Energy management system data filling method and architecture based on time sequence analysis

Abstract

The invention discloses an energy management system data filling method and architecture based on time sequence analysis, and the method achieves the intelligent prediction and filling of a data missing segment in an energy management system through constructing a filling model based on a naive Bayes algorithm, and improves the data integrity and continuity. The system firstly dynamically manages historical data and new data through a sliding window mechanism to ensure that the model can fully consider the change trend of time sequence data in the filling process. The filling model combines the distribution characteristics of historical data and current data during initialization, and endows the data in different time periods with weights, so that the model can adapt to the equipment state and environment change. And after the data missing segment is detected, the system immediately calls the filling model to carry out prediction filling, replaces the original missing data with the result and writes the result into the database, so that the accuracy of subsequent analysis and anomaly detection is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of energy management system data filling, and in particular, relates to an energy management system data filling method and architecture based on time series analysis. Background Art

[0002] In the field of new energy applications, the energy management system undertakes tasks such as data collection and data visualization in the energy dispatch process, and provides assistance for decision-making in the energy dispatch process. However, the application of functions such as early warning of equipment failures and energy consumption data prediction based on energy data in energy management systems is still in its infancy, and the deployment and implementation of functions in this regard still requires more exploration. In order to further explore the deployment and implementation of intelligent applications of energy management systems, the present invention proposes a fault detection method for energy management systems based on time series analysis, designs and implements an energy data filling and anomaly detection system, and the system will serve as a subsystem of the energy management system to provide services for the energy management system, thereby improving the functional system of the original system.

[0003] In actual operation, due to the diversity of application scenarios, the large number of sensors on terminal devices and the high sampling frequency, the data accumulation speed is extremely fast, and the generated data has the characteristics of time series arrangement, high data dimension, a large amount of unlabeled data, and complex mechanism models. Factors such as network fluctuations and electromagnetic interference often cause the collection of some equipment operation data to be interrupted. In severe cases, the interruption period is as long as several hours, which seriously affects the integrity of the collected data. In general, the mean, mode filling or other simple filling methods are used to deal with missing data. However, in a complex industrial environment, the missing data does not only have simple numerical changes, and the use of a single simple missing data filling scheme will affect the reliability of the collected data. Especially in the data statistical environment of multiple devices and multi-level meters, the uncertainty of some data can easily cause misunderstandings for data analysts. On the other hand, factors such as the aging and replacement of terminal equipment components, changes in working conditions and climate conditions will also cause the concepts implied by the equipment operation status to change, resulting in concept drift. Therefore, the research on system anomaly detection also needs to consider the characteristics of streaming data, update it in a timely manner, adapt to data that may change, and ensure the effectiveness of the model.

[0004] However, for the general problem of missing data, there are several common solutions, namely: ① directly ignore the missing data and other data points related to it; ② manually analyze and fill the missing data; ③ use a global constant to fill the missing values; ④ use the mean, mode or median to fill the missing values; ⑤ use the attribute center measurement of samples similar to the data points to be filled; ⑥ use regression methods, decision trees and other calculation methods to determine the most likely value to fill the missing data. There are many sensors in the energy management system, and the missing data is often a large continuous gap of unequal length. In addition, the aging and replacement of equipment components, changes in working conditions and climate conditions will cause sensor sampling drift, which places high requirements on the efficiency and adaptability of the data filling algorithm used. Although the current research methods have their own advantages, they still cannot achieve the dual advantages of filling accuracy and computing time when dealing with the above problems.

[0005] From a statistical point of view, normal data all come from the same generation mechanism, and vice versa, they deviate from this generation mechanism. A complete anomaly detection process can be summarized into three stages: parameterization stage, training stage, and detection stage. In the parameterization stage, the data is preprocessed into a preset format so that the data can be used in the target system; in the training stage, a certain anomaly detection method is used to train the model based on normal or abnormal data according to the needs; in the detection stage, the observed value is calculated using the obtained model and compared with the predefined value. If the deviation exceeds the threshold, an alarm is triggered. At present, most anomaly detection models are based on this basic process. In general, the research on anomaly detection technology can be divided into three categories: clustering-based anomaly detection, classification-based anomaly detection, and hybrid model-based anomaly detection. However, in the study of the application of these methods, it is found that due to the noise problems caused by factors such as current and voltage instability and communication anomalies, the concept drift phenomenon of energy process data in the form of streaming data, and the high real-time requirements for anomaly feedback in actual operation, these methods cannot be effectively applied.

