A time series data prediction method and system based on time series large model optimization

Abstract

The present invention discloses a time series data prediction method and system based on time series big model optimization, which belongs to the field of time series data prediction technology and is used to solve the technical problem that the existing time series data prediction model depends on future covariate data, and the future covariate data is difficult to obtain accurately, which limits the practicality and application scope of the model. The method includes: collecting target time series data and past covariate data sets within a preset historical time period; training a random forest regression model through the target time series data; predicting the time series prediction data within the most recent time period based on the standardized past covariate data set and the random forest regression model; obtaining the residual time series prediction results of the time series prediction data within the most recent time period and the actual time series data through the optimized time series big model, performing error compensation on the random forest regression model, and predicting the time series prediction data within a custom time period through the random forest regression model after error compensation.

Description

Technical Field

[0001] The present invention relates to the technical field of time series data prediction, and in particular to a time series data prediction method and system based on time series large model optimization. Background Art

[0002] With the development of science and technology and the surge in data, time series data prediction is increasingly used in many fields, including financial market forecasting, weather forecasting, energy consumption forecasting, medical diagnosis, and intelligent transportation. However, existing time series data prediction methods face many challenges and shortcomings: First, time series data usually has a complex structure, including long-term trends, seasonal fluctuations, and short-term fluctuations. In addition, the correlation and nonlinear relationship between different data sources further increase the difficulty of modeling.

[0003] In many practical applications, the prediction ability of single-variable prediction models is limited, and multivariate collaborative prediction can significantly improve the accuracy of prediction. However, the modeling and prediction of multivariate time series data face higher complexity, especially when dealing with the interdependence between different variables, which often lacks sufficient flexibility and robustness, and it is difficult to fully capture the interaction between variables. In addition, the time series prediction model needs to maintain good prediction performance on new data outside the training data, which requires the model to have strong generalization ability. Nowadays, many models will have overfitting problems in practical applications, resulting in decreased prediction performance on new data. How to improve the generalization ability of the model and reduce overfitting is an important challenge in the field of time series prediction. It is inevitable that time series data has missing and noise problems, which will seriously affect the prediction accuracy of the model. Traditional methods have limited effect in dealing with missing data and noise. How to effectively fill missing data and reduce noise is an important direction to improve prediction accuracy.

[0004] In order to solve the above problems, time series prediction models based on deep learning have gradually emerged. Among them, the basic time series prediction model TimesFM is a time series prediction model based on the transformer architecture, which can effectively process large-scale, multivariate time series data. However, the TimesFM model requires future covariate data when making predictions based on covariates, and in practical applications, future covariate data is often difficult to obtain accurately, which limits the practicality and application scope of the model. Summary of the invention

[0005] The embodiments of the present invention provide a time series data prediction method and system based on time series large model optimization, which are used to solve the following technical problems: the existing time series data prediction model relies on future covariate data, but in practical applications, future covariate data is difficult to obtain accurately, which limits the practicality and application scope of the model.

[0006] The embodiment of the present invention adopts the following technical solutions:

[0007] On the one hand, an embodiment of the present invention provides a time series data prediction method based on time series big model optimization, the method comprising: collecting target time series data within a preset historical time period of a scenario to be predicted and a corresponding past covariate data set;

[0008] Train the random forest regression model using the standardized target time series data;

[0009] Based on the standardized past covariate dataset and the trained random forest regression model, predict the time series forecast data in the recent time period;

[0010] Based on the long short-term memory network LSTM, the time series large model is optimized; through the optimized time series large model, the residual time series prediction result of the time series prediction data and the actual time series data in the recent time period is obtained;

[0011] According to the residual time series prediction result, the random forest regression model is error compensated to obtain an optimized random forest regression model;

[0012] The optimized random forest regression model is used to predict time series forecast data within a custom time period.

