Abnormality detection method based on terminal power utilization intelligent equipment

Abstract

The invention discloses an anomaly detection method based on terminal power utilization intelligent equipment. The anomaly detection method comprises the following steps: S1, drawing up a comprehensive real-time and regular detection power utilization safety monitoring method; s2, drawing up an abnormal power consumption time detection algorithm; and S3, designing a comprehensive electric power safety monitoring and energy-saving optimization system. According to the method, the power utilization characteristics of each node are considered, real-time data anomaly detection and long-term power system health diagnosis are combined, and comprehensive and hierarchical analysis and processing are performed on sudden problems and hidden safety hazards of the power system; according to the method, the STL decomposition and Mann-Kendall trend test are combined to judge whether the node current has an abnormal growth phenomenon, and the method has good robustness and the capability of judging the abnormal growth trend of the current, and provides a new thought for the safety detection of a power system of a terminal user.

Description

Technical Field

[0001] The present invention relates to the technical field of power safety monitoring, and in particular to an abnormality detection method based on terminal power consumption intelligent equipment. Background Art

[0002] In today's digital and information age, electricity is increasingly used in production and life, such as lighting, using electricity as a power to start machines and equipment, converting electrical energy into thermal energy for smelting, welding, cutting, drying, heating, metal heat treatment, etc., converting electrical energy into chemical energy for electroplating, electrolysis, electrochemical processing, etc. Electricity is also widely used in medical, communication, computing, measurement and other fields. The ubiquitous application of electricity has promoted the development of science and technology, especially the rapid development of new energy technology, the Internet of Things, big data, cloud computing, and artificial intelligence in today's era, which has further promoted the progress of science and technology. However, if it is improperly applied or not effectively supervised, electricity will not only hurt people, but also bring other disasters such as electric shock, fire and explosion. Therefore, when we use electricity, we must focus on electrical safety issues.

[0003] In the existing technology, traditional power monitoring systems may lack intelligence and real-time performance, and are unable to detect power safety issues and abnormal power consumption problems in a timely manner. On the other hand, most of the current safety monitoring and energy-saving optimization systems only collect data on the main line and are unable to explore the power consumption characteristics of each node. They have poor ability to locate faults or safety hazards, poor pertinence for personalized energy-saving suggestions, and are difficult to obtain data and complex to process. Summary of the invention

[0004] The purpose of the present invention is to provide an anomaly detection method based on terminal power-consuming intelligent devices to solve the problems proposed in the above-mentioned background technology that the traditional power monitoring system may lack intelligence and real-time performance, and cannot detect power safety problems and abnormal power consumption problems in time. On the other hand, most of the current safety monitoring and energy-saving optimization systems only collect data on the main line and cannot explore the power consumption characteristics of each node. The ability to locate faults or safety hazards is poor, and the targetedness of personalized energy-saving suggestions is poor. Data acquisition is difficult and complex to process.

[0005] To achieve the above object, the present invention provides the following technical solution: an abnormality detection method based on terminal power consumption intelligent device, comprising the following steps:

[0006] S1. Formulate a comprehensive real-time and periodic power safety monitoring method, analyze the shortcomings of the current end-user power system safety monitoring without segmented monitoring, including real-time power anomaly detection method and power system health detection method, and conduct comprehensive and hierarchical monitoring and investigation of sudden problems and hidden safety hazards in the power system of end-users;

[0007] S2. Formulate abnormal power consumption time detection algorithm to solve the power saving problem of end users, improve the power consumption efficiency of end users and control the power consumption cost, judge the power consumption equipment reminder at abnormal time according to the power consumption rhythm of nodes, explore the potential energy saving possibility, and provide complete and timely effective help for users to save energy;

[0008] S3. Design a comprehensive power safety monitoring and energy-saving optimization system to provide scientific data support for users to understand the system status in real time, formulate power consumption plans, plan power costs and locate faults and repairs, and provide new ideas for real-time safety monitoring of power systems, control power costs and fault location methods.