[0006] In view of this, the present invention is proposed. Summary of the invention

[0007] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide an energy management system data filling method and architecture based on time series analysis, which solves the problems raised in the above-mentioned background technology.

[0008] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0009] A method for filling data in an energy management system based on time series analysis comprises the following steps:

[0010] Step (1): Obtain operating status data and energy consumption data from sensors and devices in the energy management system, store the data in the system database, and mark the time periods when data collection is interrupted or missing as targets for subsequent filling;

[0011] Step (2): Clean and format the collected time series data, then use the sliding window mechanism to establish a dynamic data warehouse to manage historical data and new data in real time;

[0012] Step (3): Construct a filling model based on the Naive Bayes algorithm and combine historical data with current data distribution when initializing the model, assign weights to data in different time periods, and adapt to data change trends;

[0013] Step (4): When a data missing segment is detected, the filling model is called to predict and fill the marked missing segment, and the filling result is written into the database to replace the missing data.

[0014] Optionally, before detecting the data missing segment, an isolation forest algorithm model is established and initialized based on historical data to provide a benchmark model for detecting anomalies in energy flow data, wherein the steps of establishing and initializing the isolation forest algorithm model are:

[0015] Extract historical data from the energy management database and perform denoising, normalization and missing value processing on the historical data to generate a clean data set as the training input of the model. Then, set the number of subsamples n of the isolation forest algorithm model. s , number of isolated trees t, anomaly score threshold 0;

[0016] Subsamples are randomly drawn from the historical data set to construct multiple isolated trees. Each tree recursively randomly selects features and partition points until all data points are completely isolated. Then, the average path length of all isolated trees is used through the formula Calculate the anomaly score s(x) for each data point, and then compare the calculated anomaly score with the threshold θ as the benchmark for the model to determine whether the data point is abnormal, where s(x) is the anomaly score of data point x, E(H) is the average maximum depth of the tree, and is related to the number of samples n. s Related, h(x) is the path length of data point x in the isolation tree;

[0017] After the model initialization is completed, the model is deployed to the application service layer and is ready to receive future energy flow data input.

[0018] Optionally, when detecting data, enter the detection cycle at a preset time interval, read the current energy flow data, obtain the latest energy flow data from the energy database within the current detection cycle, prepare to input the anomaly detection module for analysis, use the MHSIF model to perform anomaly detection on the energy flow data within the current cycle, identify whether there are short-term sequence anomalies, mark the detected anomalies, and determine whether the current detection cycle has ended; if not, continue to read data and perform anomaly detection until the end of the cycle. At the end of the detection cycle, the system determines whether the anomaly rate within the cycle exceeds a preset threshold to determine whether the detection model needs to be updated. If the anomaly rate exceeds the threshold, use the data of the current detection cycle to update the model so that the model adapts to the new system status and data changes.

[0019] Optionally, the steps for establishing a dynamic data warehouse using a sliding window mechanism are:

[0020] Step 201: Set the time window size W t and sampling frequency f s , then, according to the set time window size W t and sampling frequency f s Initialize the sliding window mechanism, window size W t Indicates the length of time that can be retained in the sliding window, that is, W t =n·T s , where n is the number of sampling points in the window, T s is the time span of a single data point;

[0021] Step 202: In the sliding window, whenever new data D is received t Then, the data is stored in the dynamic data warehouse in chronological order. t When entering, the sliding window mechanism automatically checks whether there is data outside the window range W t , data outside the window D t -W t Will be removed according to the first-in-first-out rule to ensure that the warehouse only retains the most recent W t Data within a time period;

[0022] Step 203: Each time after removing old data and loading new data, the data in the window is reordered by time. As time goes by, the window keeps rolling. Whenever new data enters, the start and end points of the window move at the same time. The update formula for window rolling is: W t (t+1)=W t (t)+D t+1 -D t-n , which means that each time the window is updated, the new data D t+1 Add the earliest data D t-nFinally, the data managed by the sliding window is stored in the dynamic data warehouse to complete the establishment of the dynamic data warehouse.