[0013] In a feasible implementation, collecting target time series data within a preset historical time period of the scenario to be predicted and the corresponding past covariate data set specifically includes:

[0014] Collect the target time series data within the preset historical time period in the scenario to be predicted, as well as the past covariate data set with time difference;

[0015] The collected target time series data and past covariate data sets are smoothed by the moving average method, and missing data are filled;

[0016] The filled data is standardized and normalized to remove noise and outliers.

[0017] In a feasible implementation, the random forest regression model is trained by using the standardized target time series data, specifically including:

[0018] Divide the target time series data after standardization within the preset historical time period into several batches of data according to timestamps;

[0019] The random forest regression model is trained in batches according to the order of the batches using the batches of data, so that the random forest regression model can obtain the nonlinear relationship between the covariate data and the time series data.

[0020] In a feasible implementation, based on the standardized past covariate data set and the trained random forest regression model, the time series forecast data in the nearest time period is predicted, specifically including:

[0021] The standardized past covariate data set is input into the trained random forest regression model to perform time series data prediction, and the time series prediction data of the most recent time period is obtained.

[0022] In a feasible implementation, based on the long short-term memory network LSTM, the time series large model is optimized, specifically including:

[0023] A long short-term memory network LSTM is added between the input layer and the transformer layer of the TimesFM time series large model to obtain an optimized time series large model; wherein the optimized time series large model extracts the long-term features of the input data through the LSTM network, and uses the long-term features as the input data of the transformer layer.

[0024] In a feasible implementation, obtaining the residual time series prediction result between the time series prediction data and the actual time series data in the most recent time period through the optimized time series big model specifically includes:

[0025] Acquire the actual time series data of the most recent time period of the scenario to be predicted;

[0026] Calculating the residual between the time series prediction data of the most recent time period and the actual time series data;

[0027] The residual is input into the optimized time series large model to obtain the residual time series prediction result between the time series prediction data and the actual time series data.

[0028] In a feasible implementation, according to the residual time series prediction result, the random forest regression model is error compensated to obtain an optimized random forest regression model, which specifically includes:

[0029] The residual time series prediction result is added to the output layer of the random forest regression model, and a mean smoothing compensation algorithm is attached to obtain an optimized random forest regression model, so that the prediction result output by the random forest regression model is superimposed on the residual time series prediction result to achieve error compensation.

[0030] In a feasible implementation, predicting time series forecast data within a custom time period by using the optimized random forest regression model specifically includes:

[0031] The target time series data in the current time period and the corresponding covariate data set are input into the optimized random forest regression model to obtain the time series forecast data in the future custom time period.

[0032] In a feasible implementation, the target time series data in the current time period and the corresponding covariate data set are input into the optimized random forest regression model to obtain the time series prediction data in the future custom time period, specifically including:

[0033] according to , and obtain the prediction results of the optimized random forest regression model;

[0034] in, Forecasting time series data for a future custom time period, A random forest regression model is based on the past covariate dataset within the preset historical time period The prediction results, is the actual time series data corresponding to the x' time point, The optimized time series model takes the prediction results of the random forest regression model and the residuals of the actual time series data as input, and fits and outputs the residuals of the future custom time period T, which is consistent with the length of the future data to be predicted; is the covariate dataset for the current time period.

[0035] On the other hand, an embodiment of the present invention further provides a time series data prediction system based on time series large model optimization, the system comprising:

[0036] The data collection module is used to collect the target time series data and the corresponding past covariate data set within the preset historical time period of the scenario to be predicted, and standardize the collected data;

[0037] The covariate regression training and prediction module is used to train the random forest regression model through the standardized target time series data; based on the standardized past covariate data set and the trained random forest regression model, the time series prediction data in the recent time period is predicted;

[0038] A time series large model module is used to optimize the time series large model and obtain the residual time series prediction result between the time series prediction data and the actual time series data in the most recent time period through the optimized time series large model;

[0039] A compensation module is used to perform error compensation on the random forest regression model according to the residual time series prediction result to obtain an optimized random forest regression model; and predict the time series prediction data within a custom time period through the optimized random forest regression model.

