Intelligent converged terminal program upgrading method and system

Abstract

The application discloses an intelligent fusion terminal program upgrading method and system, which comprises the following steps: collecting prediction data of a data collection device according to historical data corresponding to a target program, and screening a candidate time window with stable changes according to the change trend of the prediction data; predicting various working condition data of the data collection device in the candidate time window through historical data of various working condition parameters; combining the working condition prediction data and the prediction data in the candidate time window, screening multiple sample data by using a KNN algorithm, and calculating the collected data in the candidate time window by weighting to improve the accuracy of the collected data in the candidate time window; evaluating the stability of the collected data in each candidate time window, and selecting the candidate time window with the highest stability as a time window of an upgraded target program, so that a suitable time window for upgrading the program is quickly and effectively screened, and the target program is prevented from being upgraded in a time period with large fluctuation of the collected data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of software updating, in particular to a smart converged terminal program upgrading method and system. BACKGROUND

[0002] The smart converged terminal distributes and deploys corresponding collection business APPs according to the end device asset attribution to perform information interaction with the end device, so as to realize data collection and control instruction issuing. Most smart converged terminals support online software downloading through a remote communication channel. However, when upgrading the software, the smart converged terminal needs to suspend normal work to upgrade the program, which makes the smart converged terminal unable to continue normal data collection during the software upgrading process, resulting in data loss.

[0003] The end device may cause abnormal system state operation due to its own condition or external impact during data collection. If the program is upgraded during this time period, the smart converged terminal will be unable to collect relevant data, thereby affecting subsequent abnormal analysis. Therefore, it is particularly important to select a suitable time window to upgrade the program. SUMMARY

[0004] In view of the deficiencies in the prior art, the present application provides a smart converged terminal program upgrading method and system, which can quickly and effectively screen a suitable time window to upgrade the program. The specific technical solutions are as follows:

[0005] In a first aspect, a smart converged terminal program upgrading method is provided. In a first implementable manner of the first aspect, the method comprises:

[0006] Obtaining historical collection data corresponding to a target program to perform data prediction, and preliminarily screening a time period with stable data variation trend as a candidate time window according to the predicted collection prediction data;

[0007] According to the historical data of each working condition parameter of the data collection device, predicting working condition prediction data of each working condition parameter of the data collection device within the candidate time window;

[0008] Using KNN algorithm to determine the collection data corresponding to each candidate time window by combining the working condition prediction data of each working condition parameter within the candidate time window and the collection prediction data within the candidate time window;

[0009] According to the collection data corresponding to each candidate time window, respectively evaluating the data stability degree within each candidate time window, and selecting the candidate time window with the highest stability degree to upgrade the target program.

[0010] In combination with the first implementable manner of the first aspect, in a second implementable manner of the first aspect, the candidate time window is screened according to the data variation trend, comprising:

[0011] set a sliding window based on the required upgrade duration of the target program, and intercept a plurality of time window data from the collected prediction data obtained from the trend prediction according to the set sliding window;

[0012] respectively analyze the corresponding time window data, determine the stability index of each time window, and select a time window with stable changes as a candidate time window according to the stability index.

[0013] In a third implementation manner of the first aspect, in the second implementation manner of the first aspect, the setting of the sliding window based on the upgrade duration comprises:

[0014] adopting an autocorrelation coefficient algorithm to calculate the autocorrelation coefficient corresponding to the collected prediction data, and adjusting the time step of the sliding window according to the autocorrelation coefficient with the upgrade duration as a constraint.

[0015] In a fourth implementation manner of the first aspect, in the first implementation manner of the first aspect, the working condition parameters comprise a network state, a device state, an environment state of a data collection device, and / or a concurrency of an intelligent fusion terminal.

[0016] In a fifth implementation manner of the first aspect, in the first implementation manner of the first aspect, the KNN algorithm is adopted to predict the collected data of the candidate time window, comprising:

[0017] respectively extracting features of each type of working condition prediction data and collected prediction data in the candidate time window, to obtain a data feature vector corresponding to each candidate time window;

[0018] extracting a sample feature vector corresponding to each data sample in the same time period according to the time period corresponding to the candidate time window;

[0019] combining the corresponding data feature vector and sample feature vector, respectively calculating the Euclidean distance between each candidate time window and different data samples;

[0020] selecting a plurality of data samples according to the Euclidean distance, and fusing collected sample data in the corresponding time period range in all selected data samples to calculate the collected data of the candidate time window.

