Prediction method and system based on time sequence prediction reconstruction model and medium
By adopting a time-series prediction reconstruction model method in the automotive industry, using autocorrelation functions and time-series symbolic algorithms to process data, combining recursive neural networks and random forest metamodels for prediction, the problem that traditional methods are difficult to accurately analyze and predict vehicle events is solved, and more efficient data processing and prediction accuracy is achieved.
Patent Information
- Application Number
- CN202411714944.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-27
- Publication Date
- 2025-04-25
AI Technical Summary
Traditional time series prediction methods are difficult to meet the requirements of the automotive industry for accurate analysis and prediction of vehicle events, resulting in the inability to effectively improve vehicle operation efficiency.
The prediction method based on time series prediction reconstruction model is adopted, and the data is processed through the autocorrelation function, the time series symbolization algorithm is used for data compression and symbolization, and the time series prediction reconstruction model is determined by combining recursive neural networks and random forest metamodels.
It significantly reduces the dimension of the time series, effectively compresses the amount of data, reduces storage and calculation overhead, improves the prediction accuracy of vehicle events, and is suitable for large-scale time series data processing.
Smart Images
Figure CN119884945A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of time series prediction, and in particular to a prediction method, system and medium based on a time series prediction reconstruction model. Background Art
[0002] The automotive industry is in a period of rapid development, facing increasingly complex market environments and technological advances. In this context, time series forecasting has become an important tool for improving decision-making capabilities and operational efficiency.
[0003] Time series prediction technology can effectively analyze and predict vehicle performance, faults, operating status, supply chain management, and other aspects, so as to timely discover potential problems with the vehicle and formulate effective maintenance strategies. However, traditional time series prediction methods are difficult to meet the requirements of accurate prediction, and thus cannot effectively analyze the vehicle's operating status, equipment status, management and maintenance, resulting in the inability to improve the vehicle.
[0004] It can be seen that how to effectively analyze and predict vehicle events has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the invention
[0005] The present invention provides a prediction method based on a time series prediction reconstruction model to solve the problem of how to improve the time series prediction technology so as to improve the analysis and prediction effect of vehicle events.
[0006] In order to solve the above technical problems, an embodiment of the present invention provides a method for effectively analyzing and predicting vehicle events, including:
[0007] Generate a target time series dataset corresponding to the constructed autocorrelation function;
[0008] Based on a time series symbolization algorithm, the target time series data set is sequentially compressed and symbolized to convert the target time series data set into symbolized data; wherein the compression stage is configured to control the compression error of the time series data using a tolerance;
[0009] Training the selected recursive neural network model with the symbolized data to obtain a first prediction result output by the recursive neural network model; and obtaining a second prediction result output by the trained random forest meta-model;
[0010] Comparing and analyzing the first prediction result and the second prediction result to determine a time series prediction reconstruction model;
[0011] Obtain a target prediction result output by the time series prediction reconstruction model.
[0012] Furthermore, before the target time series data set is compressed and symbolized in sequence, the method further includes:
[0013] Calculating the collected target time series data set through an autocorrelation function, and obtaining the data peak value in the autocorrelation array;
[0014] The target time series data set is segmented according to the peak position.
[0015] Furthermore, the step of sequentially compressing and symbolizing the target time series data set based on a time series symbolization algorithm to convert the target time series data set into symbolized data includes:
[0016] In the process of compressing the target time series data set, the approximation error of the current segment of the target time series data set is calculated, and when the approximation error meets a preset tolerance threshold, the current segment is fixed, and the next round of segmentation is performed;
[0017] Repeat the compression process until a compressed data set to be standardized is obtained;
[0018] Using the time series symbolization algorithm to perform standardization processing on the data set to be standardized to obtain a data set to be symbolized;
[0019] A scaling parameter is introduced to cluster the data set to be symbolized to obtain the symbolized data.
[0020] Further, the training of the selected recursive neural network model with the symbolized data to obtain a first prediction result output by the recursive neural network model; and obtaining a second prediction result output by the trained random forest meta-model include:
[0021] After performing one-hot encoding processing on the symbolized data, the symbolized data is input into the recursive neural network model for training, and the first prediction result output by the trained recursive neural network model is obtained;
[0022] The divided validation set, training set and the first prediction result are input into the random forest meta-model for training, and the second prediction result is obtained.