[0040] Compared with the prior art, the time series data prediction method and system based on time series large model optimization provided by the embodiment of the present invention have the following beneficial effects:

[0041] The method and system proposed in the present invention have the advantages of low training cost, low data requirements, simplicity, etc. The present invention adopts the improved TimesFM basic time series prediction model for data prediction, which can reduce training costs, save training time, and improve the efficiency and accuracy of time series data prediction. At the same time, the advantages of the random forest regression model in processing nonlinear relationships are utilized to further improve the prediction performance of the model. In the case of limited computing resources, efficient time series data prediction is achieved through a lightweight model, which significantly reduces the consumption and cost of computing resources.

[0042] In addition, the present invention combines LSTM with the transformer and introduces the random forest model to replace the traditional linear regression model to realize time difference regression. The covariate data with a certain time difference from the data to be tested is used as the input of the regression model. The practical performance of the model is improved without reducing the prediction accuracy of the model, and the prediction performance and applicability of the time series prediction model are effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0044] Figure 1 A flow chart of a time series data prediction method based on time series large model optimization provided by an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of a time series data prediction system architecture based on time series large model optimization provided by an embodiment of the present invention;

[0046] Figure 3 A schematic diagram of the structure of a time series data prediction system based on time series large model optimization provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0048] The embodiment of the present invention provides a time series data prediction method based on time series large model optimization, such as Figure 1 As shown, the time series data prediction method based on time series large model optimization specifically includes steps S101-S106:

[0049] S101. Collect target time series data within a preset historical time period of the scenario to be predicted and the corresponding past covariate data set, and standardize the collected data.

[0050] Specifically, the target time series data within a preset historical time period in the scenario to be predicted and the past covariate data set with a certain time difference therewith are collected.

[0051] Furthermore, the collected target time series data and past covariate data sets are smoothed by the moving average method, and missing data are filled in. The filled data are standardized and normalized to remove noise and outliers.

[0052] S102. Train a random forest regression model using the standardized target time series data.

[0053] Specifically, the target time series data that has been standardized within a preset historical time period is divided into several batches of data according to timestamps.

[0054] Furthermore, the random forest regression model is trained in batches according to the order of the batches through several batches of data, so as to obtain the nonlinear relationship between the covariate data and the time series data through the random forest regression model.

[0055] S103. Based on the standardized past covariate data set and the trained random forest regression model, predict the time series forecast data in the nearest time period.

[0056] Specifically, the standardized past covariate data set is input into the trained random forest regression model for time series data prediction to obtain the time series prediction data for the most recent time period.

[0057] S104. Optimize the time series big model based on the long short-term memory network LSTM; obtain the residual time series prediction results of the time series prediction data in the most recent time period and the actual time series data through the optimized time series big model.

[0058] Specifically, a long short-term memory network LSTM is added between the input layer and the transformer layer of the TimesFM time series model to obtain an optimized time series model; wherein the optimized time series model extracts the long-term features of the input data through the LSTM network, and uses the long-term features as the input data of the transformer layer.

[0059] Furthermore, the actual time series data of the most recent time period of the scene to be predicted is obtained, and the residual between the time series prediction data and the actual time series data of the most recent time period is calculated. The calculated residual is input into the optimized time series large model to obtain the residual time series prediction result between the time series prediction data and the actual time series data.

[0060] S105. According to the residual time series prediction result, the random forest regression model is error compensated to obtain an optimized random forest regression model.

[0061] Specifically, the residual time series prediction results are added to the output layer of the random forest regression model, and a mean smoothing compensation algorithm is attached to obtain an optimized random forest regression model, so that the prediction results output by the random forest regression model are superimposed on the residual time series prediction results to achieve error compensation.

[0062] S106. Predict the time series forecast data within the custom time period through the optimized random forest regression model.

[0063] Specifically, the target time series data in the current time period and the corresponding covariate data set are input into the optimized random forest regression model to obtain the time series prediction data in the future custom time period.