[0009] Preferably, in step S1, the specific steps of the real-time power consumption anomaly detection method are as follows:

[0010] A1. Group nodes by region and set threshold alarm ranges;

[0011] A2. The real-time power consumption abnormality alarm module collects power connection information, conducts collaborative horizontal comparison and analysis of nodes in the region, and then clusters the node voltage to obtain the normal value range of the regional voltage. If the node voltage is abnormal, the current surges, and the temperature rises, the node overload warning will be issued;

[0012] Compare and analyze the voltages in each area to determine whether there are any regional voltage anomalies;

[0013] When the node data is within the threshold alarm range, the abnormal data information alarm will be pushed.

[0014] Preferably, in step S1, the power system health detection method is: low equipment energy efficiency or line aging problems are obtained through a time series analysis method of longitudinal data, and unbalanced node load problems in the region are analyzed through horizontal comparison between nodes.

[0015] Preferably, the method for analyzing the longitudinal data time series of each node includes the following steps:

[0016] B1. Data preprocessing: Preprocess the raw data, including filling in missing values ​​to ensure data quality.

[0017] B2. Draw a time series graph: Draw the current data into a time series graph to visually observe the changing trend of the current.

[0018] B3. Seasonal analysis: If there is seasonal variation, use seasonal decomposition method to decompose the data and obtain seasonal intensity and residual components.

[0019] B4. Statistical test: If the seasonal intensity is high, the Mann-Kendall trend test in the statistical test method is applied to the remaining components to test whether the data has a significant daily growth trend. If the seasonal intensity is low, the Mann-Kendall trend test is directly applied to the original sequence data to analyze whether the current has a significant daily growth trend.

[0020] B5. Result analysis: Comprehensively analyze the results of the above steps to determine whether there is a daily increasing trend in the current and evaluate the strength of the trend.

[0021] Preferably, in step B3, the seasonal decomposition method is applied in the following specific steps: decomposing the time series data into seasonality, trend and residual by using a local weighted scatter plot smoothing technique, wherein the trend component represents the changing trend of the data; the seasonal component represents the periodic fluctuation of the data on different time scales; and the residual component represents the random noise that cannot be explained by the trend and seasonal components;

[0022] y t =T t +S t +R t ;

[0023] In the formula, y t is the current value, T t , S t , R t They are trend component, seasonal component and residual component respectively. STL is robust to outliers, so occasional abnormal observations will not affect the estimation of trend cycle and seasonal components.

[0024] After decomposition, the seasonal intensity of the current value can be calculated:

[0025]

[0026] In the formula, F s is the seasonal intensity, S t , R t They are the seasonal component and the residual component after decomposition. The seasonal intensity is between 0 and 1. The closer it is to 1, the stronger the periodicity of the data series. Similarly, the calculation formula for trend intensity is:

[0027]

[0028] In the formula, F T is the trend strength, T t , R t are the trend component and the residual component after decomposition.

[0029] STL decomposition can better determine the power consumption characteristics of nodes, seasonal intensity can be used to determine the power consumption pattern of nodes; trend intensity can be used to analyze whether the current of nodes has an abnormal growth trend. Under normal circumstances, the amplitude of node data with strong seasonality increases and decreases steadily with the change of seasons in the long term; the trend component maintains a continuous development trend on a long-term scale and remains unchanged in the short term. The remaining components show irregular scattered distribution in the short term.

[0030] After excluding seasonal influence, the Mann-Kendall trend test is used to judge the trend of the sequence data.

[0031] For a time series X with a sample size of n, the Mann-Kendall statistic first calculates S:

[0032]

[0033] Calculate the variance:

[0034]

[0035] Where: p represents the number of tie groups of data X, q k Refers to the number of data in the kth tie group. In the Mann-Kendall trend test, a tie group refers to a group of data points with the same observation value in the sample. In time series data, sometimes there are multiple observations with the same value, and these same values ​​form a tie group.

[0036] Calculate the standard normal distribution statistic Z:

[0037]

[0038] Propose hypothesis: H0: There is no monotonic trend; contrast alternative hypothesis: Ha: There is a monotonic increasing trend.

