Power and environment data anomaly prediction method and system based on time sequence analysis
By performing periodic feature extraction and prediction model establishment of the environmental parameter time series in the dynamic ring monitoring system, the problems of low prediction accuracy and insufficient periodic feature processing in the prior art are solved, and the prediction accuracy of dynamic ring data and the stability of the system are improved.
Patent Information
- Application Number
- CN202411651620.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-19
- Publication Date
- 2025-02-25
AI Technical Summary
The existing dynamic ring monitoring system has low prediction accuracy in data abnormality prediction and insufficient processing of periodic features, resulting in false alarm information or missed reports, reducing the accuracy and reliability of prediction.
By preprocessing, decomposing and reconstructing the environmental parameters time series in the dynamic ring monitoring system, periodic features are extracted, and prediction models are established using long-term and short-term memory networks and graph neural networks to improve the prediction accuracy of dynamic ring data.
It improves the accuracy and reliability of dynamic ring data prediction, can understand the relationship between environmental parameters and dynamic ring data more comprehensively, reduces false alarms and missed reports of early warning information, and improves operation and maintenance efficiency and system stability.
Smart Images
Figure CN119512047A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data monitoring technology, and in particular to a method and system for predicting anomaly of dynamic environment data based on time series analysis. Background Art
[0002] Dynamic environment monitoring systems, as an indispensable part of modern data centers, communication base stations and other key facilities, play a vital role. They are widely used in various occasions to achieve real-time monitoring of the operating status and environmental parameters of various equipment. These parameter data include but are not limited to current, voltage, temperature and humidity, water immersion status, equipment online and offline status, and operating failures, which together constitute a comprehensive portrait of the operating status of the facility. What is particularly important is that these data do not exist statically, but show distinct time series characteristics, that is, they will continue to change over time. This dynamism not only reflects the real-time changes in the operating status of the facility, but also provides a valuable information basis for subsequent abnormal predictions.
[0003] In practical applications, the data anomaly prediction function of the dynamic environment monitoring system is particularly important. By timely discovering and predicting data anomalies, operation and maintenance personnel can quickly locate potential equipment failures or environmental problems, and take necessary preventive measures to avoid the occurrence or expansion of failures. This has far-reaching significance for ensuring the stable operation of the system, extending the service life of equipment, and reducing maintenance costs.
[0004] However, although the dynamic environment monitoring system plays an important role in predicting data anomalies, the prediction methods in the existing technology have many shortcomings. Among them, the low prediction accuracy and insufficient processing of periodic characteristics are the two most prominent problems. Periodic characteristics refer to the data repeating similar patterns or trends at certain time intervals in a continuous time period. This periodicity may be caused by the operating rules of the equipment, changes in environmental factors or the influence of other external factors.
[0005] Many existing prediction methods often ignore the periodic characteristics of data when processing data, or simply treat it as a linear trend. This processing method cannot fully utilize the periodic information in the data, thus limiting the improvement of prediction accuracy. This leads to false alarms or omissions of warning information. Insufficient processing of periodic characteristics may cause the system to be unable to fully utilize the time information in the data, thereby reducing the accuracy and reliability of the prediction. If the prediction model cannot accurately reflect the periodic changes of environmental parameters, the prediction results may have large deviations, thereby affecting the warning effect of the dynamic environment monitoring system. Summary of the invention
[0006] The purpose of the present invention is to provide a method and system for predicting anomalies in dynamic environment data based on time series analysis, so as to improve the prediction accuracy of dynamic environment data by performing periodic analysis on environmental parameter data.
[0007] In order to solve the above technical problems, the present invention provides the following technical solution: a method for predicting abnormality of dynamic environment data based on time series analysis, the steps of which include:
[0008] Acquire an environmental parameter time series from the dynamic environment monitoring system, and preprocess the acquired environmental parameter time series, wherein the environmental parameters include temperature, humidity, voltage, and current; the preprocessing includes normalization processing; by preprocessing the environmental parameter time series (such as normalization processing), the dimension differences between different parameters can be eliminated, making the data more consistent and comparable;
[0009] Decompose and reconstruct the preprocessed environmental parameter time series to obtain periodic characteristics;
[0010] A prediction model is established based on the periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment data;
[0011] The dynamic environment data is predicted according to the prediction model, and the contribution of the periodic characteristics of each environmental parameter sequence to the dynamic environment data is determined to determine the environmental parameters that affect the abnormal dynamic environment data.