[0064] The specific implementation method is: according to , and obtain the prediction results of the optimized random forest regression model;

[0065] in, Forecasting time series data for a future custom time period, Based on the past covariate dataset for the preset historical time period for the random forest regression model The prediction results, is the actual time series data corresponding to the x' time point, The optimized time series model takes the prediction results of the random forest regression model and the residuals of the actual time series data as input, and fits and outputs the residuals of the future custom time period T, which is consistent with the length of the future data to be predicted; is the covariate dataset for the current time period.

[0066] As a feasible implementation method, Figure 2 A schematic diagram of a time series data prediction system architecture based on time series large model optimization provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the present invention proposes a method for predicting time series data based on optimization of a large time series model. First, LSTM is added to the TimesFM basic time series prediction model to optimize the existing architecture. Then, the random forest regression model is trained using lagged covariate data to predict and fit future data to be tested, thereby assisting the transformer in TimesFM to capture the changing trend of future data and achieve prediction. Specifically, first, LSTM is used to extract the long-term features of the data, and this feature is used as the input of the transformer. Then, the random forest regression model is trained with the sorted data, and then the past covariates relative to the data to be tested, that is, the lagged covariates, are input into the trained random forest for prediction, and the optimized model is used to fit the trend of the difference between the prediction result of the random forest and the actual value, and finally this trend is attached to the actual prediction result to achieve error compensation.

[0067] The specific implementation steps are as follows:

[0068] 1) Collect a certain amount of time series data and a lagged covariate data set with a certain time difference , ensuring that the data covers different time periods and multiple scenarios.

[0069] 2) Standardize the collected data, remove noise and outliers, and ensure data quality. This includes but is not limited to using the moving average method for smoothing, filling in missing data, and performing data standardization and normalization, etc., to finally obtain .

[0070] 3) Adjust the timing model The architecture introduces LSTM, based on the processed lagged covariate data set Training a Random Forest Model , so as to capture the nonlinear relationship between the covariate and the target variable. The time difference regression method is used to complete the time series prediction based on the lagged covariate, avoiding the dependency problem of future covariates and enhancing the generalization performance of the model.

[0071] 4) Based on the optimized large model , for future target data Make predictions and generate confidence estimates to help users assess the accuracy of the prediction results.

[0072] The main feature of this method is that the collected time series data and the corresponding covariate data cover different time periods and multiple scenarios. This step ensures the diversity and comprehensiveness of the data and enables the model to adapt to various complex time series patterns. Then the collected data is standardized to remove noise and outliers. This method improves the availability and accuracy of the data. The architecture of the time series model TimesFM is adjusted, and LSTM is used to extract input features. The extracted features are input into the transformer Transformer to assist the transformer in capturing long-term time relationships. The random forest model introduced in the training reduces the training cost. The introduction of past covariates avoids the problem of dependence on future covariates. Among them, there is a certain time difference between the corresponding covariate data set collected in step 1) and the predicted text data. Usually, the covariate is a certain time point earlier than the data to be tested. The large model used in step 3) is the TimesFM model improved by optimization, and the hyperparameters of the model are adjusted according to the time difference concept proposed in step 1). In addition, the training of the regression model needs to be trained in batches to improve the fitting ability of the regression model. The evaluation method provided in step 4) is based on the mean average error (MAE).

[0073] The system provided by the present invention includes: a data collection module, a time series large model module, a covariate regression training and prediction module, and a compensation module. Among them, the data collection module is used to collect covariate data that is related to the prediction target and has a lag effect on the prediction target; the large time series model training module is used to add an LSTM auxiliary transformer to capture time series information, modify the covariate input from future data requirements to past data requirements, and realize time difference prediction based on lagged covariates; the covariate regression training and prediction module is used to use a random forest regression model to preliminarily fit the target data based on the covariate input; the compensation module is used to realize mean smoothing compensation based on the distance between the actual value and the prediction result.

[0074] In order to facilitate understanding of the present invention, specific embodiments are given below:

[0075] To predict the number of available parking spaces in a parking lot in the next hour, first collect the number of available parking spaces in the parking lot every 20 minutes, as well as the weather conditions and temperature at that time that affect the number of available parking spaces. Separate these data into groups and preprocess them, including removing noise and useless samples.

