileage of an electric vehicle is not far, the user may choose to charge every few days, while some users will choose to charge every day. This application makes the following assumptions about the charging behavior of electric vehicle users: First, the starting point and end point of the electric vehicle user's entire day's journey are both in residential areas; whether the electric vehicle is charged after arriving at a certain destination depends on whether the remaining SOC can ensure that the car can travel to the next destination. In addition, for safety reasons, the remaining battery capacity of the electric vehicle should be no less than 15%, so the charging condition S of the electric vehicle after arriving at the destination is obtained. n -ω n+1 l n+1 <0.15C,S n is the battery state of charge when the vehicle arrives at the destination n, ω n+1 is the power consumption per kilometer of the next trip, l n+1 is the driving distance of the next trip; C is the battery capacity.
[0115] Road conditions and weather have a certain impact on the range of electric vehicles, such as Figure 5 and Figure 6 As shown, Figure 5 is the attribution function of road condition and temperature, Figure 6 is the membership function of power consumption per kilometer. In addition to the two external factors mentioned above, the cruising range is also related to vehicle factors such as the efficiency of the battery energy management system and the age of the vehicle. It is very difficult to accurately estimate the actual battery consumption per kilometer of an electric vehicle. This application mainly considers the impact of weather and traffic conditions, and uses fuzzy theory to calculate the impact of weather and traffic conditions on power consumption. The traffic conditions and temperature are standardized, and the traffic conditions at various times of the day are normalized from smooth to congested to [0,1], presented in a triangular shape; secondly, when the weather is too cold or too hot, the air conditioner needs to be turned on, and the temperature also needs to be normalized. Then, the membership function of power consumption per kilometer is obtained according to the fuzzy rules, and finally the power consumption per kilometer is defuzzified.
[0116] This application first considers the correlation of travel activities and uses the improved LSTM algorithm to predict the type of itinerary chain; then uses the feature type and itinerary start time obtained in the previous layer to predict the end time and the start time of the next trip through the BP neural network; finally, the random forest is used to obtain the driving distance. Considering that the BP algorithm in the traditional algorithm is prone to fall into the problem of local optimality, if the improved BP algorithm search algorithm is used, the weight and threshold are optimized to improve the prediction accuracy. The improved LSTM algorithm adopts recursive multi-step prediction training and only establishes a single-step prediction model that can be used iteratively. It has high prediction accuracy and simplifies the complexity of the model.
[0117] It should be noted that this application is applied to cloud space. By proposing a user travel prediction technology, it belongs to the same technical field as the technical low-code template recommendation in cloud space products, solves the same technical problem, namely user behavior analysis and prediction, and realizes the same function, predicting users' future behavior through historical data.
[0118] In actual use, charging load forecasting mainly consists of three parts: data record management, charging load forecasting and map display service.
[0119] Among them, the first part of the system needs to receive and record the travel data sent by users. Users will generate a lot of information about their location during driving. This information has a temporal relationship. Loss or incorrect order will reduce the accuracy of model prediction. Relational databases such as MySql cannot cope with a large number of users initiating requests at the same time, which will cause problems such as data loss. To solve the above problems, non-relational data Redis is introduced as a cache to record the data sent by users. The data in Redis is stored in the form of key-value pairs, and then the data in Redis is written to the MySql database in sequence and in a fixed and quantitative manner through the background, which not only avoids performance degradation under a large number of requests but also realizes the persistence of travel data. Due to the limited storage capacity of the service, each user's data will only be retained for two weeks.
[0120] Charging load prediction mainly consists of two steps: data preprocessing and model calculation. The specific function of data preprocessing is to convert the user's trajectory records from GPS coordinates and time into appropriate trip chain types and spatiotemporal characteristics, and normalize the temperature and traffic conditions to improve the speed of data storage, query and analysis, and prepare for model input; then call the model, that is, the multi-layer machine learning algorithm; before calling the model, first retrieve the user's historical data. If the data to be queried is temporarily stored in Redis, use it directly. If not, query the user's data through MySql; input the historical data into the proposed multi-layer machine learning model to obtain the predicted travel characteristics; and then generate the final load forecast through the calculated power consumption per kilometer and the proposed charging judgment conditions.
[0121] The map display part needs to call the third-party interface of Baidu Map to realize the intuitive display of the location of charging stations. The charging stations in the area are highlighted and marked with different shades of colors according to the size of the charging load, and the ID and predicted load value of each station are displayed at the same time.