[0021] In a sixth implementation manner of the first aspect, in the fifth implementation manner of the first aspect, the selecting of a plurality of data samples according to the Euclidean distance comprises:

[0022] setting the number of selected data samples according to the time step of the candidate time window.

[0023] In a seventh implementation manner of the first aspect, in the fifth implementation manner of the first aspect, the Euclidean distance between the candidate time window and the data sample is calculated, including:

[0024] According to the characteristics of the working condition parameters and the collected data and the corresponding weights, the Euclidean distance between the candidate time window and the data sample is calculated by weighting.

[0025] In an eighth implementation manner of the first aspect, in the first implementation manner of the first aspect, the stability of the collected data in the candidate time window is evaluated, including:

[0026] The stability of the collected parameters in the candidate time window is evaluated by using an autocorrelation analysis method and / or an ADF test method.

[0027] In a ninth implementation manner of the first aspect, in the first implementation manner of the first aspect, an ARIMA model is used for data prediction.

[0028] In a second aspect, an intelligent fusion terminal program upgrading system is provided, including:

[0029] A trend prediction module is configured to obtain historical collected data corresponding to a target program for data prediction, and preliminarily filter a time period with stable data variation trend as a candidate time window according to the predicted collected data variation trend.

[0030] A working condition prediction module is configured to predict working condition prediction data of each working condition parameter of a data collection device in the candidate time window according to historical data of each working condition parameter of the data collection device.

[0031] A data prediction module is configured to determine collected data corresponding to each candidate time window by using a KNN algorithm through the working condition prediction data of each working condition parameter in the candidate time window and the collected prediction data in the candidate time window.

[0032] A window decision module is configured to evaluate the data stability in each candidate time window according to the collected data corresponding to each candidate time window, and select a candidate time window with the highest stability to upgrade the target program.

[0033] Beneficial effects: the intelligent fusion terminal program upgrading method and system can quickly determine the future change trend of the collection index by predicting the historical collection data corresponding to the target program to be upgraded, and based on the future change trend of the collection index, preliminarily screen out the time period with stable device state as the candidate time window for program upgrading. Then, combined with the working condition prediction data of the collection terminal in the candidate time window and the prediction data of the collection index, the collection data in each candidate time window is accurately predicted through the KNN algorithm. Finally, the stability of the collection data of each candidate time window is evaluated respectively, and the candidate time window with the highest stability is selected as the time window for upgrading the target program, so as to upgrade the target program in the time period with stable device operation, avoiding upgrading the target program in the time period with large data fluctuation. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the specific embodiments of the present application, the drawings required to be used in the specific embodiments will be briefly introduced below. In all the drawings, the elements or parts are not necessarily drawn according to the actual proportion.

[0035] Figure 1 The flow chart of the intelligent fusion terminal program upgrading method provided by an embodiment of the present application;

[0036] Figure 2 The system block of the intelligent fusion terminal program upgrading system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0037] The embodiments of the technical solutions of the present application will be described in detail below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, but cannot limit the protection scope of the present application.

[0038] Embodiment one

[0039] As shown in the flow chart of the intelligent fusion terminal program upgrading method, the upgrading method comprises: Figure 1

[0040] Step 1, obtaining the historical collection data corresponding to the target program for data prediction, and preliminarily screening the time period with stable data change as the candidate time window according to the change trend of the predicted collection prediction data;

[0041] Step 2, predicting the working condition prediction data of each working condition parameter of the data collection device in the candidate time window according to the historical data of each working condition parameter of the data collection device;

[0042] ​Step 3, respectively, through the working condition prediction data of each working condition parameter in the alternative time window, combined with the collected prediction data in the alternative time window, the KNN algorithm is used to determine the collected data corresponding to each alternative time window;

[0043] Step 4, respectively, according to the collected data corresponding to each alternative time window, the data stability degree in each alternative time window is evaluated, and the alternative time window with the highest stability degree is selected to upgrade the target program.