[0023] Furthermore, the comparing and analyzing the first prediction result and the second prediction result to determine the time series prediction reconstruction model includes:
[0024] Calculating a first root mean square error between the first prediction result and the divided validation set;
[0025] Calculating a second root mean square error between the second prediction result and the validation set;
[0026] The first root mean square error and the second root mean square error are compared to determine the time series prediction reconstruction model.
[0027] Furthermore, obtaining the target prediction result output by the time series prediction reconstruction model includes:
[0028] Analyzing and predicting vehicle events using the time series prediction reconstruction model to obtain analysis and prediction results;
[0029] The analysis prediction result is decoded into a corresponding symbol sequence, and the symbol sequence is reconstructed and analyzed and the length is optimized to obtain the target prediction result.
[0030] Further, the reconstructing and analyzing the symbol sequence includes:
[0031] Based on the cluster center value pair corresponding to each symbol in the symbol sequence, each symbol is replaced;
[0032] The replaced symbol sequence is subjected to inverse compression and linear interpolation to achieve reconstruction analysis of the symbol sequence.
[0033] Another embodiment of the present invention provides a prediction system based on a time series prediction reconstruction model, comprising:
[0034] A data preprocessing module is used to generate a target time series data set corresponding to the constructed autocorrelation function;
[0035] A data compression and symbolization module, used for sequentially compressing and symbolizing the target time series data set based on a time series symbolization algorithm to convert the target time series data set into symbolized data; wherein the compression stage is configured to use a tolerance to control the compression error of the time series data;
[0036] A model training module, used to train the selected recursive neural network model with the symbolized data to obtain a first prediction result output by the recursive neural network model; and obtain a second prediction result output by the trained random forest meta-model;
[0037] A model determination module, used for comparing and analyzing the first prediction result and the second prediction result to determine a time series prediction reconstruction model;
[0038] An analysis and prediction module is used to obtain a target prediction result output by the time series prediction reconstruction model. Another embodiment of the present invention provides a vehicle controller, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the prediction method based on the time series prediction reconstruction model as described above when executing the computer program.
[0039] Yet another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the prediction method based on the time series prediction reconstruction model as described above is implemented.
[0040] Compared with the prior art, the embodiments of the present invention have the following advantages:
[0041] The time series symbol aggregation algorithm significantly reduces the dimension of the time series through symbolization technology. While retaining the main trend and shape of the time series, it can effectively compress the data volume and reduce the storage and computing overhead. It is suitable for large-scale time series data processing. The time series symbol aggregation algorithm and neural network model are used to perform two-way processing and adaptive learning mechanism on the collected vehicle data, which can more accurately predict vehicle events. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a flowchart of a prediction method based on a time series prediction reconstruction model in one embodiment of the present invention;
[0043] Figure 2 is a logic diagram of a time series prediction scheme in one embodiment of the present invention;
[0044] Figure 3 It is a schematic diagram of the structure of a prediction system based on a time series prediction reconstruction model in one embodiment of the present invention. DETAILED DESCRIPTION
[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0046] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.
[0047] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0048] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.
[0049] It can be understood that time series prediction technology is of great significance for vehicle performance monitoring, fault prediction and maintenance management, supply chain management, market trend prediction, and analysis and prediction of the surrounding environment during autonomous driving. Based on this, an embodiment of the present invention provides a prediction method based on a time series prediction reconstruction model. For details, please refer to Figure 1 , Figure 1 The figure shows a flow chart of a prediction method based on a time series prediction reconstruction model in one embodiment of the present invention, which includes the following steps:
[0050] S1. Generate a target time series dataset corresponding to the constructed autocorrelation function.
[0051] As you can understand, the autocorrelation function is a tool used in statistics to describe the dependency between a moment in a time series and its past moments. It reflects the correlation of the time series at different time lags and reveals the patterns, trends or periodicity in the series.
[0052] The autocorrelation function ACF(k) measures the similarity of the time series at the time lag k, indicating the degree of correlation between the data at the current time point and the data at the previous k time points.
[0053] When the user uploads a dataset, the system checks whether the dataset file is in .npy or .csv format. If it is not, the system prompts the user to re-upload the dataset in the correct format. If it is in .npy format, it is directly converted to np.ndarray. If it is in .csv format, it is first converted to pd.DataFrame and then converted to np.ndarray.