[0122] This application uses three deep learning models to predict the trip chain type, target time period and cruising distance of the current trip, and determines the charging load of the vehicle to be predicted by combining charging information, thereby improving the accuracy of predicting vehicle load.
[0123] See also Figure 7 , Figure 7 is a structural diagram of a vehicle load prediction device provided in an embodiment of the present application. Figure 7 As shown, the vehicle load prediction device 700 includes:
[0124] The first prediction module 710 is used to input the starting time point of the current trip of the vehicle to be predicted into the first deep learning model for prediction during the driving process of the vehicle to be predicted, and obtain the trip chain type of the current trip, wherein the trip chain type is used to indicate the starting position and the end position of the current trip;
[0125] The second prediction module 720 is used to predict the information and obtain a target time period, wherein the starting time of the target time period is the starting time point of the next parking, and the ending time point of the target time period is the ending time point of the next parking. The second deep learning model is a deep learning model generated by improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation;
[0126] A third prediction module 730 is used to input the target time period into a third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and obtain a cruising distance;
[0127] The generating module 740 is used to generate the charging load of the current trip according to the cruising distance and the charging information.
[0128] Optionally, the first prediction module 710 includes:
[0129] The first acquisition submodule is used to acquire the type of the trip chain of the previous trip of the current trip and the end time point of the previous trip;
[0130] A first generating submodule, used for generating a cell state of a previous trip according to the type of the trip chain of the previous trip and the end time point of the previous trip, wherein the cell state is used for indicating a driving state parameter of the vehicle to be predicted during the previous trip;
[0131] The prediction submodule is used to input the type of the trip chain of the previous trip, the starting time point of the current trip and the cell state of the previous trip into the first deep learning model for prediction, so as to obtain the type of the trip chain of the current trip.
[0132] Optionally, the prediction submodule includes:
[0133] a calculation unit, configured to perform weighted calculation on the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip according to preset weight information to obtain forgetting information, wherein the preset weight information includes a first weight value, a second weight value, and a third weight value, wherein the first weight value corresponds to the type of the trip chain of the previous trip, the second weight value corresponds to the start time point of the current trip, and the third weight value corresponds to the cell state of the previous trip, and the forgetting information is used to indicate the influence of the previous trip on the current trip;
[0134] A first updating unit, configured to update the forgetting information according to preset bias information to obtain the target forgetting information;
[0135] A second updating unit, configured to update the cell state of the previous trip according to the target forgetting information to obtain the cell state of the current trip;
[0136] The prediction unit inputs the cell state of the current trip into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
[0137] Optionally, the second deep learning model is obtained by updating a preset second deep learning model through target parameter information, and the preset second deep learning model is constructed based on initial parameter information, and the target parameter information is obtained by updating the initial parameter information based on an improved BP algorithm and a back propagation algorithm.
[0138] Optionally, also include:
[0139] An acquisition module, used to acquire a training sample set, wherein the training sample set includes a plurality of training samples, and the plurality of training samples correspond one to one with a plurality of historical trips, wherein the driving distances corresponding to the plurality of historical trips are different;
[0140] A screening module, used to screen the multiple training samples according to a random forest algorithm to determine multiple target training samples;
[0141] A training module is used to train a preset third deep learning model based on the multiple target training samples to obtain the third deep learning model.
[0142] Optionally, the generating module 740 includes:
[0143] A second generating submodule, used for generating a state of charge according to the cruising range, wherein the state of charge is used for indicating the remaining power information of the vehicle battery;
[0144] A second acquisition submodule is used to acquire error information, wherein the error information includes weather parameters and traffic parameters in the current trip;
[0145] a third generating submodule, configured to generate a charging power curve based on the state of charge, the error information and the charging information, wherein the charging power curve is used to indicate a charging state of a user's car;
[0146] The fourth generating submodule is used to generate the charging load of the current trip based on the difference information between the charging power curve and the preset charging power curve, the preset charging power curve is the predicted charging power curve of the current trip, the charging information is used to indicate the charging amount of the vehicle to be predicted in the current trip, and the charging load is used to indicate the charging power of the vehicle to be predicted in the current trip.
[0147] This application uses three deep learning models to predict the trip chain type, target time period and cruising distance of the current trip, and determines the charging load of the vehicle to be predicted by combining charging information, thereby improving the accuracy of predicting vehicle load.
[0148] The present application also provides an electronic device. Figure 8 , the electronic device may include a processor 801, a memory 802, and a program 8021 stored in the memory 802 and executable on the processor 801.