[0023] Optionally, the steps of constructing a filling model based on the Naive Bayes algorithm and combining historical data with current data distribution when initializing the model and assigning weights to data in different time periods are as follows:

[0024] Step 301: Historical data D H ={d 1 , d 2 , dots, d m} and current time period data D c ={d m+1 , dots, d n}, and then, using the naive Bayes principle, calculate the prior probability P(T H ) and P(T C ) to quantify the importance of data in each time period, which is expressed as: In the formula, |D H | and |D C |Indicates the number of samples of historical data and current data;

[0025] Step 302: According to the data distribution in different time periods, weights w are assigned to the historical data and the current data respectively. H and w C , its expression is: H =P(T H )·C H , w C =P(T C )·C C Among them, C H and C C It is based on the prior probability of the time period and the credibility coefficient of the data;

[0026] Step 303: Calculate the conditional probability based on the weighted data and construct a naive Bayes imputation model, which is expressed as: Among them, P(x|D) is the probability of filling the missing value x after integrating historical and current data.

[0027] Optionally, when a data missing segment is detected, the filling model is called to predict and fill the marked missing segment, and the filling result is written into the database to replace the missing data in the following steps:

[0028] Scan the time series data to find the missing segments in the data. According to the marked missing segments, call the naive Bayes filling model to calculate the predicted value of each missing point. For each missing point, calculate the probability of all values ​​and select the value corresponding to the probability as the filling result. Then, replace the corresponding missing values ​​with the filling results of all missing points one by one, and update the filled data to the database.

[0029] The architecture of the energy management system data filling method based on time series analysis is characterized by comprising:

[0030] Data access layer, the data access layer consists of the collection server database and the application server database. The collection server database is used to store the original energy data collected by the sensor, and the application server database is used to store the complete data after preprocessing and filling;

[0031] Data preprocessing module, which obtains energy data from the acquisition server database, cleans the data, converts the format and aligns the time to ensure the quality and consistency of the data, and stores the preprocessed data in the application server database;

[0032] Missing data filling module, when missing data is detected, the missing data filling module calls the filling model based on the naive Bayes algorithm to fill the missing segment and update the filled data to the application server database;

[0033] Anomaly detection and early warning module, which is used to read complete data from the application server database, detect abnormal states in energy flow data based on the isolation forest algorithm, trigger an alarm when an anomaly is detected, and record abnormal data segments;

[0034] Data display module,The data display module is used to visualize the filled data and anomaly detection results through the presentation layer, and to display the abnormal data segments and filling status to managers.

[0035] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:

[0036] The present invention realizes intelligent prediction and filling of data missing segments in the energy management system by constructing a filling model based on the naive Bayes algorithm, thereby improving data integrity and continuity. The system first dynamically manages historical data and new data through a sliding window mechanism to ensure that the model can fully consider the changing trend of time series data during the filling process. The filling model combines the distribution characteristics of historical data and current data during initialization, and assigns weights to data in different time periods, so that the model can adapt to changes in equipment status and environment. After detecting a missing data segment, the system immediately calls the filling model for prediction and filling, and replaces the original missing data with the result and writes it into the database, thereby ensuring the accuracy of subsequent analysis and anomaly detection.

[0037] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0039] In the figure:

[0040] Figure 1 It is a flow chart of data collection and processing;

[0041] Figure 2 Anomaly detection graph for energy flow data;

[0042] Figure 3 This is a hierarchical structure diagram of the energy management system based on C / S architecture.

[0043] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0044] The present invention will now be described in further detail with reference to the accompanying drawings.

[0045] See also Figure 1-3 As shown, in this embodiment, a method for filling data in an energy management system based on time series analysis is provided, comprising the following steps:

[0046] Step (1): Obtain operating status data and energy consumption data from sensors and devices in the energy management system, store the data in the system database, and mark the time periods when data collection is interrupted or missing as targets for subsequent filling;

[0047] Step (2): Clean and format the collected time series data, then use the sliding window mechanism to establish a dynamic data warehouse to manage historical data and new data in real time;

[0048] Step (3): Construct a filling model based on the Naive Bayes algorithm and combine historical data with current data distribution when initializing the model, assign weights to data in different time periods, and adapt to data change trends;

[0049] Step (4): When a data missing segment is detected, the filling model is called to predict and fill the marked missing segment, and the filling result is written into the database to replace the missing data.