[0012] Periodicity refers to the fact that similar patterns or trends recur in data at certain time intervals over a continuous period of time. Environmental parameters, such as temperature, humidity, and air pressure, often change periodically with changes in seasons, weather, time, and other factors. This periodic change has certain regularity and predictability, but it is also affected by a variety of factors, making its change process complicated. Due to the existence of periodic changes, the prediction model needs to capture this change pattern more accurately to improve the accuracy of the prediction.
[0013] According to the above technical solution, the periodic feature extraction step includes:
[0014] Decomposing the preprocessed environmental parameter time series by VMD to obtain several environmental parameter time series, and learning the periodic characteristics of each environmental parameter time series by long short-term memory network; the VMD decomposition is variational mode decomposition;
[0015] The periodic characteristics of each sub-environmental parameter time series are weighted to obtain the equivalent periodic characteristics of the sub-environmental parameter time series;
[0016] Inputting the preprocessed environmental parameter time series into a long short-term memory network to obtain original periodic features, and weighting the original periodic features to obtain equivalent original periodic features;
[0017] The equivalent periodic characteristics and original periodic characteristics of all sub-environmental parameter time series are reconstructed to obtain the first periodic characteristics;
[0018] Using a graph neural network to learn the preprocessed environmental parameter time series to obtain a second periodic feature;
[0019] The first periodic feature and the second periodic feature together constitute the periodic feature of the environmental parameter time series.
[0020] The regularity of each variable in the environmental parameter time series is complex and may be affected by multiple other variables. Therefore, it is often difficult to intuitively observe the periodicity of the environmental parameter time series. The periodicity of the environmental parameter time series may be the superposition of multiple periodic characteristics and has a certain degree of complexity.
[0021] Therefore, when analyzing the periodic characteristics of environmental parameter time series, it is necessary to consider decomposing the environmental parameter time series, and then use long short-term memory networks or neural networks to learn the periodic characteristics, and then integrate the periodicity of each decomposed sequence to determine the overall periodic characteristics.
[0022] According to the above technical solution, the step of calculating the equivalent periodic characteristics includes:
[0023] Obtain the frequency and peak value corresponding to each periodic feature of the secondary environmental parameter time series;
[0024] Calculate the weight of each periodic feature of the secondary environmental parameter time series; the weight of the periodic feature
[0025]
[0026] In the formula, represents the weight corresponding to the periodic feature i, H i represents the peak value corresponding to the periodic feature i, w i represents the frequency corresponding to the periodic feature i, H j represents the peak value corresponding to the periodic feature j, w j represents the frequency corresponding to the periodic feature j, and k represents the number of periodic features contained in the sub-environmental parameter time series;
[0027] The periodic characteristics of the corresponding sub-environmental parameter time series are weighted according to the weight of each periodic characteristic to obtain the equivalent periodic characteristics of the sub-environmental parameter time series; the weighting is:
[0028]
[0029] In the formula, a n represents the equivalent periodic feature corresponding to the sub-environmental parameter time series n, k represents the number of periodic features contained in the sub-environmental parameter time series, represents the weight corresponding to the periodic feature j in the environmental parameter time series n, a j Represents the periodic feature j in the environmental parameter time series n.
[0030] According to the above technical solution, the first periodic feature is:
[0031]
[0032] Where P1 represents the first periodic characteristic, a n represents the equivalent periodic characteristics corresponding to the sub-environmental parameter time series n, a0 represents the equivalent original periodic characteristics, and m represents the number of sub-environmental parameter time series.
[0033] According to the above technical solution, the preprocessed environmental parameter time series is input into an adaptive graph network structure to obtain a graph adjacency matrix, and multi-scale spatiotemporal graph convolution is used to learn the time and variable dependency in the graph adjacency matrix to obtain the second periodic feature.
[0034] Environmental parameter time series data are all multivariate. When analyzing multivariate time series, we should not only consider the change pattern of a single variable over time, but also the mutual influence between variables. Therefore, the graph adjacency matrix method can be used to consider the mutual influence between variables.