[0044] Specifically, first, according to the business type of the target program to be upgraded, the corresponding historical collected data can be retrieved, such as for electric energy metering business, the historical data of the electric meter can be retrieved, and for water metering business, the historical data of the water meter can be retrieved. Since high-precision prediction models such as LSTM model require a large amount of data and computing resources, and the training time is long. The efficiency is low. Moreover, the intelligent fusion terminal undertakes many types of business, and the collected data types for different business types are different, which requires a lot of resources and time to train data prediction adapted to different business types. Therefore, the application can use ARIMA model, SARIMA model and other traditional statistical models, and only rely on the collected historical data to predict the collected prediction data of the data collection device in the future set time period. The trend of data change can reflect the working state of the data collection device or the external impact bearing condition. The change stable time period can be selected as the alternative time window from the time period.

[0045] The traditional ARIMA model, SARIMA model and other traditional statistical models have high computing efficiency and are easy to explain, and can quickly filter out multiple alternative time windows, but it is difficult to integrate external variables, the accuracy of the prediction data is limited, and the prediction data error in the selected alternative time window is large.

[0046] In order to improve the accuracy of the data in the alternative time window, the historical data of each working condition parameter of the data collection device can be retrieved first, and the working condition data of the data collection device in the alternative time window can be predicted. Then, combined with the working condition prediction data and the collected prediction data of the data collection device in the alternative time window, the KNN algorithm is used to select multiple sample data with high similarity to the alternative time window from the pre-constructed sample library, and the collected data in the alternative time window is calculated by weighting, so as to integrate multiple external variables and improve the accuracy of the collected data in the alternative time window, and provide a basis for selecting a suitable upgrade time window. Finally, the existing data stability evaluation algorithm can be used to evaluate the stability degree of the collected data corresponding to each alternative time window, and the alternative time window with the highest stability degree is selected as the time window of the target program to be upgraded, so as to quickly and effectively select a suitable time window to upgrade the program, so as to avoid upgrading the target program in the time period with large fluctuation of collected data.

[0047] In the embodiment, optionally, the candidate time window is screened according to the data variation trend, including:

[0048] A sliding window is set based on the required upgrade duration of the target program, and a plurality of time window data is intercepted from the collected prediction data obtained from the trend prediction according to the set sliding window;

[0049] The corresponding time window data is statistically analyzed respectively to determine the stability index of each time window, and the time window with stable variation is selected as the candidate time window according to the stability index.

[0050] Specifically, in step 1, a sliding window can be set based on the required upgrade duration of the target program, and a plurality of time window data is intercepted from the collected prediction data sequence obtained by statistical model prediction according to the set sliding window. Then the stability index of each time window, such as variance, standard deviation or interquartile range, can be statistically analyzed respectively, and the time window with high stability is selected as the candidate time window according to the stability index. For example, the time window with small variance, standard deviation or interquartile range is selected as the candidate time window. The algorithm of the stability index is simple, and a plurality of candidate time windows can be quickly screened through the stability index, thereby improving the efficiency of screening the time window of the upgrade target program.

[0051] In the embodiment, optionally, the working condition parameters include network status, device status, environment status of the data acquisition device and / or concurrency of the intelligent fusion terminal.

[0052] Specifically, in step 2, the working condition parameters include network status, device status, environment status of the data acquisition device and concurrency of the intelligent fusion terminal. The device status includes device temperature, working current, working voltage and the like, which reflect the working status of the data acquisition device. The environment status includes temperature and humidity of the environment where the data acquisition device is located, which will affect the working of the data acquisition device. The network status and concurrency will affect the data interaction efficiency between the data acquisition device and the intelligent fusion terminal. By fusing the network status, device status, environment status and concurrency of the intelligent fusion terminal and other external variables, the collected data in the candidate time window is predicted, thereby improving the accuracy of the collected data in the candidate time window.

[0053] In the embodiment, optionally, the KNN algorithm is used to predict the collected data of the candidate time window, including:

[0054] The feature extraction is performed on the working condition prediction data and the collected prediction data in each candidate time window respectively to obtain the data feature vector corresponding to each candidate time window;

[0055] According to the time period corresponding to the alternative time window, the sample feature vectors corresponding to the same time period of each data sample are extracted respectively;

[0056] In combination with the corresponding data feature vectors and sample feature vectors, the Euclidean distances between each alternative time window and different data samples are calculated respectively;

[0057] According to the Euclidean distances, a plurality of data samples are selected, and the collected sample data within the corresponding time period range of all selected data samples are fused to calculate the collected data of the alternative time window.