[0039] Preferably, before conducting a hypothesis test, a significance level is set, represented by α. The significance level represents the probability of accepting an error when rejecting the null hypothesis. MK ≥Z 1-α , we reject the null hypothesis H0 and accept the alternative hypothesis Ha, where is the percentile of the standard normal distribution;

[0040] If the daily growth trend of the current trend data meets the characteristics of line aging or low-efficiency equipment nodes, the user will be reminded of the risk, and the calculated daily average current will be compared horizontally to determine whether there is an unbalanced node load in the area. The load distribution diagram will be drawn so that the user can intuitively obtain the load distribution situation in the area.

[0041] Preferably, in step S2, the abnormal power consumption time detection algorithm is specifically as follows: the historical load curves of the nodes are clustered by DBSCAN to obtain the clustering of the node power load characteristic curves, and the decision boundary of the power consumption rhythm is obtained by using the One-Class SVM algorithm, and the real-time power consumption data of the node is within the decision boundary as the basis for abnormal judgment.

[0042] Preferably, the training data of One-Class SVM contains only positive samples. The goal is to build a model to describe the normal situation so as to detect anomalies in the test phase. The goal of One-Class SVM is to find a hyperplane in the feature space that can enclose the normal data, which is expressed by the following convex optimization problem:

[0043]

[0044] The following constraints are met:

[0045] Where w is the hyperplane normal vector, is the feature map of the input sample, ρ is the distance to the hyperplane, ξ i is the slack variable, v is the specified hyperparameter, and n is the number of samples;

[0046] Use Lagrangian dual solution: define the Lagrangian function and introduce the constraints into the objective function:

[0047]

[0048] Among them, α and β are Lagrange multipliers. SMO is then used to solve the dual problem and calculate the optimal parameters as shown in formula (2.3) (2.4), where s is the support vector set;

[0049]

[0050] Preferably, in step S3, the power safety monitoring and energy-saving optimization system includes a server side and a browser side, the server side is responsible for processing the business logic of the application, including data processing, calculation, storage and interaction with the database, the server side includes a database, an energy-saving optimization module, a power detection module and a remote control module, and the database adopts a time series database;

[0051] The browser is responsible for rendering the user interface and converting the data provided by the server into a page visible to the user, including a real-time display module of the power system topology, a multi-dimensional data visualization and report generation module, an abnormal data alarm and push module, an energy-saving optimization reminder module, and remote control;

[0052] The server and the browser use the HTTP protocol for communication, and the server and the gateway use the UDP protocol for communication.

[0053] Preferably, energy-saving optimization reminders include power usage time analysis and power consumption analysis. Real-time power usage analysis uses an abnormal power usage time detection model to detect abnormal power usage time and excessive power usage time. Power consumption analysis mines potential energy savings to discover equipment with low energy efficiency. For power usage time analysis of high-energy-consuming nodes, it helps users achieve energy saving or reduce electricity costs.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] In the present invention, considering the power consumption characteristics of each node, real-time data anomaly detection is combined with long-term power system health diagnosis, and a comprehensive and hierarchical analysis and processing are made for sudden problems and hidden safety hazards of the power system. A method combining STL decomposition with Mann-Kendall trend test is used to determine whether there is an abnormal growth phenomenon in the node current. The method has good robustness and the ability to distinguish the abnormal growth trend of current, and provides a new idea for the power system safety detection of terminal users. The clustering method is combined with One-Class SVM to mine the node power consumption rhythm, and the power consumption is accurately analyzed and abnormal judgment is made from the time dimension. The abnormal power consumption rhythm can be timely reminded to the user, and the potential energy saving possibility in the system is discovered, and the user is provided with more comprehensive and in-place energy-saving suggestions and more timely energy-saving optimization reminders. The data collection and analysis and mining of terminal power equipment provide users with a comprehensive, detailed and intelligent power monitoring system, which provides scientific data support for users to understand the system status, formulate power consumption plans, plan power consumption costs and fault location and maintenance in real time, and provide new methods for real-time safety monitoring, control power consumption costs and fault location methods of power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a real-time power data alarm flow chart in an abnormality detection method based on terminal power consumption intelligent equipment of the present invention;

[0057] Figure 2 This is a flow chart of regular power system monitoring and diagnosis in an abnormality detection method based on terminal power consumption intelligent equipment of the present invention.