[0076] Secondly, based on the collected data, the regression model is trained in batches to improve the fitting ability of the regression model. The trained regression model is saved for later use.

[0077] Subsequently, the related past covariates are used as the input of the regression model, and the residuals of the predicted results of the regression model and the parking lot vacant spaces in the most recent time period are used as the input of the optimized time series large model.

[0078] Finally, based on the optimized time series model, there will be a residual time series prediction result. This residual time series prediction result is superimposed on the regression model result, and mean smoothing compensation is added to obtain the final prediction, providing detailed results of parking space occupancy in the future custom time period.

[0079] The method proposed in the present invention has the advantages of low training cost, low data requirements, simplicity, etc. The improved TimesFM basic time series prediction model is used for data prediction, which can reduce training costs, save training time, and improve the efficiency and accuracy of time series data prediction. At the same time, the advantages of the random forest model in processing nonlinear relationships are used to further improve the prediction performance of the model. In the case of limited computing resources, efficient time series data prediction is achieved through a lightweight model, which significantly reduces the consumption and cost of computing resources.

[0080] In addition, the embodiment of the present invention also provides a time series data prediction system based on time series large model optimization, such as Figure 3 As shown, the time series data prediction system 300 based on time series large model optimization specifically includes:

[0081] The data collection module 310 is used to collect the target time series data and the corresponding past covariate data set within the preset historical time period of the scenario to be predicted, and to perform standardization processing on the collected data;

[0082] The covariate regression training and prediction module 320 is used to train the random forest regression model through the standardized target time series data; based on the standardized past covariate data set and the trained random forest regression model, predict the time series prediction data in the recent time period;

[0083] The time series large model module 330 is used to optimize the time series large model and obtain the residual time series prediction result between the time series prediction data and the actual time series data in the most recent time period through the optimized time series large model;

[0084] The compensation module 340 is used to perform error compensation on the random forest regression model according to the residual time series prediction result to obtain an optimized random forest regression model; and predict the time series prediction data within a custom time period through the optimized random forest regression model.

[0085] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0086] The above describes specific embodiments of the present invention. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0087] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present invention should be included in the protection scope of the present invention.

Claims

1. A time series data prediction method based on time series large model optimization, characterized in that: The method comprises: Collect target time series data and corresponding past covariate data sets within a preset historical time period in a parking lot scene; the target time series data at least includes a sequence of vacant parking space data within the historical time period; the past covariate data set at least includes the following data: weather conditions and temperature corresponding to the time when the vacant parking space data is located; Train the random forest regression model using the standardized target time series data; Based on the standardized past covariate dataset and the trained random forest regression model, the time series prediction data of vacant parking spaces in the recent time period is predicted; Based on the long short-term memory network LSTM, the time series large model is optimized, specifically including: adding the long short-term memory network LSTM between the input layer and the transformer layer of the TimesFM time series large model to obtain the optimized time series large model; wherein the optimized time series large model extracts the long-term features of the input data through the LSTM network, and uses the long-term features as the input data of the transformer layer; Obtaining residual time series prediction results of the time series prediction data of the vacant parking spaces in the recent time period and the actual time series data through the optimized time series big model; According to the residual time series prediction result, the random forest regression model is error compensated to obtain an optimized random forest regression model, specifically including: adding the residual time series prediction result to the output layer of the random forest regression model, and adding a mean smoothing compensation algorithm to obtain the optimized random forest regression model, so that the prediction result output by the random forest regression model is superimposed with the residual time series prediction result to achieve error compensation; The optimized random forest regression model is used to predict the time series prediction data within the custom time period to obtain the parking space occupancy prediction result within the custom time period in the parking lot scene.

2. According to claim 1, a time series data prediction method based on time series large model optimization is characterized in that: Collect the target time series data and the corresponding past covariate data set within the preset historical time period in the parking lot scene, including: Collect the target time series data within the preset historical time period in the parking lot scenario, as well as the past covariate data set with time difference; The collected target time series data and past covariate data sets are smoothed by the moving average method, and missing data are filled; The filled data is standardized and normalized to remove noise and outliers.