[0149] When the program 8021 is executed by the processor 801, it can achieve Figure 1 Any step in the corresponding method embodiment:
[0150] During the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction, and the trip chain type of the current trip is obtained, and the trip chain type is used to indicate the starting position and the end position of the current trip;
[0151] Input the trip chain type into the second deep learning model to predict the next stop information of the current trip, and obtain a target time period, wherein the starting time of the target time period is the start time point of the next stop, and the end time point of the target time period is the end time point of the next stop;
[0152] Inputting the target time period into a third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and obtaining the cruising distance;
[0153] The charging load of the current trip is generated according to the cruising distance and the charging information.
[0154] Optionally, during the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction to obtain the trip chain type of the current trip, including:
[0155] Get the type of the trip chain of the previous trip of the current trip and the end time of the previous trip;
[0156] generating a cell state of the previous trip according to the type of the trip chain of the previous trip and the end time point of the previous trip, wherein the cell state is used to indicate a driving state parameter of the vehicle to be predicted during the previous trip;
[0157] The type of the trip chain of the previous trip, the start time point of the current trip and the cell state of the previous trip are input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
[0158] Optionally, the inputting the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip into the first deep learning model for prediction to obtain the type of the trip chain of the current trip includes:
[0159] Performing weighted calculation on the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip according to preset weight information to obtain forgetting information, wherein the preset weight information includes a first weight value, a second weight value, and a third weight value, wherein the first weight value corresponds to the type of the trip chain of the previous trip, the second weight value corresponds to the start time point of the current trip, and the third weight value corresponds to the cell state of the previous trip, and the forgetting information is used to indicate the influence of the previous trip on the current trip;
[0160] Updating the forgetting information according to preset bias information to obtain the target forgetting information;
[0161] The cell state of the previous trip is updated according to the target forgetting information to obtain the cell state of the current trip;
[0162] The cell state of the current trip is input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
[0163] Optionally, the second deep learning model is obtained by updating a preset second deep learning model through target parameter information, and the preset second deep learning model is constructed based on initial parameter information, and the target parameter information is obtained by updating the initial parameter information based on an improved BP algorithm and a back propagation algorithm.
[0164] Optionally, before inputting the target time period into the third deep learning model to predict the remaining cruising distance of the vehicle to be predicted and obtaining the cruising distance, the method further includes:
[0165] Acquire a training sample set, the training sample set comprising a plurality of training samples, the plurality of training samples corresponding one to one with a plurality of historical trips, wherein the driving distances corresponding to the plurality of historical trips are different;
[0166] Screening the multiple training samples according to a random forest algorithm to determine multiple target training samples;
[0167] A preset third deep learning model is trained based on the multiple target training samples to obtain the third deep learning model.
[0168] Optionally, generating a charging load for a current trip according to the cruising distance and the charging information includes:
[0169] Generating a state of charge according to the cruising distance, the state of charge being used to indicate remaining power information of a vehicle battery;
[0170] Acquiring error information, wherein the error information includes weather parameters and traffic parameters in the current trip;
[0171] generating a charging power curve based on the state of charge, the error information and the charging information, wherein the charging power curve is used to indicate a charging state of a user's car;
[0172] The charging load of the current trip is generated based on the difference information between the charging power curve and the preset charging power curve, the preset charging power curve is the predicted charging power curve of the current trip, the charging information is used to indicate the charging amount of the vehicle to be predicted in the current trip, and the charging load is used to indicate the charging power of the vehicle to be predicted in the current trip.
[0173] This application uses three deep learning models to predict the trip chain type, target time period and cruising distance of the current trip, and determines the charging load of the vehicle to be predicted by combining charging information, thereby improving the accuracy of predicting vehicle load.
[0174] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, each process of the above-mentioned vehicle load prediction method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0175] The embodiment of the present application further provides a computer program product, which is stored in a storage medium. The computer program product is executed by at least one processor to implement the various processes of the above-mentioned vehicle load prediction method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0176] It should be noted that, in this article, the terms "include", "comprises" or any other variations 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, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0177] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0178] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A method for predicting vehicle load, characterized in that: The method comprises: During the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction, and the trip chain type of the current trip is obtained, and the trip chain type is used to indicate the starting position and the end position of the current trip; Input the trip chain type into the second deep learning model to predict the next stop information of the current trip, and obtain a target time period, wherein the time starting point of the target time period is the start time point of the next stop, and the time ending point of the target time period is the end time point of the next stop. The second deep learning model is a deep learning model generated by improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation; Inputting the target time period into a third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and obtaining the cruising distance; The charging load of the current trip is generated according to the cruising distance and the charging information.