[0058] Figure 3 This is a flowchart of time series analysis of longitudinal data of each node in an abnormality detection method based on terminal power consumption intelligent equipment of the present invention;

[0059] Figure 4 This is a diagram of the power consumption analysis system in the abnormality detection method based on the terminal power consumption intelligent device of the present invention. DETAILED DESCRIPTION

[0060] 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 implementation regulations described 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.

[0061] Example: Refer to Figure 1-4 As shown: A method for detecting abnormalities based on terminal power-using intelligent devices, comprising the following steps:

[0062] Step 1: Formulate a comprehensive real-time and periodic power safety monitoring method, analyze the shortcomings of the current end-user power system safety monitoring without segmented monitoring, including real-time power anomaly detection methods and power system health detection methods, and conduct comprehensive and hierarchical monitoring and investigation of sudden problems and hidden safety hazards in the end-user power system;

[0063] The specific steps of the real-time power consumption anomaly detection method are as follows:

[0064] 1. Group nodes by region and set threshold alarm ranges;

[0065] 2. The real-time power consumption abnormality alarm module collects power connection information, conducts collaborative horizontal comparison and analysis of each node in the area, and then clusters the node voltage to obtain the normal value range of the regional voltage. If the node voltage is abnormal, the current surges, and the temperature rises, the node overload warning will be issued;

[0066] It adopts a combination of threshold alarm and data mining technology, which is specifically used to quickly detect significant mutations in voltage, current and other data. By conducting collaborative horizontal analysis of the voltage of each node in the region and the voltage of each region, it can more accurately identify abnormal situations and provide hierarchical and comprehensive abnormality detection. This method can not only effectively monitor the abnormalities of small-scale data, but also determine whether the node is overloaded. Considering that the devices monitored by devices such as smart sockets are highly volatile, and the loads of nodes in the region are different, the analysis of voltage is particularly critical in horizontal comparison. This comprehensive analysis method improves the accuracy and reliability of the system and effectively copes with the complexity of real-time power consumption data monitoring.

[0067] Compare and analyze the voltages in each area to determine whether there are any regional voltage anomalies;

[0068] When the node data is within the threshold alarm range, the abnormal data information alarm will be pushed;

[0069] The power system health detection method is: the low energy efficiency of equipment or line aging problems are obtained through the time series analysis method of longitudinal data, and the unbalanced load of nodes in the region is analyzed through horizontal comparison between nodes. The time series analysis method of longitudinal data of each node includes the following steps:

[0070] ①. Data preprocessing: Preprocess the raw data, including filling in missing values ​​to ensure data quality.

[0071] ②. Draw a time series graph: Draw the current data into a time series graph to visually observe the changing trend of the current.

[0072] ③ Seasonal analysis: If there is seasonal variation, use seasonal decomposition method to decompose the data and obtain seasonal intensity and residual components;

[0073] The specific steps of seasonal decomposition method are as follows: the time series data is decomposed into seasonality, trend and residual by local weighted scatter plot smoothing technology. The trend component represents the changing trend of the data; the seasonal component represents the periodic fluctuation of the data on different time scales; the residual component represents the random noise that cannot be explained by the trend and seasonal components.

[0074] y t =T t +S t +R t ;

[0075] In the formula, y t is the current value, T t , S t , R t They are trend component, seasonal component and residual component respectively. STL is robust to outliers, so occasional abnormal observations will not affect the estimation of trend cycle and seasonal components.

[0076] After decomposition, the seasonal intensity of the current value can be calculated:

[0077]

[0078] In the formula, F s is the seasonal intensity, S t , R t They are the seasonal component and the residual component after decomposition. The seasonal intensity is between 0 and 1. The closer it is to 1, the stronger the periodicity of the data series. Similarly, the calculation formula for trend intensity is:

[0079]

[0080] In the formula, F T is the trend strength, T t , R tare the trend component and the residual component after decomposition.