[0035] According to the above technical solution, the prediction model building step includes:
[0036] The periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment anomaly data are divided into a training set and a validation set;
[0037] Training a prediction model using a neural network model based on the training set;
[0038] Based on the validation set, the coefficient of determination and the root mean square error of the trained prediction model are calculated, and the coefficient of determination and the root mean square error are used as the judgment criteria to determine whether the prediction model meets the prediction accuracy. If the prediction accuracy is not met, the interpretable Bayesian convolutional network training is continued to be used to optimize and adjust the prediction model until the prediction accuracy is met.
[0039] According to the above technical scheme, the periodic feature sequence after reconstructing the environmental parameter time series is input into the prediction model to predict the dynamic environment data, and it is determined whether the predicted dynamic environment data contains abnormal data. If so, the SHAP interpretation algorithm is used to obtain the contribution of the periodic characteristics after reconstructing the environmental parameter time series to the changes in the dynamic environment data. According to the contribution of the periodic characteristics after reconstructing the environmental parameter time series to the abnormal changes in the dynamic environment data, the main environmental parameters of the abnormal dynamic environment data are determined.
[0040] Another embodiment, a dynamic environment data anomaly prediction system based on time series analysis, includes:
[0041] A data acquisition module is used to acquire a time series of environmental parameters from the dynamic environment monitoring system and preprocess the acquired time series of environmental parameters, wherein the environmental parameters include temperature, humidity, voltage, and current; the preprocessing includes normalization processing;
[0042] The periodic feature reconstruction module is used to decompose and reconstruct the preprocessed environmental parameter time series to obtain periodic features;
[0043] The prediction model building module builds a prediction model based on the periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment data;
[0044] The traceability module is used to predict dynamic environmental data based on the prediction model, the contribution of the periodic characteristics of each environmental parameter sequence to the dynamic environmental data, and determine the environmental parameters that affect the abnormal dynamic environmental data.
[0045] According to the above technical solution, the periodic feature reconstruction module performs the following steps:
[0046] Decomposing the preprocessed environmental parameter time series by VMD to obtain several sub-environmental parameter time series, and learning the periodic characteristics of each sub-environmental parameter time series by long short-term memory network;
[0047] The periodic characteristics of each sub-environmental parameter time series are weighted to obtain the equivalent periodic characteristics of the sub-environmental parameter time series;
[0048] Inputting the preprocessed environmental parameter time series into a long short-term memory network to obtain original periodic features, and weighting the original periodic features to obtain equivalent original periodic features;
[0049] The equivalent periodic characteristics and original periodic characteristics of all sub-environmental parameter time series are reconstructed to obtain the first periodic characteristics;
[0050] Using a graph neural network to learn the preprocessed environmental parameter time series to obtain a second periodic feature;
[0051] The first periodic feature and the second periodic feature together constitute the periodic feature of the environmental parameter time series.
[0052] It also includes another technical solution, a computer-readable storage medium, in which computer execution instructions are stored. When the computer execution instructions are executed by a computer, the dynamic environment data anomaly prediction method based on time series analysis as described in any of the above technical solutions is implemented.
[0053] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: by analyzing and extracting the periodic characteristics of the environmental parameter time series, and using the periodic characteristics of the environmental parameter time series to predict the dynamic environment data, the present invention can more comprehensively understand the relationship between environmental parameters and dynamic environment data, and improve the stability and robustness of the prediction system. By determining the contribution of each environmental parameter sequence periodic characteristic to the dynamic environment data, it can be determined which environmental parameters have the greatest impact on the abnormal dynamic environment data. This helps operation and maintenance personnel to focus on these key parameters in a targeted manner, reasonably allocate maintenance resources, and improve operation and maintenance efficiency. By predicting the abnormal dynamic environment data in advance and taking corresponding preventive measures, the downtime and maintenance costs caused by equipment failure or data abnormalities can be reduced. At the same time, it can effectively monitor and predict abnormal changes in environmental parameters to improve the stability and security of the data center. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0055] Figure 1 It is a flow chart of the steps of the method for predicting abnormal dynamic environment data based on time series analysis of the present invention;
[0056] Figure 2 It is a flow chart of the periodic feature extraction steps of the environmental parameter time series of the present invention. DETAILED DESCRIPTION
[0057] 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. 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.
[0058] Embodiment 1: The present invention provides a technical solution for a method for predicting abnormal dynamic environment data based on time series analysis, the steps of which include:
[0059] S1. Obtaining a time series of environmental parameters from a dynamic environment monitoring system, and preprocessing the obtained time series of environmental parameters, where the environmental parameters include but are not limited to temperature, humidity, voltage, and current; the preprocessing includes normalization processing.