[0058] Specifically, first, according to the time period corresponding to the alternative time window, each type of working condition prediction data corresponding to the alternative time window can be extracted from the corresponding working condition prediction data, and feature extraction is performed on each type of working condition prediction data and the collected prediction data within the alternative time window extracted previously. The extracted features can include mean, median, variance, standard deviation, etc. Thus, the data feature vectors corresponding to each item of working condition data feature and collected data feature of the alternative time window are obtained.

[0059] Then, according to the time period corresponding to the alternative time window, the sample data of the same time period can be extracted from each data sample respectively, and feature extraction is performed on the sample data of the time period to obtain sample feature vectors of the same feature type.

[0060] The sample data can be historical collected data obtained by other intelligent fusion terminals of the same business type, such as historical electric meter data collected by intelligent fusion terminals covering different residential communities. It can also be historical collected data obtained by the same intelligent fusion terminal in different collection periods, such as historical collected data obtained by the intelligent fusion terminal every day in the past.

[0061] Then, according to each feature in the standardized data feature vectors and sample feature vectors, the existing multi-dimensional Euclidean distance algorithm can be used to calculate the Euclidean distance between the alternative time window and the sample data. Finally, according to the Euclidean distance between the alternative time window and different sample data, a plurality of sample data closest to the alternative time window can be selected, and the weight corresponding to each sample data can be set according to the Euclidean distance, and the collected sample data of the same time period as the alternative time window is extracted from each selected sample data for weighted average, thereby obtaining the collected data corresponding to the alternative time window. In this way, the collected data of the alternative time window can be accurately predicted without training, improving the efficiency of the upgrade time window screening. Moreover, it can adapt to different types of collected data prediction, thereby meeting the target program update of different business types.

[0062] In this embodiment, optionally, the stability of the collected data within the alternative time window is evaluated, including:

[0063] The stability of the collected parameters in the alternative time window is evaluated by using autocorrelation analysis method and / or ADF test method.

[0064] Specifically, the existing autocorrelation analysis method or ADF test method can be used alone to quickly evaluate the stability of the collected data in the alternative time window, and the most suitable alternative time window is selected from all the alternative time windows according to the autocorrelation coefficient or test statistic and P value obtained by evaluation. For example, the alternative time window with the autocorrelation coefficient closest to 1 is selected, or the alternative time window with the smallest test statistic and P value is selected. In order to improve the evaluation accuracy of the stability, the autocorrelation analysis method and the ADF test method can also be combined to evaluate the stability of the collected data in the alternative time window, and the alternative time window is selected according to the evaluation results of the two evaluation methods.

[0065] Embodiment Two

[0066] Embodiment Two is substantially the same as Embodiment One, and the main difference is that in this embodiment, the sliding window is set based on the upgrade duration, which includes:

[0067] The autocorrelation coefficient of the collected prediction data is calculated by using the autocorrelation coefficient algorithm, and the time step of the sliding window is adjusted according to the autocorrelation coefficient with the upgrade duration as the constraint.

[0068] Specifically, the time step of the sliding window needs to match the stationarity of the collected prediction data. If the fluctuation of the collected prediction data is large, a smaller time step needs to be set to quickly respond to data changes, or the collected prediction data needs to be differentiated to make it stationary. If the fluctuation of the collected prediction data is small, a larger time step can be set to capture longer trends when filtering sample data. Therefore, when setting the time step of the sliding window, the collected prediction data obtained by prediction can be first subjected to ADF test to determine whether the collected prediction data is stable. When the collected prediction data is not stable, the collected prediction data can be differentiated, and then the autocorrelation coefficient of the differentiated collected prediction data under different lag orders is calculated by using the existing autocorrelation coefficient algorithm, and the autocorrelation coefficient is plotted into an ACF graph. Finally, a plurality of significant lag orders are selected from the ACF graph, and the lag order with a time step longer than the upgrade duration is selected as the time step of the sliding window. In this way, the sliding window can be deeply matched with the time variation rule of the collected prediction data, the time-dependent capturing ability of the subsequent KNN algorithm can be improved, and the data prediction accuracy and stability of the alternative time window can be improved.