[0054] After the conversion is completed, check whether the missing value ratio of the dataset is greater than 50%. If so, discard the dataset and prompt the user to upload it again. Otherwise, remove the missing values in the dataset to obtain the dataset {x1,x2,…,x n In the embodiment of the present invention, the data is divided in advance into a ratio of 90% training set and 10% validation set to serve as input for the subsequent model training process.
[0055] In some embodiments of the present invention, the collected target time series data set is calculated by an autocorrelation function, and the data peak value in the autocorrelation array is obtained, specifically:
[0056] For the above dataset {x1,x2,…,x n}Construct the autocorrelation function and calculate ACF(1), ACF(2), …, ACF(n-1) for time series data of length n.
[0057] These values are placed in the autocorrelation array acf_array to obtain the data peak value in acf_array, which is larger than the values on the left and right sides. The target time series data set is segmented according to the peak position.
[0058] As an example, if ACF(a1), ..., ACF(a k ) is the peak value (1≤a1 <a2<…<a k ≤n-1), then a1, …, a k is the segmentation point. The target time series dataset is segmented according to the determined segmentation point, which is expressed as:
[0059]
[0060] In the data preprocessing stage, the embodiment of the present invention uses the autocorrelation function to determine the lag order, and can segment the time series with periodicity, so that the periodicity in the time series can be better utilized during model prediction.
[0061] S2. Compression and symbolization are performed on the target time series data set in sequence based on a time series symbolization algorithm to convert the target time series data set into symbolized data; wherein the compression stage is configured to use a tolerance to control the compression error of the time series data.
[0062] It can be understood that the time series symbolization algorithm (symbol clustering) converts continuous time series data into discrete symbol sequences, which can reduce data dimensions and thus reduce data complexity. That is, automatic feature extraction is completed through the time series symbolization process, which reduces data complexity and improves quality.
[0063] Based on this, the embodiment of the present invention uses a time series symbolization algorithm (in the embodiment of the present invention, the time series symbolization algorithm is named ABBA algorithm) to compress and symbolize the preprocessed and segmented data. For details, please refer to Figure 2 , Figure 2 Shown is a logic diagram of a time series prediction solution in one embodiment of the present invention.
[0064] from Figure 2 In the process of symbolizing the sequence using the ABBA algorithm, when the target time series data set has not completed the segmentation of the entire sequence, the time series is first compressed.
[0065] Specifically, in the process of compressing the target time series data set, the embodiment of the present invention first starts from the first data point x1 of the time series data set and performs line segment fitting of the first segment until a preset tolerance threshold is reached. The data points in the time series are traversed, and the segment length is gradually increased to calculate the length len and increment inc of the current segment, which are expressed as follows:
[0066] len=ji,inc=x j -x i
[0067] Where i is the starting point of the current segment and j is the end point of the current segment.
[0068] Next, calculate the approximation error of the current segment of the target time series dataset. When the approximation error meets the preset tolerance threshold, fix the current segment and start the next round of segmentation. In the embodiments of the present invention, the mean square error (MSE) is used to calculate the approximation error of the current segment. If the MSE of the current segment is less than the set tolerance threshold, the segment length is continuously increased; if the MSE exceeds the tolerance, the current segment is fixed. In this embodiment, the fixed current segment is set to (len, inc), and then a new segmentation is started.
[0069] Repeat the above compression process until the dataset to be standardized after compression is obtained. Based on the comparison and judgment of the tolerance threshold and the error, continuous segmentation is performed until all data points of the time series are covered. Exemplarily, the final generated sequence of (len, inc) pairs is the compressed representation of the time series.
[0070] As Figure 2 shown, after the compression process is completed, in order to avoid the influence of the absolute value difference between the length and the increment on the clustering result, the time series data after compression is first standardized to ensure that the length and the increment have the same scale during clustering and will not affect the result due to dimensional differences. In the embodiments of the present invention, a time series symbolization algorithm is used to standardize the dataset to be standardized, and a dataset to be symbolized is obtained.
[0071] A scaling parameter is introduced to cluster the dataset to be symbolized, and the symbolized data is obtained.
[0072] In some embodiments of the present invention, exemplarily, clustering is performed according to the scaling parameter scl. If scl = 0, only one-dimensional clustering is performed on inc; if scl = ∞, only one-dimensional clustering is performed on len; if 0 < scl < ∞, the k-means clustering algorithm is used. Each clustering center point represents a symbol, and all clustering centers form a symbol set. Each (len, inc) pair is symbolized and mapped to the symbol corresponding to the clustering center point, thereby completing the symbolization of the sequence.