[0060] S2. Decompose and reconstruct the preprocessed environmental parameter time series to obtain periodic characteristics. The specific execution steps include:
[0061] S201, decomposing the preprocessed environmental parameter time series by VMD (variational mode decomposition) to obtain several sub-environmental parameter time series, and learning the periodic characteristics of each sub-environmental parameter time series by long short-term memory network; the variational mode decomposition is an adaptive filter bank method for processing non-stationary signals, and the main advantage of VMD is that it can automatically determine the number of modes, thereby avoiding the problem of manually setting the number of modes in traditional methods;
[0062] S202, weighting the periodic characteristics of each secondary environmental parameter time series to obtain equivalent periodic characteristics of the secondary environmental parameter time series;
[0063] S203, inputting the preprocessed environmental parameter time series into a long short-term memory network to obtain original periodic features, and weighting the original periodic features to obtain equivalent original periodic features;
[0064] S204, reconstructing the equivalent periodic characteristics and original periodic characteristics of all secondary environmental parameter time series to obtain a first periodic characteristic; the first periodic characteristic:
[0065]
[0066] Where P1 represents the first periodic characteristic, a n represents the equivalent periodic characteristics corresponding to the sub-environmental parameter time series n, a0 represents the equivalent original periodic characteristics, and m represents the number of sub-environmental parameter time series;
[0067] S205, using a graph neural network to learn the preprocessed environmental parameter time series to obtain a second periodic feature, that is, inputting the preprocessed environmental parameter time series into an adaptive graph network structure to obtain a graph adjacency matrix, and using multi-scale spatiotemporal graph convolution to learn the time and variable dependency in the graph adjacency matrix to obtain the second periodic feature;
[0068] S206: The first periodic feature and the second periodic feature together form a periodic feature of the environmental parameter time series.
[0069] The calculation steps of the equivalent periodic characteristics include:
[0070] Obtain the frequency and peak value corresponding to each periodic feature of the secondary environmental parameter time series;
[0071] Calculate the weight of each periodic feature of the secondary environmental parameter time series; the weight of the periodic feature
[0072]
[0073] In the formula, represents the weight corresponding to the periodic feature i, H i represents the peak value corresponding to the periodic feature i, w i represents the frequency corresponding to the periodic feature i, H j represents the peak value corresponding to the periodic feature j, w j represents the frequency corresponding to the periodic feature j, and k represents the number of periodic features contained in the sub-environmental parameter time series;
[0074] The periodic characteristics of the corresponding sub-environmental parameter time series are weighted according to the weight of each periodic characteristic to obtain the equivalent periodic characteristics of the sub-environmental parameter time series; the weighting is:
[0075]
[0076] In the formula, a n represents the equivalent periodic feature corresponding to the sub-environmental parameter time series n, k represents the number of periodic features contained in the sub-environmental parameter time series, represents the weight corresponding to the periodic feature j in the environmental parameter time series n, a j Represents the periodic feature j in the environmental parameter time series n.
[0077] S3, establishing a prediction model based on the periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment data, the specific execution steps include: obtaining the periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment abnormal data, and dividing them into a training set and a validation set;
[0078] Training a prediction model using a neural network model based on the training set;
[0079] Based on the validation set, the coefficient of determination and the root mean square error of the trained prediction model are calculated, and the coefficient of determination and the root mean square error are used as the judgment criteria to determine whether the prediction model meets the prediction accuracy. If the prediction accuracy is not met, the interpretable Bayesian convolutional network training is continued to be used to optimize and adjust the prediction model until the prediction accuracy is met.
[0080] S4. Predict the dynamic environment data according to the prediction model, and determine the contribution of the periodic characteristics of each environmental parameter sequence to the dynamic environment data, and determine the environmental parameters that affect the abnormal dynamic environment data;
[0081] Among them, the periodic feature sequence after reconstructing the environmental parameter time series is input into the prediction model to predict the dynamic environment data, and it is determined whether the predicted dynamic environment data contains abnormal data. If so, the SHAP interpretation algorithm is used to obtain the contribution of the periodic characteristics after reconstructing the environmental parameter time series to the changes in the dynamic environment data. According to the contribution of the periodic characteristics after reconstructing the environmental parameter time series to the abnormal changes in the dynamic environment data, the main environmental parameters of the abnormal dynamic environment data are determined.