[0081] STL decomposition can better determine the power consumption characteristics of nodes, seasonal intensity can be used to determine the power consumption pattern of nodes; trend intensity can be used to analyze whether the current of nodes has an abnormal growth trend. Under normal circumstances, the amplitude of node data with strong seasonality increases and decreases steadily with the change of seasons in the long term; the trend component maintains a continuous development trend on a long-term scale and remains unchanged in the short term. The remaining components show irregular scattered distribution in the short term.

[0082] After excluding seasonal influence, the Mann-Kendall trend test is used to judge the trend of the sequence data.

[0083] For a time series X with a sample size of n, the Mann-Kendall statistic first calculates S:

[0084]

[0085]

[0086] Calculate the variance:

[0087]

[0088] Where: p represents the number of tie groups of data X, q k Refers to the number of data in the kth tie group. In the Mann-Kendall trend test, a tie group refers to a group of data points with the same observation value in the sample. In time series data, sometimes there are multiple observations with the same value, and these same values ​​form a tie group.

[0089] Calculate the standard normal distribution statistic Z:

[0090]

[0091] Proposed hypothesis: H0: There is no monotonic trend; contrast alternative hypothesis: Ha: There is a monotonic increasing trend;

[0092] Before conducting a hypothesis test, set the significance level, represented by α. The significance level represents the probability of error accepted when rejecting the null hypothesis. MK ≥Z 1-α , we reject the null hypothesis H0 and accept the alternative hypothesis Ha, where is the percentile of the standard normal distribution;

[0093] If the daily growth trend of the current trend data meets the characteristics of line aging or low-efficiency equipment nodes, the user will be reminded of the risk, and the calculated daily average current will be compared horizontally to determine whether there is an unbalanced node load in the area. The load distribution diagram will be drawn so that the user can intuitively obtain the load distribution situation in the area.

[0094] ④. Statistical test: If the seasonal intensity is high, the Mann-Kendall trend test in the statistical test method is applied to the remaining components to test whether the data has a significant daily growth trend. If the seasonal intensity is low, the Mann-Kendall trend test is directly applied to the original sequence data to analyze whether the current has a significant daily growth trend.

[0095] ⑤. Result analysis: Comprehensively analyze the results of the above steps to determine whether there is a daily increasing trend in the current and evaluate the strength of the trend.

[0096] Step 2: Develop an abnormal power consumption time detection algorithm: cluster the historical load curves of the nodes through DBSCAN to obtain the clustering of the node power load characteristic curves, use the One-Class SVM algorithm to obtain the decision boundary of the power consumption rhythm, and use the real-time power consumption data of the nodes within the decision boundary as the basis for abnormal judgment;

[0097] The training data of One-Class SVM contains only positive samples. The goal is to build a model to describe the normal situation so as to detect anomalies in the test phase. The goal of One-Class SVM is to find a hyperplane in the feature space that can enclose the normal data, which is expressed by the following convex optimization problem:

[0098]

[0099] The following constraints are met:

[0100] Where w is the hyperplane normal vector, is the feature map of the input sample, ρ is the distance to the hyperplane, ξ i is the slack variable, v is the specified hyperparameter, and n is the number of samples;

[0101] Use Lagrangian dual solution: define the Lagrangian function and introduce the constraints into the objective function:

[0102]

[0103] Among them, α and β are Lagrange multipliers. SMO is then used to solve the dual problem and calculate the optimal parameters as shown in formula (2.3) (2.4), where s is the support vector set;

[0104]

[0105] Solve the problem of power saving for end users, improve the power efficiency and control the power cost of end users, judge the power equipment reminder at abnormal time according to the power rhythm of nodes, explore the potential energy saving possibility, and provide complete and timely effective help for users to save energy. Solve the problem of detecting power consumption at abnormal time of nodes, fully consider the power consumption characteristics of each node, solve the drawbacks of the threshold method, and provide new ideas for energy saving optimization for end users.