[0069] Embodiment Three

[0070] Example 3 is largely the same as Example 1, with the main difference being that multiple data samples are selected based on the Euclidean distance, including:

[0071] The number of data samples to be selected is set according to the time step of the alternative time window.

[0072] Specifically, the number of data samples has a significant impact on the performance of the KNN algorithm. Too few or too many data samples can lead to overfitting and affect computational efficiency. In this embodiment, the number of data samples can be set according to the time step of the alternative time window. When the time step is large, the KNN algorithm is prone to losing local features and rapid changing trends, potentially leading to underfitting. Therefore, a smaller number of data samples is needed in this case to effectively capture local features and avoid underfitting. Conversely, when the time step is small, a larger number of data samples can be set to avoid overfitting caused by noise and short-term fluctuations.

[0073] Example 4

[0074] Example 4 is largely the same as Example 1, with the main difference being: calculating the Euclidean distance between the candidate time window and the data sample, including:

[0075] Based on the characteristics of each operating condition parameter and the characteristics of the collected data and their corresponding weights, the Euclidean distance between the candidate time window and the data sample is calculated using a weighted average method.

[0076] Specifically, different operating parameters and data acquisition characteristics have different dimensions and varying importance to the data acquired by the data acquisition equipment. Therefore, when calculating the Euclidean distance between the acquired data and the sample data within the candidate time window, different weights can be assigned to each operating parameter and data acquisition characteristic to highlight the more important characteristics and make the distance calculation results more reasonable and accurate. For example, data acquisition characteristics directly reflect the changes in data within the candidate time window, so they can be assigned the highest weight. The equipment status of the data acquisition equipment itself reflects its working state when acquiring data, so it can be assigned a relatively high weight. Environmental parameter characteristics reflect the environmental conditions when the data acquisition equipment acquires data and have a relatively small impact on the acquired data, so they can be assigned a relatively low weight.

[0077] like Figure 2 The system block diagram shown is for the intelligent converged terminal program upgrade system. The upgrade system includes:

[0078] The trend prediction module is configured to acquire historical data corresponding to the target program for data prediction, and preliminarily select time periods with stable data changes as alternative time windows based on the predicted trend of the acquired data.

[0079] The working condition prediction module is configured to predict working condition prediction data of each working condition parameter of the data acquisition device in the candidate time window according to historical data of each working condition parameter of the data acquisition device.

[0080] The data prediction module is configured to determine the acquisition data corresponding to each candidate time window by using the KNN algorithm on the basis of the working condition prediction data of each working condition parameter in the candidate time window and the acquisition prediction data in the candidate time window.

[0081] The window decision module is configured to respectively evaluate the data stability degree in each candidate time window according to the acquisition data corresponding to each candidate time window, and select the candidate time window with the highest stability degree as the time window of the target program to be upgraded.

[0082] Specifically, the upgrading system includes a trend prediction module, a working condition prediction module, a data prediction module, and a window decision module. The trend prediction module can retrieve corresponding historical acquisition data according to the business type to which the target program to be upgraded is directed, and predict acquisition prediction data of the data acquisition device in a future set time period range by using traditional statistical models such as ARIMA and SARIMA models through the acquired historical data, and filter a time period with stable changes from the time period range as a candidate time window according to the change trend of the acquisition prediction data.

[0083] The working condition prediction module can first retrieve historical data of each working condition parameter of the data acquisition device, and predict each working condition data of the data acquisition device in the candidate time window. The data prediction module can combine the working condition prediction data and the acquisition prediction data of the data acquisition device in the candidate time window, select a plurality of sample data with high similarity to the candidate time window from a pre-constructed sample library by using the KNN algorithm, and weightedly calculate the acquisition data in the candidate time window, so as to integrate various external variables to improve the accuracy of the acquisition data in the candidate time window, and provide a basis for selecting a suitable upgrading time window. The window decision module can evaluate the stability degree of the acquisition data corresponding to each candidate time window by using an existing data stability evaluation algorithm, and select the candidate time window with the highest stability degree as the time window of the target program to be upgraded, so as to quickly and effectively filter a suitable time window to upgrade the program.