[0050] The present invention realizes intelligent prediction and filling of data missing segments in the energy management system by constructing a filling model based on the naive Bayes algorithm, thereby improving data integrity and continuity. The system first dynamically manages historical data and new data through a sliding window mechanism to ensure that the model can fully consider the changing trend of time series data during the filling process. The filling model combines the distribution characteristics of historical data and current data during initialization, and assigns weights to data in different time periods, so that the model can adapt to changes in equipment status and environment. After detecting a missing data segment, the system immediately calls the filling model for prediction and filling, and replaces the original missing data with the result and writes it into the database, thereby ensuring the accuracy of subsequent analysis and anomaly detection.

[0051] In this embodiment, before detecting the data missing segment, an isolation forest algorithm model is also established and initialized based on historical data to provide a benchmark model for detecting anomalies in energy flow data, wherein the steps of establishing and initializing the isolation forest algorithm model are:

[0052] Extract historical data from the energy management database and perform denoising, normalization and missing value processing on the historical data to generate a clean data set as the training input of the model. Then, set the number of subsamples n of the isolation forest algorithm model. s , number of isolated trees t, anomaly score threshold θ;

[0053] Subsamples are randomly drawn from the historical data set to construct multiple isolated trees. Each tree recursively randomly selects features and partition points until all data points are completely isolated. Then, the average path length of all isolated trees is used through the formula Calculate the anomaly score s(x) for each data point, and then compare the calculated anomaly score with the threshold θ as the benchmark for the model to determine whether the data point is abnormal, where s(x) is the anomaly score of data point x, E(H) is the average maximum depth of the tree, and is related to the number of samples n. s Related, h(x) is the path length of data point x in the isolation tree;

[0054] After the model initialization is completed, the model is deployed to the application service layer and is ready to receive future energy flow data input.

[0055] In this embodiment, when detecting data, the detection cycle is entered at a preset time interval, the current energy flow data is read, the latest energy flow data is obtained from the energy database within the current detection cycle, and is prepared to be input into the anomaly detection module for analysis. The MHSIF model is used to perform anomaly detection on the energy flow data within the current cycle to identify whether there are short-term sequence anomalies, and the detected anomalies are marked to determine whether the current detection cycle has ended; if not, the data continues to be read and anomaly detection is performed until the cycle ends. At the end of the detection cycle, the system determines whether the anomaly rate within the cycle exceeds a preset threshold to determine whether the detection model needs to be updated. If the anomaly rate exceeds the threshold, the model is updated using the data of the current detection cycle to adapt the model to the new system state and data changes.

[0056] In this embodiment, the steps of establishing a dynamic data warehouse using a sliding window mechanism are as follows:

[0057] Step 201: Set the time window size W t and sampling frequency f s , then, according to the set time window size W t and sampling frequency f s Initialize the sliding window mechanism, window size W t Indicates the length of time that can be retained in the sliding window, that is, W t =n·T s , where n is the number of sampling points in the window, T s is the time span of a single data point;

[0058] Step 202: In the sliding window, whenever new data D is received t Then, the data is stored in the dynamic data warehouse in chronological order. t When entering, the sliding window mechanism automatically checks whether there is data outside the window range W t , data outside the window Will be removed according to the first-in-first-out rule to ensure that the warehouse only retains the most recent W t Data within a time period;

[0059] Step 203: Each time after removing old data and loading new data, the data in the window is reordered by time. As time goes by, the window keeps rolling. Whenever new data enters, the start and end points of the window move at the same time. The update formula for window rolling is: W t (t+1)=W t (t)+D t+1 -D t-n, which means that each time the window is updated, the new data D t+1 Add the earliest data D t-n Finally, the data managed by the sliding window is stored in the dynamic data warehouse to complete the establishment of the dynamic data warehouse.

[0060] In this embodiment, the steps of constructing a filling model based on the Naive Bayes algorithm and combining historical data with current data distribution when initializing the model to assign weights to data in different time periods are as follows:

[0061] Step 301: Historical data D H ={d 1 , d 2 , dots, d m} and current time period data D C ={d m+1 , dots, d n}, and then, using the naive Bayes principle, calculate the prior probability P(T H ) and P(T C ) to quantify the importance of data in each time period, which is expressed as: In the formula, |D H | and |D C |Indicates the number of samples of historical data and current data;

[0062] Step 302: According to the data distribution in different time periods, weights w are assigned to the historical data and the current data respectively. H and w C , its expression is: H =P(T H )·C H , w C =P(T C )·C C Among them, C H and C C It is based on the prior probability of the time period and the credibility coefficient of the data;

[0063] Step 303: Calculate the conditional probability based on the weighted data and construct a naive Bayes imputation model, which is expressed as: Among them, P(x|D) is the probability of filling the missing value x after integrating historical and current data.