[0106] Step 3: Design a comprehensive power safety monitoring and energy-saving optimization system to provide scientific data support for users to understand the system status in real time, formulate power consumption plans, plan power costs, and locate and repair faults, and provide new ideas for real-time safety monitoring of power systems, control power costs, and locate faults;

[0107] The power safety monitoring and energy-saving optimization system includes a server side and a browser side. The server side is responsible for processing the business logic of the application, including data processing, calculation, storage and interaction with the database. The server side includes a database, an energy-saving optimization module, a power detection module and a remote control module. The database uses a time series database, which is an open source database system specially designed to process time series data. Time series data is a type of data sorted in chronological order, which is commonly used in monitoring, sensor data, logs, application performance monitoring and other scenarios;

[0108] The browser is responsible for rendering the user interface and converting the data provided by the server into a page visible to the user, including the real-time display module of the power system topology map, the multi-dimensional data visualization and report generation module, the abnormal data alarm and push module, the energy-saving optimization reminder module, and the remote control. The real-time display module of the power system topology map is used to drag the device icon to build the topology map according to the actual installation of the monitoring equipment, and establish the mapping relationship between the chart and the actual equipment through the GUID. This function will display the real-time data of the equipment next to the corresponding icon, and the topology map can be built and viewed by area and level. In the event of a system failure, it assists management personnel to quickly locate the fault location and cause. The multi-dimensional data visualization and report generation module is used to support a variety of chart formats, including pie charts, line charts, bar charts, etc., to display various key data of the power system. Users can use charts to visually observe the changing trends of power load, energy consumption and other information and various data year-on-year and month-on-month, customize the export report content, and provide reliable data support for the power management and planning of enterprises and institutions. The abnormal data alarm and push module is used to use different color labels on the front-end page. Users can define special emergency event types such as (equipment GUID, abnormal conditions and other attributes). When a special emergency occurs, the system actively pushes information to users through pop-ups, text messages or emails. Users can quickly locate abnormal equipment based on the power system topology map, optimize the process of fault location and maintenance, and remote control is used to connect with the safety monitoring and energy-saving optimization modules to achieve remote control and timing control of dispatchable equipment.

[0109] The server and the browser use HTTP protocol for communication, and the server and the gateway use UDP protocol for communication;

[0110] Energy-saving optimization reminders include power usage time analysis and power consumption analysis. Real-time power consumption analysis uses an abnormal power usage time detection model to detect abnormal power usage time and excessive power usage time. Power consumption analysis mines potential energy savings and discovers equipment with low energy efficiency. It analyzes the power usage time of high-energy-consuming nodes to help users achieve energy conservation or reduce electricity costs.

[0111] The present invention combines real-time data anomaly detection with long-term power system health diagnosis from the perspective of power consumption characteristics of each node, makes a comprehensive and hierarchical analysis and processing of sudden problems and hidden safety hazards of the power system, and uses a method combining STL decomposition with Mann-Kendall trend test to determine whether there is an abnormal growth phenomenon in the node current. The method has good robustness and the ability to distinguish the abnormal growth trend of current, and provides a new idea for the safety detection of the power system of the terminal user. The clustering method is combined with One-Class SVM to mine the node power consumption rhythm, and the power consumption is accurately analyzed and abnormal judgment is made from the time dimension. The abnormal power consumption rhythm can be timely reminded to the user, and the potential energy saving possibility in the system can be found, and the user can be provided with more comprehensive and in-place energy-saving suggestions and more timely energy-saving optimization reminders. The data collection and analysis of the terminal power equipment provide users with a comprehensive, detailed and intelligent power monitoring system, which provides scientific data support for users to understand the system status, formulate power consumption plans, plan power consumption costs and locate and repair faults in real time. Provide a new method for real-time safety monitoring, control power consumption costs and fault location methods of the power system.

