Energy consumption anomaly identification method

By grouping and clustering energy consumption data, a dynamic energy-saving evaluation benchmark is constructed, which solves the problems of high false negative rate and insufficient adaptability of existing energy consumption anomaly identification methods, and realizes accurate identification and refined management of energy consumption anomalies.

CN116628607BActive Publication Date: 2026-11-10ZHUHAI PILOT TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310321571.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-03-29
Publication Date
2026-11-10
Estimated Expiration
2043-03-29

AI Technical Summary

Technical Problem

Existing methods for identifying energy consumption anomalies have a high false negative rate, cannot accurately identify energy consumption in specific rooms, lack adaptability, and cannot provide objective and accurate guidance for energy consumption management.

Method used

By converting the sampled data into standard hourly data, and combining it with air conditioning parameters and outdoor temperature, grouping and clustering analysis is performed to identify the same type of hourly load, construct a dynamic energy-saving evaluation benchmark, identify the boundary threshold between energy saving and non-energy saving, and use KMeans clustering to determine energy consumption anomalies.

Benefits of technology

It improves the accuracy and objectivity of energy consumption anomaly identification, enabling precise identification of abnormal energy consumption periods in each room, reducing operation and maintenance costs, and providing a basis for building energy conservation management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116628607B_ABST
    Figure CN116628607B_ABST
Patent Text Reader

Abstract

This invention discloses a method for identifying energy consumption anomalies. The method includes: converting sampled data into standard hourly data; identifying energy consumption data from the hour prior to startup; processing any anomalies; grouping the hourly load of rooms with similar areas within a building based on air conditioning parameters; searching for similar hourly loads within each group based on outdoor temperature and / or set temperature; filtering similar hourly loads based on room IDs; identifying peaks in the similar hourly loads and their corresponding frequency distributions to determine energy-saving and non-energy-saving peaks; and performing statistical and cluster analysis on the similar hourly loads between the energy-saving and non-energy-saving peaks to construct evaluation indicators and determine the threshold between energy-saving and non-energy-saving peaks. This energy consumption anomaly identification method solves the problems of high false negative rates and insufficient adaptability in existing identification methods.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to an energy consumption management system, and more particularly to a method for identifying energy consumption anomalies in an energy consumption management system. Background Technology

[0002] With the deepening of energy conservation efforts across society, energy consumption management systems have gradually become a standard feature of the information technology construction of various energy-consuming units. To truly maximize the effectiveness of energy consumption management systems, it is naturally necessary to have algorithms capable of identifying abnormal energy consumption within the building energy consumption management system.

[0003] Traditional energy consumption anomaly identification algorithms mainly fall into two categories: benchmarking and outlier identification. The first is the benchmarking method, which assesses energy system energy efficiency by checking whether actual energy consumption or efficiency meets relevant standard requirements. The second is unsupervised learning-based outlier identification, which uses clustering or statistical methods to analyze the data holistically and identify outliers as abnormal energy consumption values. The significant drawbacks of these two methods are twofold: firstly, the relevant standard requirements are set as conservative and fixed values, and the results of directly clustering or statistically analyzing all data often reveal obvious outliers, leading to a high false negative rate. Secondly, the relevant standard requirements are set for large public buildings, and the clustering or statistical analysis is also a holistic analysis, thus failing to understand the specific energy consumption within individual rooms and unable to provide objective, accurate, and clear guidance for energy consumption management. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the existing technology by providing a method for identifying energy consumption anomalies. This invention solves the problems of high false negative rates, lack of objectivity, and insufficient adaptability in existing identification methods.

[0005] The energy consumption anomaly identification method of the present invention includes: The sampled data is converted into standard hourly data, and the energy consumption data of the hour before power-on is identified and abnormal data is identified and processed. The energy consumption data of all rooms in the corresponding building are used to group the hourly load of each room with a similar area based on the corresponding parameters of the air conditioner. Find the same type of hourly load within the group based on outdoor temperature and / or set temperature; Filter hourly loads of the same type based on room ID; determine peaks by analyzing the two variables of hourly load of the same type and the corresponding sub-container statistical frequency, and identify energy-saving peaks and non-energy-saving peaks. Statistical and cluster analysis of the same type of hourly load between the energy-saving peak and the non-energy-saving peak is performed to construct evaluation indicators and find the boundary threshold between energy-saving and non-energy-saving. If the energy consumption is higher than the threshold, the energy consumption is abnormal, and vice versa.