3. The method for predicting time series data based on time series large model optimization according to claim 1, characterized in that: The random forest regression model is trained through the standardized target time series data, including: Divide the target time series data after standardization within the preset historical time period into several batches of data according to timestamps; The random forest regression model is trained in batches according to the order of the batches using the batches of data, so that the random forest regression model can obtain the nonlinear relationship between the covariate data and the time series data.

4. The method for predicting time series data based on time series large model optimization according to claim 1, characterized in that: Based on the standardized past covariate dataset and the trained random forest regression model, the time series prediction data of vacant parking spaces in the recent time period is predicted, including: The standardized past covariate data set is input into the trained random forest regression model to perform time series data prediction, so as to obtain time series prediction data of vacant parking spaces in the parking lot within the recent time period.

5. The method for predicting time series data based on time series large model optimization according to claim 1, characterized in that: The residual time series prediction results of the time series prediction data of the vacant parking spaces in the recent time period and the actual time series data are obtained through the optimized time series big model, specifically including: Acquire actual time series data of available parking spaces in the parking lot scene in the most recent time period; Calculating the residual between the time series prediction data of the most recent time period and the actual time series data; The residual is input into the optimized time series large model to obtain the residual time series prediction result between the time series prediction data and the actual time series data.

6. The method for predicting time series data based on time series large model optimization according to claim 1, characterized in that: The optimized random forest regression model is used to predict the time series forecast data within the custom time period, specifically including: The target time series data in the current time period and the corresponding covariate data set are input into the optimized random forest regression model to obtain the time series forecast data in the future custom time period.

7. The method for predicting time series data based on time series large model optimization according to claim 6 is characterized in that: Input the target time series data in the current time period and the corresponding covariate data set into the optimized random forest regression model to obtain the time series forecast data in the future custom time period, including: according to , and obtain the prediction results of the optimized random forest regression model; in, Forecasting time series data for a future custom time period, A random forest regression model is based on the past covariate dataset within the preset historical time period The prediction results, is the actual time series data corresponding to the x' time point, The optimized time series model takes the prediction results of the random forest regression model and the residuals of the actual time series data as input, and fits and outputs the residuals of the future custom time period T, which is consistent with the length of the future data to be predicted; is the covariate dataset for the current time period.

8. A time series data prediction system based on time series large model optimization, characterized in that: The system comprises: The data collection module is used to collect target time series data and corresponding past covariate data sets within a preset historical time period in the parking lot scene; the target time series data at least includes a sequence of vacant parking space data within the historical time period; the past covariate data set at least includes the following data: weather conditions and temperature corresponding to the time when the vacant parking space data is located; The covariate regression training and prediction module is used to train the random forest regression model through the standardized target time series data; based on the standardized past covariate data set and the trained random forest regression model, the time series prediction data of the available parking spaces in the recent time period is predicted; The time series large model module is used to optimize the time series large model based on the long short-term memory network LSTM, specifically including: adding the long short-term memory network LSTM between the input layer and the transformer layer of the TimesFM time series large model to obtain the optimized time series large model; wherein the optimized time series large model extracts the long-term features of the input data through the LSTM network, and uses the long-term features as the input data of the transformer layer; through the optimized time series large model, the residual time series prediction results of the time series prediction data of the vacant parking spaces in the recent time period and the actual time series data are obtained; The compensation module is used to perform error compensation on the random forest regression model according to the residual time series prediction result to obtain an optimized random forest regression model, specifically including: adding the residual time series prediction result to the output layer of the random forest regression model, and appending a mean smoothing compensation algorithm to obtain an optimized random forest regression model, so that the prediction result output by the random forest regression model is superimposed on the residual time series prediction result to achieve error compensation; through the optimized random forest regression model, the time series prediction data within the custom time period is predicted to obtain the parking space occupancy prediction result within the custom time period in the parking lot scene.

Images