[0112] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting abnormalities based on terminal power-consuming intelligent devices, characterized in that: The following steps are involved: S1. Formulate a comprehensive real-time and periodic power safety monitoring method, analyze the shortcomings of the current end-user power system safety monitoring without segmented monitoring, including real-time power anomaly detection method and power system health detection method, and conduct comprehensive and hierarchical monitoring and investigation of sudden problems and hidden safety hazards in the power system of end-users; S2. Formulate abnormal power consumption time detection algorithm to solve the power saving problem of end users, improve the power consumption efficiency of end users and control the power consumption cost, judge the power consumption equipment reminder at abnormal time according to the power consumption rhythm of nodes, explore the potential energy saving possibility, and provide complete and timely effective help for users to save energy; S3. Design a comprehensive power safety monitoring and energy-saving optimization system to provide scientific data support for users to understand the system status in real time, formulate power consumption plans, plan power costs and locate faults and repairs, and provide new ideas for real-time safety monitoring of power systems, control power costs and fault location methods.

2. The abnormality detection method based on terminal power consumption intelligent device according to claim 1 is characterized in that: In step S1, the specific steps of the real-time power consumption anomaly detection method are as follows: A1. Group nodes by region and set threshold alarm ranges; A2. The real-time power consumption abnormality alarm module collects power connection information, conducts collaborative horizontal comparison and analysis of nodes in the region, and then clusters the node voltage to obtain the normal value range of the regional voltage. If the node voltage is abnormal, the current surges, and the temperature rises, the node overload warning will be issued; Compare and analyze the voltages in each area to determine whether there are any regional voltage anomalies; When the node data is within the threshold alarm range, the abnormal data information alarm will be pushed.

3. The abnormality detection method based on terminal power consumption intelligent device according to claim 1 is characterized in that: In step S1, the power system health detection method is: the low energy efficiency of equipment or line aging problems are obtained through the time series analysis method of longitudinal data, and the unbalanced node load problems in the region are analyzed through horizontal comparison between nodes.

4. The abnormality detection method based on terminal power consumption intelligent device according to claim 3 is characterized in that: The longitudinal data time series analysis method of each node includes the following steps: B1. Data preprocessing: Preprocess the raw data, including filling in missing values ​​to ensure data quality. B2. Draw a time series graph: Draw the current data into a time series graph to visually observe the changing trend of the current. B3. Seasonal analysis: If there is seasonal variation, use seasonal decomposition method to decompose the data and obtain seasonal intensity and residual components. B4. Statistical test: If the seasonal intensity is high, the Mann-Kendall trend test in the statistical test method is applied to the remaining components to test whether the data has a significant daily growth trend. If the seasonal intensity is low, the Mann-Kendall trend test is directly applied to the original sequence data to analyze whether the current has a significant daily growth trend. B5. Result analysis: Comprehensively analyze the results of the above steps to determine whether there is a daily increasing trend in the current and evaluate the strength of the trend.

5. The abnormality detection method based on terminal power consumption intelligent device according to claim 4 is characterized in that: In step B3, the specific steps of applying the seasonal decomposition method are as follows: decomposing the time series data into seasonality, trend and residual components by means of local weighted scatter plot smoothing technology, wherein the trend component represents the changing trend of the data; the seasonal component represents the periodic fluctuation of the data on different time scales; and the residual component represents the random noise that cannot be explained by the trend and seasonal components; y t =T t +S t +R t ; In the formula, y t is the current value, T t , S t , R t They are trend component, seasonal component and residual component respectively. STL is robust to outliers, so occasional abnormal observations will not affect the estimation of trend cycle and seasonal components. After decomposition, the seasonal intensity of the current value can be calculated: In the formula, F s is the seasonal intensity, S t , R t They are the seasonal component and the residual component after decomposition. The seasonal intensity is between 0 and 1. The closer it is to 1, the stronger the periodicity of the data series. Similarly, the calculation formula for trend intensity is: In the formula, F T is the trend strength, T t , R t are the trend component and the residual component after decomposition. The power consumption characteristics of nodes can be better determined by STL decomposition, and the power consumption pattern of nodes can be determined by seasonal intensity; The trend strength can be used to analyze whether the current of the node has an abnormal growth trend. Under normal circumstances, the amplitude of the node data with strong seasonality increases and decreases steadily with the change of seasons in the long term; The trend component maintains a continuous development trend in the long term and remains unchanged in the short term. The residual component shows an irregular scattered distribution in the short term. After excluding seasonal influence, the Mann-Kendall trend test is used to judge the trend of the sequence data. For a time series X with a sample size of n, the Mann-Kendall statistic first calculates S: Calculate the variance: Where: p represents the number of tie groups of data X, q k Refers to the number of data in the kth tie group. In the Mann-Kendall trend test, a tie group refers to a group of data points with the same observation value in the sample. In time series data, sometimes there are multiple observations with the same value, and these same values ​​form a tie group. Calculate the standard normal distribution statistic Z: Propose hypothesis: H0: There is no monotonic trend; contrast alternative hypothesis: Ha: There is a monotonic increasing trend.