[0073] In this embodiment, the compression error of the time series data is controlled by the tolerance, and the time series is symbolized based on clustering, so that the trend characteristics in the sequence can be effectively maintained.
[0074] S3 - S4: Train the selected recurrent neural network model with the symbolized data to obtain the first prediction result output by the recurrent neural network model; and, obtain the second prediction result output by the trained random forest meta-model. Then, perform a comparison and analysis on the first prediction result and the second prediction result to determine the time series prediction reconstruction model.
[0075] After obtaining the symbolized data, in order to achieve a better model training effect, the embodiment of the present invention performs one-hot encoding on the symbolized data. That is, a unique vector containing only 0 and 1 is generated for each symbol, exemplarily, a→(0,0,0), b→(0,0,1), c→(0,1,0), ..., and it is input into a recursive neural network model (LSTM) for training, and then the first prediction result output by the trained recursive neural network model is obtained.
[0076] The following is a detailed description of the LSTM training process:
[0077] Because in the data preprocessing stage, the dimension of the segmentation process of the segmentation point increases, such as Figure 2 Therefore, the time series needs to be dimensionally reduced before training.
[0078] After dimensionality reduction, the number of models that need to be trained is calculated based on the different values of the training parameters. By default, the values based on the training parameters are as follows:
[0079] i.scl (values are 0, 0.5, 2, inf)
[0080] ii.tolerance (values are 0.05, 0.1)
[0081] iii.learning_rate (values are 0.05, 0.1)
[0082] iv.num_layers (values are 1, 2, 3)
[0083] v.state (values are stateful, stateless)
[0084] Among them, scl is the scaling factor, tolerance is the tolerance, learning_rate is the learning rate, num_layers is the number of layers of the LSTM model, and state is a parameter to determine whether LSTM uses the cell state.
[0085] In some embodiments of the present invention, the values of the parameters can be specified by yourself, and the program automatically combines and generates different training parameters before the training starts. Exemplarily, this embodiment trains a total of: 4×2×2×3×2=96 models. It is worth noting that in actual situations, the number of CPU or GPU cards used for parallel training is specified according to the specific situation. For example, if an 8-core GPU uses 8 cards, the 96 models will be evenly distributed to 8 cards for training.
[0086] Then, after the LSTM model is trained, the first root mean square error between the first prediction result and the divided validation set is calculated. It can be understood that after each model is trained, the model (.pth format), the prediction output result of the model, i.e., the first prediction result, and the relative mean square error RMSE (i.e., the first root mean square error) compared with the validation set will be saved. The smaller the RMSE, the more accurate the model prediction.
[0087] Then, the divided validation set, training set and the above-mentioned first prediction result are input into the random forest meta-model for training, the second prediction result is obtained, and the second root mean square error between the second prediction result and the validation set is calculated. In this embodiment, Stacking is used for model fusion, and the prediction results of each model are combined to form a new feature matrix. Each column represents the prediction result of a model on a training set or a test set. The prediction results of the basic model on the training set are used as a new feature matrix, and the validation set is used as the target value to train the random forest meta-model. After the training is completed, the prediction result of the output model, that is, the second prediction result, is compared with the validation set, and the RMSE, that is, the second root mean square error, is calculated.
[0088] Finally, the first root mean square error and the second root mean square error are compared, that is, the RMSE of the random forest metamodel is compared with the RMSE of each LSTM model. If the RMSE of the random forest metamodel is not greater than the RMSE of each LSTM model, the random forest metamodel is used for prediction output; otherwise, the LSTM model with the smallest RMSE is selected for prediction output to determine the time series prediction reconstruction model.
[0089] S5. Obtain the target prediction result output by the time series prediction reconstruction model.
[0090] After the time series prediction reconstruction model used for prediction is determined, the vehicle event is analyzed and predicted using the time series prediction reconstruction model to obtain the analysis and prediction results.
[0091] In the embodiment of the present invention, a test set (generally the original sequence, ie, training set + validation set) is input into the time series prediction reconstruction model, and the analysis and prediction results are output.