[0006] The sampled data includes rated power, rated cooling capacity, outdoor temperature and / or set temperature, or includes air conditioning parameters and / or operating environment data for all rooms in the corresponding building, including air conditioning nameplate parameters, room electrical parameters, air conditioning operating data and outdoor weather information.

[0007] The relevant parameters of the air conditioner include rated cooling capacity and rated cooling power.

[0008] This invention's energy consumption anomaly identification method combines clustering and statistics, using a dynamic energy-saving evaluation benchmark. It performs correlation analysis and identification based on factors such as hourly load (same type of hourly load data) in a single room, outdoor temperature, indoor temperature, set temperature, rated cooling capacity of the air conditioner, rated power of the air conditioner, on / off status, and room area. It provides dynamic energy consumption thresholds that change with outdoor and set temperatures, improving the accuracy of energy consumption anomaly identification. This significantly enhances the objectivity, authenticity, and accuracy of energy consumption anomaly identification. By finely grouping all rooms in a building and performing correlation analysis of the same type of hourly load, it can identify specific periods of energy consumption anomalies for each room. Furthermore, for energy management projects, it reduces operation and maintenance manpower costs and provides a basis for building energy-saving management decisions, demonstrating strong adaptability and applicability. Attached Figure Description

[0009] Figure 1 This is a flowchart of an embodiment of the present invention. Implementation

[0010] To better understand the present invention, the technical solution of the present invention will now be clearly and completely described through embodiments and accompanying drawings.

[0011] This invention uses hourly data grouped by parameters such as rated power, rated cooling capacity, outdoor temperature, and set temperature. Based on unsupervised methods such as clustering and statistics, it determines the boundary point (boundary threshold) between "energy saving" and "non-energy saving". If the energy consumption is higher than the threshold, the energy consumption is abnormal; otherwise, the energy consumption is normal.

[0012] That is, collecting the air conditioner nameplate parameters, room electrical parameters, air conditioner operation data, and outdoor weather information of all rooms in the corresponding building; the air conditioner nameplate parameters include rated cooling capacity, rated power, etc., the room electrical parameters include hourly load, etc., the air conditioner operation data includes indoor temperature, set temperature, on / off status, etc., and the weather data includes outdoor temperature, etc.

[0013] Convert the frequency converter sampling data into standard hourly data, and identify and mark abnormal values ​​such as energy consumption data, jump data, and / or negative changes in cumulative load value in the hour before startup; The energy consumption data (i.e. hourly load) of all rooms in the corresponding building are used to refine the hourly load of each room with a similar area by using the corresponding parameters of the air conditioner (rated cooling capacity, rated cooling power). Search for the same type of hourly load within the group based on factors including outdoor temperature and / or set temperature; Filter hourly loads of the same type based on room ID; determine peaks by analyzing the two variables of hourly load of the same type and the corresponding sub-container statistical frequency, and identify energy-saving peaks and non-energy-saving peaks. Statistical and cluster analysis is performed on the hourly loads of the same type between the energy-saving peak and the non-energy-saving peak to construct evaluation indicators and identify the boundary threshold between energy-saving and non-energy-saving. If energy consumption exceeds the threshold, energy consumption is considered abnormal; otherwise, energy consumption is considered normal.

[0014] Its air conditioning parameters include the parameters on the air conditioner nameplate, and its air conditioning operating environment data includes the room area, the corresponding room electrical parameters, air conditioning operating data and / or outdoor weather information.

[0015] The specific steps in this example include: a data preprocessing module, a module for searching similar data, and a module for identifying abnormal energy consumption.

[0016] 1. Data Preprocessing Module: This mainly includes data conversion, boot data marking, abnormal data identification and processing, etc. Its data conversion includes converting frequency-sampled data into standard hourly data; The sampled data includes the nameplate parameters of the air conditioners in all rooms of a building, room electrical parameters, air conditioner operation data and / or outdoor weather information, such as outdoor temperature, indoor temperature, set temperature, rated cooling capacity of the air conditioner, rated power of the air conditioner, on / off status, and / or room area.