6. The abnormality detection method based on terminal power consumption intelligent device according to claim 5 is characterized in that: Before conducting a hypothesis test, set the significance level, represented by α. The significance level represents the probability of error accepted when rejecting the null hypothesis. MK ≥Z 1-α , we reject the null hypothesis H0 and accept the alternative hypothesis Ha, where is the percentile of the standard normal distribution; If the daily growth trend of the current trend data meets the characteristics of line aging or low-efficiency equipment nodes, the user will be reminded of the risk, and the calculated daily average current will be compared horizontally to determine whether there is an unbalanced node load in the area. The load distribution diagram will be drawn so that the user can intuitively obtain the load distribution situation in the area.

7. The abnormality detection method based on terminal power consumption intelligent device according to claim 1 is characterized in that: In step S2, the abnormal power consumption time detection algorithm is specifically as follows: the historical load curves of the nodes are clustered by DBSCAN to obtain the clustering of the node power load characteristic curves, and the decision boundary of the power consumption rhythm is obtained by using the One-Class SVM algorithm. The real-time power consumption data of the node is within the decision boundary as the basis for abnormal judgment.

8. The abnormality detection method based on terminal power consumption intelligent device according to claim 7 is characterized in that: The training data of One-Class SVM contains only positive samples. The goal is to build a model to describe the normal situation so as to detect anomalies in the test phase. The goal of One-Class SVM is to find a hyperplane in the feature space that can enclose the normal data, which is expressed by the following convex optimization problem: The following constraints are met: ξ i ≥0,i=1,2,…,n; Where w is the hyperplane normal vector, is the feature map of the input sample, ρ is the distance to the hyperplane, ξ i is the slack variable, v is the specified hyperparameter, and n is the number of samples; Use Lagrangian dual solution: define the Lagrangian function and introduce the constraints into the objective function: Among them, α and β are Lagrange multipliers. SMO is then used to solve the dual problem and calculate the optimal parameters as shown in formula (2.3) (2.4), where s is the support vector set; 9. The abnormality detection method based on terminal power consumption intelligent device according to claim 1 is characterized in that: In step S3, the power safety monitoring and energy-saving optimization system includes a server side and a browser side. The server side is responsible for processing the business logic of the application, including data processing, calculation, storage and interaction with the database. The server side includes a database, an energy-saving optimization module, a power detection module and a remote control module. The database adopts a time series database. The browser is responsible for rendering the user interface and converting the data provided by the server into a page visible to the user, including a real-time display module of the power system topology, a multi-dimensional data visualization and report generation module, an abnormal data alarm and push module, an energy-saving optimization reminder module, and remote control; The server and the browser use the HTTP protocol for communication, and the server and the gateway use the UDP protocol for communication.

10. The abnormality detection method based on terminal power consumption intelligent device according to claim 8 is characterized in that: Energy-saving optimization reminders include power usage time analysis and power consumption analysis. Real-time power consumption analysis uses an abnormal power usage time detection model to detect abnormal power usage time and excessive power usage time. Power consumption analysis mines potential energy savings and discovers equipment with low energy efficiency. It analyzes the power usage time of high-energy-consuming nodes to help users achieve energy conservation or reduce electricity costs.

Images