[0082] Among them, according to the actual operation of the computer room and business needs, reasonable alarm thresholds are set for each predicted dynamic environment data. Once the data exceeds or falls below these thresholds, the system will automatically trigger the alarm mechanism and determine that the dynamic environment data is abnormal.
[0083] Embodiment 2: A dynamic environment data anomaly prediction system based on time series analysis, comprising:
[0084] A data acquisition module is used to acquire a time series of environmental parameters from the dynamic environment monitoring system and preprocess the acquired time series of environmental parameters, wherein the environmental parameters include temperature, humidity, voltage, and current; the preprocessing includes normalization processing;
[0085] The periodic feature reconstruction module is used to decompose and reconstruct the preprocessed environmental parameter time series to obtain periodic features. The specific execution steps include: decomposing the preprocessed environmental parameter time series using VMD to obtain several sub-environmental parameter time series, and using a long short-term memory network to learn the periodic features of each sub-environmental parameter time series;
[0086] The periodic characteristics of each sub-environmental parameter time series are weighted to obtain the equivalent periodic characteristics of the sub-environmental parameter time series;
[0087] Inputting the preprocessed environmental parameter time series into a long short-term memory network to obtain original periodic features, and weighting the original periodic features to obtain equivalent original periodic features;
[0088] The equivalent periodic characteristics and original periodic characteristics of all sub-environmental parameter time series are reconstructed to obtain the first periodic characteristics;
[0089] Using a graph neural network to learn the preprocessed environmental parameter time series to obtain a second periodic feature;
[0090] The first periodic feature and the second periodic feature together constitute the periodic feature of the environmental parameter time series.
[0091] The prediction model building module uses a neural network or a long short-term memory network to build a prediction model based on the periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic ring data;
[0092] The traceability module is used to predict dynamic environmental data based on the prediction model, and use the SHAP interpretation algorithm to obtain the contribution of the periodic characteristics of each environmental parameter sequence to the dynamic environmental data, and then determine the environmental parameters that affect the abnormal dynamic environmental data.
[0093] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0094] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for predicting abnormal dynamic environment data based on time series analysis, characterized in that: The prediction method steps include: Acquire a time series of environmental parameters from a dynamic environment monitoring system, and preprocess the acquired time series of environmental parameters, wherein the environmental parameters include temperature, humidity, voltage, and current; the preprocessing includes normalization processing; Decompose and reconstruct the preprocessed environmental parameter time series to obtain periodic characteristics; A prediction model is established based on the periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment data; The dynamic environment data is predicted according to the prediction model, and the contribution of the periodic characteristics of each environmental parameter sequence to the dynamic environment data is determined to determine the environmental parameters that affect the abnormal dynamic environment data.
2. The method for predicting abnormal dynamic environment data based on time series analysis according to claim 1 is characterized in that: The step of extracting the periodic features comprises: The preprocessed environmental parameter time series is decomposed by VMD to obtain several environmental parameter time series, and the periodic characteristics of each environmental parameter time series are learned by long short-term memory network; VMD decomposition is variational mode decomposition; The periodic characteristics of each sub-environmental parameter time series are weighted to obtain the equivalent periodic characteristics of the sub-environmental parameter time series; Inputting the preprocessed environmental parameter time series into a long short-term memory network to obtain original periodic features, and weighting the original periodic features to obtain equivalent original periodic features; The equivalent periodic characteristics and original periodic characteristics of all sub-environmental parameter time series are reconstructed to obtain the first periodic characteristics; Using a graph neural network to learn the preprocessed environmental parameter time series to obtain a second periodic feature; The first periodic feature and the second periodic feature together constitute the periodic feature of the environmental parameter time series.
3. The method for predicting abnormal dynamic environment data based on time series analysis according to claim 2 is characterized in that: The calculation steps of the equivalent periodic characteristics include: Obtain the frequency and peak value corresponding to each periodic feature of the secondary environmental parameter time series; Calculate the weight of each periodic feature of the secondary environmental parameter time series; the weight of the periodic feature In the formula, represents the weight corresponding to the periodic feature i, H i represents the peak value corresponding to the periodic feature i, w i represents the frequency corresponding to the periodic feature i, H j represents the peak value corresponding to the periodic feature j, w j represents the frequency corresponding to the periodic feature j, and k represents the number of periodic features contained in the sub-environmental parameter time series; The periodic characteristics of the corresponding sub-environmental parameter time series are weighted according to the weight of each periodic characteristic to obtain the equivalent periodic characteristics of the sub-environmental parameter time series; the weighted weighting is: In the formula, a n represents the equivalent periodic feature corresponding to the sub-environmental parameter time series n, k represents the number of periodic features contained in the sub-environmental parameter time series, represents the weight corresponding to the periodic feature j in the environmental parameter time series n, a j Represents the periodic feature j in the environmental parameter time series n.