[0064] In this embodiment, when a data missing segment is detected, the filling model is called to predict and fill the marked missing segment, and the filling result is written into the database to replace the missing data. The steps are:

[0065] Scan the time series data to find the missing segments in the data. According to the marked missing segments, call the naive Bayes filling model to calculate the predicted value of each missing point, calculate the probability of all values ​​for each missing point, and select the value corresponding to the probability as the filling result. Then, replace the corresponding missing values ​​with the filling results of all missing points one by one, and update the filled data to the database. The naive Bayes algorithm is used to intelligently fill the missing segments of the data to ensure the integrity and continuity of the energy management system data. First, the historical data and current data are divided, and the prior probabilities are calculated separately, and weights are assigned to the data in different time periods to enhance the model's adaptability to data changes. The filling model is constructed based on the weighted data, effectively combining the long-term and short-term data characteristics to ensure that the filling results are highly consistent with the actual data distribution. When a missing segment is detected, the system calls the filling model for prediction and filling, and replaces the missing values ​​one by one. Finally, the filled data is updated to the database to provide reliable input for subsequent analysis and anomaly detection.

[0066] The architecture of the energy management system data filling method based on time series analysis is characterized by comprising:

[0067] Data access layer, the data access layer consists of the collection server database and the application server database. The collection server database is used to store the original energy data collected by the sensor, and the application server database is used to store the complete data after preprocessing and filling;

[0068] Data preprocessing module, which obtains energy data from the acquisition server database, cleans the data, converts the format and aligns the time to ensure the quality and consistency of the data, and stores the preprocessed data in the application server database;

[0069] Missing data filling module, when missing data is detected, the missing data filling module calls the filling model based on the naive Bayes algorithm to fill the missing segment and update the filled data to the application server database;

[0070] Anomaly detection and early warning module, which is used to read complete data from the application server database, detect abnormal states in energy flow data based on the isolation forest algorithm, trigger an alarm when an anomaly is detected, and record abnormal data segments;

[0071] Data display module,The data display module is used to visualize the filled data and anomaly detection results through the presentation layer, and to display the abnormal data segments and filling status to managers.

[0072] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. A data filling method for energy management system based on time series analysis, characterized in that: The following steps are involved: Step (1): Obtain operating status data and energy consumption data from sensors and devices in the energy management system, store the data in the system database, and mark the time periods when data collection is interrupted or missing as targets for subsequent filling; Step (2): Clean and format the collected time series data, then use the sliding window mechanism to establish a dynamic data warehouse to manage historical data and new data in real time; Step (3): Construct a filling model based on the Naive Bayes algorithm and combine historical data with current data distribution when initializing the model, assign weights to data in different time periods, and adapt to data change trends; Step (4): When a data missing segment is detected, the filling model is called to predict and fill the marked missing segment, and the filling result is written into the database to replace the missing data.

2. The method for filling data of an energy management system based on time series analysis according to claim 1 is characterized in that: Before detecting the data missing segment, an isolation forest algorithm model is also established and initialized based on historical data to provide a benchmark model for detecting anomalies in energy flow data. The steps for establishing and initializing the isolation forest algorithm model are: Extract historical data from the energy management database and perform denoising, normalization and missing value processing on the historical data to generate a clean data set as the training input of the model. Then, set the number of subsamples n of the isolation forest algorithm model. s , number of isolated trees t, anomaly score threshold θ; Subsamples are randomly drawn from the historical data set to construct multiple isolated trees. Each tree recursively randomly selects features and partition points until all data points are completely isolated. Then, the average path length of all isolated trees is used through the formula Calculate the anomaly score s(x) for each data point, and then compare the calculated anomaly score with the threshold θ as the benchmark for the model to determine whether the data point is abnormal, where s(x) is the anomaly score of data point x, E(H) is the average maximum depth of the tree, and is related to the number of samples n. s Related, h(x) is the path length of data point x in the isolation tree; After the model initialization is completed, the model is deployed to the application service layer and is ready to receive future energy flow data input.