2. The method according to claim 1, characterized in that During the driving process of the vehicle to be predicted, the starting time point of the current trip of the vehicle to be predicted is input into the first deep learning model for prediction, and the trip chain type of the current trip is obtained, including: Get the type of the trip chain of the previous trip of the current trip and the end time of the previous trip; generating a cell state of the previous trip according to the type of the trip chain of the previous trip and the end time point of the previous trip, wherein the cell state is used to indicate a driving state parameter of the vehicle to be predicted during the previous trip; The type of the trip chain of the previous trip, the start time point of the current trip and the cell state of the previous trip are input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
3. The method according to claim 2, characterized in that The step of inputting the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip into the first deep learning model for prediction to obtain the type of the trip chain of the current trip includes: Performing weighted calculation on the type of the trip chain of the previous trip, the start time point of the current trip, and the cell state of the previous trip according to preset weight information to obtain forgetting information, wherein the preset weight information includes a first weight value, a second weight value, and a third weight value, wherein the first weight value corresponds to the type of the trip chain of the previous trip, the second weight value corresponds to the start time point of the current trip, and the third weight value corresponds to the cell state of the previous trip, and the forgetting information is used to indicate the influence of the previous trip on the current trip; Updating the forgetting information according to preset bias information to obtain the target forgetting information; The cell state of the previous trip is updated according to the target forgetting information to obtain the cell state of the current trip; The cell state of the current trip is input into the first deep learning model for prediction to obtain the type of the trip chain of the current trip.
4. The method according to claim 1, characterized in that The second deep learning model is obtained by updating the preset second deep learning model through the target parameter information, and the preset second deep learning model is constructed based on the initial parameter information. The target parameter information is obtained by updating the initial parameter information based on the improved BP algorithm and the back propagation algorithm.
5. The method according to claim 1, characterized in that Before inputting the target time period into the third deep learning model to predict the remaining cruising distance of the vehicle to be predicted and obtaining the cruising distance, the method further includes: Acquire a training sample set, the training sample set comprising a plurality of training samples, the plurality of training samples corresponding one to one with a plurality of historical trips, wherein the driving distances corresponding to the plurality of historical trips are different; Screening the multiple training samples according to a random forest algorithm to determine multiple target training samples; A preset third deep learning model is trained based on the multiple target training samples to obtain the third deep learning model.
6. The method according to claim 1, characterized in that The generating the charging load of the current trip according to the cruising distance and the charging information includes: Generating a state of charge according to the cruising distance, the state of charge being used to indicate remaining power information of a vehicle battery; Acquiring error information, wherein the error information includes weather parameters and traffic parameters in the current trip; generating a charging power curve based on the state of charge, the error information and the charging information, wherein the charging power curve is used to indicate a charging state of a user's car; The charging load of the current trip is generated based on the difference information between the charging power curve and the preset charging power curve, the preset charging power curve is the predicted charging power curve of the current trip, the charging information is used to indicate the charging amount of the vehicle to be predicted in the current trip, and the charging load is used to indicate the charging power of the vehicle to be predicted in the current trip.
7. A vehicle load prediction device, characterized in that: The device comprises: A first prediction module is used to input the starting time point of the current trip of the vehicle to be predicted into the first deep learning model for prediction during the driving process of the vehicle to be predicted, and obtain the trip chain type of the current trip, wherein the trip chain type is used to indicate the starting position and the end position of the current trip; A second prediction module is used to predict the information to obtain a target time period, wherein the starting time of the target time period is the starting time point of the next parking, and the ending time point of the target time period is the ending time point of the next parking, and the second deep learning model is a deep learning model generated by improving the BP neural network, and the second deep learning model is different from the Monte Carlo simulation; A third prediction module, used for inputting the target time period into a third deep learning model to predict the remaining cruising distance of the vehicle to be predicted, and obtaining the cruising distance; A generation module is used to generate the charging load of the current trip according to the cruising range and charging information.
8. An electronic device, comprising: A memory, a processor, and a program stored in the memory and executable on the processor; wherein the processor is used to read the program in the memory to implement the steps in the vehicle load prediction method as described in any one of claims 1 to 6.
9. A readable storage medium for storing a program, characterized in that: When the program is executed by a processor, the steps of the vehicle load prediction method according to any one of claims 1 to 6 are implemented.
10. A computer program product, characterized in that The computer program product is stored in a storage medium, and the computer program product is executed by at least one processor to implement the steps in the vehicle load prediction method according to any one of claims 1 to 6.