4. The method for predicting abnormal dynamic environment data based on time series analysis according to claim 2 is characterized in that: The first periodic feature: Where P1 represents the first periodic characteristic, a n represents the equivalent periodic characteristics corresponding to the sub-environmental parameter time series n, a0 represents the equivalent original periodic characteristics, and m represents the number of sub-environmental parameter time series.
5. The method for predicting abnormal dynamic environment data based on time series analysis according to claim 2 is characterized in that: The preprocessed environmental parameter time series is input into an adaptive graph network structure to obtain a graph adjacency matrix, and multi-scale spatiotemporal graph convolution is used to learn the time and variable dependency in the graph adjacency matrix to obtain the second periodic feature.
6. The method for predicting abnormal dynamic environment data based on time series analysis according to claim 1 is characterized in that: The prediction model building step includes: The periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment anomaly data are divided into a training set and a validation set; Training a prediction model using a neural network model based on the training set; Based on the validation set, the coefficient of determination and the root mean square error of the trained prediction model are calculated, and the coefficient of determination and the root mean square error are used as the judgment criteria to determine whether the prediction model meets the prediction accuracy. If the prediction accuracy is not met, the interpretable Bayesian convolutional network training is continued to be used to optimize and adjust the prediction model until the prediction accuracy is met.
7. The method for predicting abnormal dynamic environment data based on time series analysis according to claim 1 is characterized in that: The periodic feature sequence reconstructed from the environmental parameter time series is input into the prediction model to predict the dynamic environment data, and it is determined whether the predicted dynamic environment data contains abnormal data. If so, the SHAP interpretation algorithm is used to obtain the contribution of the periodic characteristics reconstructed from the environmental parameter time series to the changes in the dynamic environment data. According to the contribution of the periodic characteristics reconstructed from the environmental parameter time series to the abnormal changes in the dynamic environment data, the main environmental parameters of the abnormal dynamic environment data are determined.
8. A dynamic environment data anomaly prediction system based on time series analysis, characterized in that: include: A data acquisition module is used to acquire a time series of environmental parameters from the dynamic environment monitoring system and preprocess the acquired time series of environmental parameters, wherein the environmental parameters include temperature, humidity, voltage, and current; the preprocessing includes normalization processing; The periodic feature reconstruction module is used to decompose and reconstruct the preprocessed environmental parameter time series to obtain periodic features; The prediction model building module builds a prediction model based on the periodic characteristics of the environmental parameter time series reconstructed and the corresponding dynamic environment data; The traceability module is used to predict dynamic environmental data based on the prediction model, the contribution of the periodic characteristics of each environmental parameter sequence to the dynamic environmental data, and determine the environmental parameters that affect the abnormal dynamic environmental data.
9. The dynamic environment data anomaly prediction system based on time series analysis according to claim 8 is characterized in that: The periodic feature reconstruction module execution steps include: Decomposing the preprocessed environmental parameter time series by VMD to obtain several sub-environmental parameter time series, and learning the periodic characteristics of each sub-environmental parameter time series by long short-term memory network; The periodic characteristics of each sub-environmental parameter time series are weighted to obtain the equivalent periodic characteristics of the sub-environmental parameter time series; Inputting the preprocessed environmental parameter time series into a long short-term memory network to obtain original periodic features, and weighting the original periodic features to obtain equivalent original periodic features; The equivalent periodic characteristics and original periodic characteristics of all sub-environmental parameter time series are reconstructed to obtain the first periodic characteristics; Using a graph neural network to learn the preprocessed environmental parameter time series to obtain a second periodic feature; The first periodic feature and the second periodic feature together constitute the periodic feature of the environmental parameter time series.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by a computer, the method for predicting anomalies in dynamic environment data based on time series analysis as described in any one of claims 1 to 7 is implemented.