3. The method for filling data of an energy management system based on time series analysis according to claim 2 is characterized in that: When detecting data, enter the detection cycle at the preset time interval, read the current energy flow data, obtain the latest energy flow data from the energy database within the current detection cycle, prepare to input the anomaly detection module for analysis, use the MHSIF model to perform anomaly detection on the energy flow data within the current cycle, identify whether there are short-term sequence anomalies, and mark the detected anomalies to determine whether the current detection cycle has ended; if not, continue to read data and perform anomaly detection until the end of the cycle. At the end of the detection cycle, the system determines whether the anomaly rate within the cycle exceeds the preset threshold to determine whether the detection model needs to be updated. If the anomaly rate exceeds the threshold, use the data of the current detection cycle to update the model so that the model adapts to the new system status and data changes.

4. The method for filling data of an energy management system based on time series analysis according to claim 3 is characterized in that: The steps to establish a dynamic data warehouse using the sliding window mechanism are: Step 201: Set the time window size W t and sampling frequency f s , then, according to the set time window size W t and sampling frequency f s Initialize the sliding window mechanism, window size W t Indicates the length of time that can be retained in the sliding window, that is, W t =n·T s , where n is the number of sampling points in the window, T s is the time span of a single data point; Step 202: In the sliding window, whenever new data D is received t Then, the data is stored in the dynamic data warehouse in chronological order. t When entering, the sliding window mechanism automatically checks whether there is data outside the window range W t , data outside the window Will be removed according to the first-in-first-out rule to ensure that the warehouse only retains the most recent W t Data within a time period; Step 203: Each time after removing old data and loading new data, the data in the window is reordered by time. As time goes by, the window keeps rolling. Whenever new data enters, the start and end points of the window move at the same time. The update formula for window rolling is: W t (t+1)=W t (t)+D t+1 -D t-n , which means that each time the window is updated, the new data D t+1 Add the earliest data D t-n Finally, the data managed by the sliding window is stored in the dynamic data warehouse, completing the establishment of the dynamic data warehouse.

5. The method for filling data of an energy management system based on time series analysis according to claim 1 is characterized in that: The steps of constructing a filling model based on the Naive Bayes algorithm and combining historical data with current data distribution when initializing the model and assigning weights to data in different time periods are as follows: Step 301: Historical data D H ={d1, d2, dots, d m } and current time period data D C ={d m+1 , dots, d n }, and then, using the naive Bayes principle, calculate the prior probability P(T H ) and T(T C ) to quantify the importance of data in each time period, which is expressed as: In the formula, |D H | and |D C |Indicates the number of samples of historical data and current data; Step 302: According to the data distribution in different time periods, weights w are assigned to the historical data and the current data respectively. H and w C , its expression is: H =P(T H )·C H , w C =P(T C )·C C , where C H and C C It is based on the prior probability of the time period and the credibility coefficient of the data; Step 303: Calculate the conditional probability based on the weighted data and construct a naive Bayes imputation model, which is expressed as: Among them, P(x|D): after integrating historical and current data, missing values x The probability of filling.

6. The method for filling data of an energy management system based on time series analysis according to claim 1 is characterized in that: When a data missing segment is detected, the filling model is called to predict and fill the marked missing segment, and the filling result is written into the database to replace the missing data. The steps are: Scan the time series data to find the missing segments in the data. According to the marked missing segments, call the naive Bayes filling model to calculate the predicted value of each missing point. For each missing point, calculate the probability of all values ​​and select the value corresponding to the probability as the filling result. Then, replace the corresponding missing values ​​with the filling results of all missing points one by one, and update the filled data to the database.

7. An architecture of a method for filling data in an energy management system based on time series analysis according to any one of claims 1 to 6, characterized in that: include: Data access layer, the data access layer consists of the collection server database and the application server database. The collection server database is used to store the original energy data collected by the sensor, and the application server database is used to store the complete data after preprocessing and filling; Data preprocessing module, which obtains energy data from the acquisition server database, cleans the data, converts the format and aligns the time to ensure the quality and consistency of the data, and stores the preprocessed data in the application server database; Missing data filling module, when missing data is detected, the missing data filling module calls the filling model based on the naive Bayes algorithm to fill the missing segment and update the filled data to the application server database; Anomaly detection and early warning module,The anomaly detection and early warning module is used to read complete data from the application server database, detect abnormal states in energy flow data based on the isolation forest algorithm, and trigger an alarm when an anomaly is detected, while recording the abnormal data segment; Data display module,The data display module is used to visualize the filled data and anomaly detection results through the presentation layer, and to display the abnormal data segments and filling status to managers.

Images