[0084] The above examples are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some or all of the technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.

Claims

1. A method for upgrading a smart converged terminal program, characterized in that, include: Historical data collected for the target program is obtained for data prediction, and time periods with stable data changes are initially selected as candidate time windows based on the predicted data change trends. Based on the historical data of various operating parameters of the data acquisition equipment, predict the operating condition data of various operating parameters of the data acquisition equipment within the alternative time window. The KNN algorithm is used to determine the collected data for each alternative time window by combining the predicted operating conditions for each operating condition parameter within the alternative time window with the collected predicted data within the alternative time window. The stability of the data within each alternative time window is evaluated based on the collected data, and the alternative time window with the highest stability is selected to upgrade the target program.

2. The intelligent converged terminal program upgrade method according to claim 1, characterized in that, Based on data trends, alternative time windows are selected, including: A sliding window is set based on the upgrade time required by the target program, and multiple time window data are extracted from the collected prediction data obtained by trend prediction according to the set sliding window. Statistical analysis was performed on the data of the corresponding time windows to determine the stability index of each time window, and time windows with stable changes were selected as alternative time windows based on the stability index.

3. The intelligent converged terminal program upgrade method according to claim 2, characterized in that, The sliding window is set based on the upgrade duration, including: The autocorrelation coefficient algorithm is used to calculate the autocorrelation coefficient corresponding to the collected prediction data, and the time step of the sliding window is adjusted according to the autocorrelation coefficient, with the upgrade duration as a constraint.

4. The intelligent converged terminal program upgrade method according to claim 1, characterized in that, The operating parameters include the network status, device status, environmental status of the data acquisition equipment, and / or the number of concurrent connections of the intelligent fusion terminal.

5. The intelligent converged terminal program upgrade method according to claim 1, characterized in that, The KNN algorithm is used to predict the collected data for the candidate time window, including: Feature extraction is performed on the prediction data and collected prediction data of various working conditions within the candidate time window to obtain the data feature vector corresponding to each candidate time window; Based on the time periods corresponding to the candidate time windows, extract the sample feature vectors of each data sample in the same time period; By combining the corresponding data feature vectors and sample feature vectors, the Euclidean distance between each candidate time window and different data samples is calculated respectively; Multiple data samples are selected based on the Euclidean distance, and the collected sample data within the corresponding time period range of all selected data samples are merged to calculate the collected data of the alternative time window.

6. The intelligent converged terminal program upgrade method according to claim 5, characterized in that, Multiple data samples are selected based on the Euclidean distance, including: The number of data samples to be selected is set according to the time step of the alternative time window.

7. The intelligent converged terminal program upgrade method according to claim 5, characterized in that, Calculate the Euclidean distance between the alternative time window and the data sample, including: Based on the characteristics of each operating condition parameter and the characteristics of the collected data and their corresponding weights, the Euclidean distance between the candidate time window and the data sample is calculated using a weighted average method.

8. The method for upgrading a smart converged terminal program according to claim 1, characterized in that, Assess the stability of data collected within the alternative time window, including: The stability of the collected parameters within the alternative time window is evaluated using autocorrelation analysis and / or ADF test.

9. The method for upgrading a smart converged terminal program according to claim 1, characterized in that, The ARIMA model is used for data prediction.

10. A smart converged terminal program upgrade system, characterized in that, include: The trend prediction module is configured to acquire historical data corresponding to the target program for data prediction, and preliminarily select time periods with stable data changes as alternative time windows based on the predicted trend of the acquired data. The operating condition prediction module is configured to predict the operating condition data of the data acquisition equipment within a selected time window based on the historical data of the various operating condition parameters of the data acquisition equipment. The data prediction module is configured to predict the operating conditions within the alternative time window using each operating condition parameter, and then use the KNN algorithm to determine the corresponding collected data for each alternative time window in combination with the collected prediction data within the alternative time window. The window decision module is configured to evaluate the data stability within each alternative time window based on the collected data corresponding to each alternative time window, and select the alternative time window with the highest stability to upgrade the target program.

Images