[0017] Its boot data flags include: The load data for the hour before the air conditioner is turned on is marked. Because the temperature difference between indoors and outdoors is large in the first hour after the air conditioner is turned on, it cannot enter a steady state in a short time. During this period, the outdoor unit of the air conditioner will continue to work, resulting in high energy consumption. In order to reduce false alarms of abnormal energy consumption, the energy consumption data for the hour before the air conditioner is turned on needs to be marked, and abnormal data needs to be identified and processed. The identification of abnormal data includes the identification of outliers in jump data, the identification of randomly inserted erroneous data, and the identification of negative changes in cumulative load values. The specific identification methods mainly include box plot method and 3sigmoid principle. The processing of abnormal data, such as the marking of outlier flags, is not included in the scope of energy consumption anomaly identification.

[0018] 2. Searching for similar data includes: Before searching for similar data, the set temperature in the standard hourly data after data conversion is compared with the reasonable set temperature. This temperature value can be set by the user or professional to the highest possible value while ensuring user comfort. If the set temperature is lower than the reasonable set temperature, an energy consumption anomaly warning will be issued directly; otherwise, an energy consumption anomaly will be identified.

[0019] The search steps include: 1) Finding historical start-up data for all air conditioners in a building with the same or approximately equal rated power and cooling capacity, and room areas that are also the same or approximately equal; 2) Among all historical start-up data that meet the criteria in step 1, finding historical start-up data with a value t+δ that is close to the set temperature and outdoor temperature of the corresponding room at a certain point in time. The δ value of this historical data can be set based on the experience of professionals. The historical start-up data includes historical data that is currently in operation, specifically including start-up status, outdoor temperature, set temperature, indoor temperature, and hourly load.

[0020] For example, on April 25, 2022, at 16:25, the rated power and rated cooling capacity of the air conditioners in building A were 1525 / 5250 respectively. The temperature was set to 24 degrees Celsius at 15:00 that afternoon, and the outdoor temperature was 30 degrees Celsius. A similar hourly load search was performed on room B, which consumes 1.5 kWh of electricity: 1) Find historical start-up data for all air conditioners in building A with rated power and rated cooling capacity of 1525 / 5250 and approximately the same room area; 2) Among all historical data that meet the requirements of step one, find historical data with a temperature setting and outdoor temperature close to t±δ (the δ value can be set based on the experience of professionals).

[0021] 3. Energy consumption anomaly identification includes: Energy consumption anomaly identification is based on the statistical analysis of the same type of hourly load to find the dividing threshold between energy saving and non-energy saving. If the energy consumption is higher than the dividing threshold, the energy consumption is abnormal, and otherwise the energy consumption is normal.

[0022] The statistical analysis of hourly load data of the same type involves first arranging the load data in ascending order, then dividing the data into bins using a sliding window approach, and counting the frequency of data within each bin. Based on the frequency statistics, the distribution of similar load data (i.e., similar hourly load data) can be determined, and peak identification can be performed on the two variables: hourly load and the corresponding frequency. Peak identification can be achieved directly using the existing method `scipy.signal.find_peaks`. If the distribution shows a clear bimodal waveform, the peak with the smaller load value is the energy-saving peak, and the peak with the larger load value is the non-energy-saving peak. The boundary between the two peaks is the dynamic threshold for distinguishing between energy-saving and non-energy-saving data. If the distribution shows a single-peak waveform, meaning that the distribution of energy-saving and non-energy-saving data is unbalanced in the similar hourly load data, then before determining the dynamic threshold, it is necessary to filter the similar hourly load data based on the room ID.

[0023] The steps for filtering room IDs include: 1) Assuming that the amount of energy-saving data is greater than the amount of non-energy-saving data, find the hourly load of the same type, calculate the average of the hourly load data of the same type by grouping by room ID, sort the room IDs from smallest to largest according to the average of the hourly load data of the same type, and reduce the number of rooms in turn to perform statistical analysis on the hourly load of the same type. 2) If the statistical analysis in step 1) can find a bimodal waveform, that is, the dividing line between energy saving and non-energy saving can be found, and the loop will be exited. Otherwise, the process of step 1) will continue. If the number of rooms is less than a certain threshold and the statistical results still do not find a bimodal waveform, then proceed to step 3). 3) Assuming the amount of non-energy-saving data is greater than the amount of energy-saving data, find the hourly load of the same type, calculate the average value of the hourly load of the same type by grouping by room ID, and perform statistical analysis on the hourly load of the same type according to the average value from large to small. 4) If the statistical analysis in step 3) can find a double-peak waveform, that is, the dividing line between energy saving and non-energy saving can be found, exit the loop; otherwise, continue the process in step 1). If the number of rooms is less than a certain threshold, the statistical results still cannot find a double-peak waveform, exit the loop, and the algorithm cannot identify energy consumption anomalies based on the existing data.