[0092] The analysis prediction result is One-Hot decoded into a corresponding symbol sequence, and the symbol sequence is reconstructed and analyzed and the length is optimized to obtain the target prediction result.
[0093] The embodiment of the present invention uses the ABBA algorithm to reconstruct the acquired symbol sequence. Specifically, based on the cluster center value pair corresponding to each symbol in the symbol sequence, each symbol is replaced; illustratively, each symbol is replaced with the value pair (len, inc) of its corresponding cluster center, and since len may not be an integer, quantization is required to realign len in the segment to an integer grid.
[0094] The replaced symbol sequence is subjected to reverse compression and linear interpolation to achieve reconstruction analysis of the symbol sequence. Reverse compression can connect these discrete segmented values through linear interpolation to generate a complete time series. It can be understood that in the reconstructed time series, the reconstruction error presents a random walk pattern similar to the Brownian bridge.
[0095] It is worth noting that in some embodiments of the present invention, after the reconstruction is completed, it will be further determined whether the length of the reconstructed time series is the same as the prediction length specified by the user. If it is shorter than the specified length, cumulative error compensation is used; if it is longer than the specified length, the first i elements are directly sliced and intercepted. This is the final analysis and prediction result of the entire model.
[0096] Finally, the final analysis and prediction results are visualized. For example, the original data set and the prediction results are spliced, and the matplotlib library is used to draw a line graph, in which the line segments of the original data set and the prediction results are represented by different colors for easy distinction.
[0097] In summary, this embodiment processes and segments the sequence data set based on an adaptive function, so that the trained model can better capture the periodicity in the time series; uses a time series symbolization algorithm to symbolize the time series based on clustering, which can effectively maintain the trend characteristics in the sequence; in the compression stage and data reconstruction stage of the time series data set, the model uses tolerance to control the compression error and reconstruction of the time series data. The smaller the tolerance, the more segments the algorithm will generate, so that the error of each segment linearly approximating the original time series is smaller and the reconstruction accuracy is higher; uses the LSTM model and Stacking machine ensemble learning method to fuse multiple models, thereby effectively improving the prediction accuracy of the model.
[0098] An embodiment of the present invention provides a prediction system based on a time series prediction reconstruction model. For details, see Figure 3 , Figure 3 The figure shows a schematic diagram of a prediction system structure based on a time series prediction reconstruction model in one embodiment of the present invention, including:
[0099] A data preprocessing module M1 is used to generate a target time series data set corresponding to the constructed autocorrelation function;
[0100] A data compression and symbolization module M2, used for sequentially compressing and symbolizing the target time series data set based on a time series symbolization algorithm to convert the target time series data set into symbolized data; wherein the compression stage is configured to use a tolerance to control the compression error of the time series data;
[0101] The model training module M3 is used to train the selected recursive neural network model with the symbolic data to obtain a first prediction result output by the recursive neural network model; and obtain a second prediction result output by the trained random forest meta-model;
[0102] A model determination module M4 is used to compare and analyze the first prediction result and the second prediction result to determine a time series prediction reconstruction model;
[0103] The analysis and prediction module M5 is used to obtain the target prediction result output by the time series prediction reconstruction model.
[0104] The embodiment of the present invention also provides a vehicle controller, which can be used for vehicle performance monitoring, vehicle fault prediction and maintenance management, and vehicle automatic driving. It can be understood that the vehicle controller is responsible for coordinating and managing the vehicle's power system, including the motor, engine (for hybrid vehicles), transmission, and battery pack. Based on time series analysis technology, the vehicle controller can adjust the output of the power system in real time according to the driver's input (such as accelerator pedal position, brake pedal position, etc.) and vehicle status (such as vehicle speed, battery power, motor temperature, etc.) to meet the vehicle's driving needs. Furthermore, through time series analysis technology, the vehicle controller can effectively process and predict traffic flow and pedestrian behavior around the vehicle, and timely warn of accidents during vehicle operation.
[0105] Accordingly, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to perform the steps in the prediction method based on the time series prediction reconstruction model in the above embodiment, for example Figure 1 Steps S1 to S5 described in .
[0106] The technical features and technical effects of the prediction system based on the time series prediction reconstruction model proposed in the embodiment of the present invention are the same as the technical features and technical effects of the prediction method based on the time series prediction reconstruction model proposed in the embodiment of the present invention, and will not be repeated here.