[0024] After filtering by room ID, the hourly load of the remaining rooms of the same type is statistically analyzed again. If a bimodal distribution is found, the boundary between "energy saving" and "non-energy saving" is determined.

[0025] There is a transitional range between energy-saving and non-energy-saving modes. To reduce the false alarm rate of energy consumption anomaly identification, the right critical value of the transitional range should be used as the boundary between "energy-saving" and "non-energy-saving" as much as possible. According to offline analysis, the right critical value of the transitional range is characterized by a small frequency and a small absolute value of frequency increment. That is, the right critical value corresponds to the maximum load value in the cluster with a small frequency and a small absolute value of frequency change.

[0026] The frequency and the absolute value of the frequency increment are evaluated.

[0027] To evaluate the small frequency (p) and frequency change Clusters with smaller absolute values, i.e. We have constructed new evaluation indicators as follows:

[0028] The steps to re-analyze the hourly load of the remaining rooms for the same type and continue to determine the boundary between "energy-saving" and "non-energy-saving" include: 1) After filtering by room ID, re-analyze the hourly load of the remaining rooms of the same type. Analyze the hourly load and its corresponding frequency to determine if a double peak exists. If so, extract the load between the two peaks and calculate the frequency change.

[0029] 2) Based on frequency (p) and frequency change These two variables were subjected to K-means ternary clustering, and the frequency (p) and frequency change of each cluster were calculated. Normalization is performed to obtain scaler_p, The normalized result yields the mean value of all points in its cluster, which is then incorporated into the new evaluation metric: Calculate the value of γ.

[0030] 3) Take the maximum load value in the cluster with the smallest γ value, which is the dividing line between "energy saving" and "non-energy saving".

[0031] This invention collects energy consumption data from all rooms in a building, using factors such as the rated cooling capacity and power of air conditioners, and room area to finely group the hourly load of each room. Then, based on outdoor temperature and set temperature, it benchmarks against massive amounts of historical energy consumption data from similar rooms. On this basis, it performs binning and frequency counting to determine the data distribution of similar energy consumption data. KMeans clustering is then used to divide the data into energy-saving data, non-energy-saving data, and transitional data between energy-saving and non-energy-saving categories. Finally, the decision interface between transitional and non-energy-saving data is identified as the final energy consumption benchmark. This invention proposes a dynamic energy-saving assessment benchmark, which can significantly improve the accuracy of energy consumption anomaly identification. For energy management projects, it can reduce operation and maintenance labor costs and provide a basis for building energy-saving management decisions.

Claims

1. A method for identifying energy consumption anomalies, characterized in that: include: The sampled data is converted into standard hourly data, and the energy consumption data of the hour before power-on is identified and abnormal data is identified and processed. The energy consumption data of all rooms in the corresponding building are used to group the hourly load of each room with a similar area based on the corresponding parameters of the air conditioner. Find the same type of hourly load within the group based on outdoor temperature and / or set temperature; Filter hourly loads of the same type based on room ID; determine peaks by analyzing the two variables of hourly load of the same type and the corresponding sub-container statistical frequency, and identify energy-saving peaks and non-energy-saving peaks. Statistical and cluster analysis of the same type of hourly load between the energy-saving peak and the non-energy-saving peak is performed to construct evaluation indicators and find the boundary threshold between energy-saving and non-energy-saving. If the energy consumption is higher than the threshold, the energy consumption is abnormal; otherwise, the energy consumption is normal. The relevant parameters of the air conditioner include rated cooling capacity and / or rated cooling power.

2. The energy consumption anomaly identification method according to claim 1, characterized in that: The sampled data includes rated power, rated cooling capacity, outdoor temperature and / or set temperature, or air conditioning parameters and / or operating environment data for all rooms in the corresponding building. The air conditioning parameters include air conditioner nameplate parameters, and the operating environment data includes room electrical parameters, air conditioning operation data and / or outdoor weather information.

Citation Information

Patent Citations

  • Decentralized flexible load monitoring system and monitoring method

    CN109660023A

  • Load decomposition method based on electrical appliance physical characteristics and residential electricity consumption behaviors

    CN111563827A