[0107] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A prediction method based on a time series prediction reconstruction model, characterized in that: Applied to vehicle event prediction, including: Generate a target time series dataset corresponding to the constructed autocorrelation function; Based on a time series symbolization algorithm, the target time series data set is sequentially compressed and symbolized to convert the target time series data set into symbolized data; wherein the compression stage is configured to control the compression error of the time series data using a tolerance; Training the selected recursive neural network model with the symbolized data to obtain a first prediction result output by the recursive neural network model; and obtaining a second prediction result output by the trained random forest meta-model; Comparing and analyzing the first prediction result and the second prediction result to determine a time series prediction reconstruction model; Obtain the target prediction result output by the time series prediction reconstruction model.
2. The prediction method based on the time series prediction reconstruction model according to claim 1, characterized in that: Before the target time series data set is compressed and symbolized in sequence, the method further includes: Calculating the collected target time series data set through an autocorrelation function, and obtaining the data peak value in the autocorrelation array; The target time series data set is segmented according to the peak position.
3. The prediction method based on the time series prediction reconstruction model according to claim 1, characterized in that: The method of sequentially compressing and symbolizing the target time series data set based on the time series symbolization algorithm to convert the target time series data set into symbolic data includes: In the process of compressing the target time series data set, the approximation error of the current segment of the target time series data set is calculated, and when the approximation error meets a preset tolerance threshold, the current segment is fixed, and the next round of segmentation is performed; Repeat the compression process until a compressed data set to be standardized is obtained; Using the time series symbolization algorithm to perform standardization processing on the data set to be standardized to obtain a data set to be symbolized; A scaling parameter is introduced to cluster the data set to be symbolized to obtain the symbolized data.
4. The prediction method based on the time series prediction reconstruction model according to claim 1, characterized in that: The step of training the selected recursive neural network model with the symbolized data to obtain a first prediction result output by the recursive neural network model; and obtaining a second prediction result output by the trained random forest meta-model comprises: After performing one-hot encoding processing on the symbolized data, the symbolized data is input into the recursive neural network model for training, and the first prediction result output by the trained recursive neural network model is obtained; The divided validation set, training set and the first prediction result are input into the random forest meta-model for training, and the second prediction result is obtained.
5. The prediction method based on the time series prediction reconstruction model according to claim 4, characterized in that: The comparing and analyzing the first prediction result and the second prediction result to determine the time series prediction reconstruction model includes: Calculating a first root mean square error between the first prediction result and the divided validation set; Calculating a second root mean square error between the second prediction result and the validation set; The first root mean square error and the second root mean square error are compared to determine the time series prediction reconstruction model.
6. The prediction method based on the time series prediction reconstruction model according to claim 1, characterized in that: The obtaining of the target prediction result output by the time series prediction reconstruction model includes: Analyzing and predicting vehicle events using the time series prediction reconstruction model to obtain analysis and prediction results; The analysis prediction result is decoded into a corresponding symbol sequence, and the symbol sequence is reconstructed and analyzed and the length is optimized to obtain the target prediction result.
7. The prediction method based on the time series prediction reconstruction model according to claim 6, characterized in that: The reconstructing and analyzing the symbol sequence includes: Based on the cluster center value pair corresponding to each symbol in the symbol sequence, each symbol is replaced; The replaced symbol sequence is subjected to inverse compression and linear interpolation to achieve reconstruction analysis of the symbol sequence.
8. A prediction system based on a time series prediction reconstruction model, characterized in that: include: A data preprocessing module is used to generate a target time series data set corresponding to the constructed autocorrelation function; A data compression and symbolization module, used for sequentially compressing and symbolizing the target time series data set based on a time series symbolization algorithm to convert the target time series data set into symbolized data; wherein the compression stage is configured to use a tolerance to control the compression error of the time series data; A model training module, used to train the selected recursive neural network model with the symbolized data to obtain a first prediction result output by the recursive neural network model; and obtain a second prediction result output by the trained random forest meta-model; A model determination module, used for comparing and analyzing the first prediction result and the second prediction result to determine a time series prediction reconstruction model; The analysis and prediction module is used to obtain the target prediction result output by the time series prediction reconstruction model.
9. A vehicle controller, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the prediction method based on the time series prediction reconstruction model as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer readable storage medium stores There is a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, Implement the prediction method based on the time series prediction reconstruction model as described in any one of claims 1 to 7.