Property energy loss assessment method and system based on deep learning

Abstract

The invention relates to the technical field of energy assessment, in particular to a property energy loss assessment method and system based on deep learning, and the method comprises the steps: obtaining multi-source energy data and environment association data of a target property, and determining a basic energy consumption sequence in the multi-source energy data and a plurality of to-be-analyzed energy monitoring sub-regions; and according to the multi-source energy data, determining a time sequence correlation degree between each energy monitoring sub-region and the basic energy consumption sequence, and according to the time sequence correlation degree and the equipment operation characteristics of each energy monitoring sub-region, determining an energy consumption characteristic vector of each energy monitoring sub-region. According to the invention, through the analysis of the multi-source energy data, the energy consumption characteristics of different sub-regions can be evaluated in detail, so that the energy consumption loss evaluation is more accurate, a property manager can be helped to determine which regions have energy waste or loss problems, and through the calculation and analysis of the environmental sensitivity coefficient, the energy loss can be evaluated under different environmental conditions.

Description

Technical Field

[0001] This invention relates to the field of energy assessment technology, specifically to a method and system for assessing property energy loss based on deep learning. Background Technology

[0002] With the global energy shortage and increasing environmental awareness, more and more property managers and related industries are beginning to focus on how to reduce energy waste and lower operating costs through precise energy management. Traditional energy assessment methods often rely on human experience and are difficult to comprehensively and accurately assess and predict energy consumption. Therefore, there is an urgent need for a more intelligent and efficient energy management method.

[0003] Currently, with the widespread application of smart devices and sensors, energy metering data has become richer and more diverse. By collecting and analyzing this multi-source data, a more comprehensive understanding of energy usage in various sub-areas of a property can be achieved. However, extracting effective information from massive amounts of energy data, especially analyzing complex time-series data, remains a challenge.

[0004] Furthermore, as the scale of building and property management increases, managers need to identify and respond to energy loss issues in various areas more promptly and accurately, and quickly take effective optimization measures. At the same time, the energy consumption characteristics of different areas or equipment vary, and personalized optimization solutions can further improve energy efficiency and reduce costs. Therefore, a technical means is needed that can provide real-time feedback and personalized solutions based on specific circumstances. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: a deep learning-based method for assessing property energy consumption, comprising: Acquire multi-source energy data and environmental correlation data of the target property, and determine the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed in the multi-source energy data; The temporal correlation degree between each energy monitoring sub-region and the basic energy consumption sequence is determined based on the multi-source energy data, and the energy consumption feature vector of each energy monitoring sub-region is determined based on the temporal correlation degree and the equipment operation characteristics of each energy monitoring sub-region. Based on the energy consumption feature vector and the environmental correlation data of each energy monitoring sub-region, multiple high-loss candidate regions are determined from all energy monitoring sub-regions, and the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region is determined. The energy loss type label for each high-loss candidate region is determined based on the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region and the equipment operating characteristics.

[0006] Preferably, determining the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed in the multi-source energy data includes: Extract raw energy consumption data from multiple energy metering nodes of the target property from the multi-source energy data; The energy consumption stability coefficient of each energy metering node is determined based on the original energy consumption data of the multiple energy metering nodes. Based on all energy consumption stability coefficients and the spatiotemporal distribution characteristics of the multi-source energy data, a baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property.

[0007] Preferably, the raw energy consumption data of multiple energy metering nodes of the target property is extracted from the multi-source energy data, including: The multi-source energy data is preprocessed to obtain a standardized energy dataset; Determine the instantaneous energy consumption value and cumulative energy consumption value of each energy metering node in the standardized energy dataset; The reliability of energy consumption collection for each energy metering node is determined based on the instantaneous energy consumption value and the cumulative energy consumption value. Based on the reliability of energy consumption data collection at each energy metering node, the raw energy consumption data of multiple energy metering nodes for the target property are extracted from all energy metering nodes.

[0008] Preferably, the energy consumption stability coefficient of each energy metering node is determined based on the raw energy consumption data of the plurality of energy metering nodes, including: Construct a time-series network of energy consumption for the target property based on the raw energy consumption data from all energy metering nodes; Randomly select an energy metering node as the selected node, and determine the energy consumption fluctuation amplitude and fluctuation frequency of the selected node within a preset time period; The energy consumption stability coefficient of the selected node is determined based on the energy consumption fluctuation amplitude and the fluctuation frequency. The remaining energy metering nodes are selected as nodes, and the energy consumption stability coefficient of the remaining energy metering nodes is determined.

[0009] Preferably, based on all energy consumption stability coefficients and the spatiotemporal distribution characteristics of the multi-source energy data, a baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property, including: Obtain the spatial location information and energy consumption collection timestamp of each energy metering node; The spatiotemporal distribution feature matrix of the multi-source energy data is constructed based on the spatial location information of each energy metering node and the energy consumption collection timestamp. Obtain all energy metering nodes of the target property, and determine the spatial correlation index of energy consumption for each energy metering node based on the spatiotemporal distribution feature matrix; Randomly select an energy metering node and obtain the raw energy consumption data and spatial location information corresponding to the energy metering node; Obtain the energy consumption stability coefficient and energy consumption spatial correlation index corresponding to the energy metering node; The regional division weight of the energy metering node is determined based on the energy consumption spatial correlation index and energy consumption stability coefficient of the energy metering node. Continue selecting the remaining energy metering nodes and determine the regional division weights of the remaining energy metering nodes; Based on the regional division weights of all energy metering nodes, a baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property.

[0010] Preferably, determining the energy consumption feature vector of each energy monitoring sub-region based on the time-series correlation degree and the equipment operation characteristics of each energy monitoring sub-region includes: Obtain a standardized energy dataset from the multi-source energy data, and transform the standardized energy dataset into a time-series feature matrix; Extract the time-series sub-matrix corresponding to each energy monitoring sub-region from the time-series feature matrix; Determine the device operating status parameters for each time series submatrix; The equipment characteristic factors for each energy monitoring sub-region are determined based on the equipment operating status parameters. The energy consumption feature vector of each energy monitoring sub-region is constructed based on the device feature factors and the time-series correlation degree.

[0011] Preferably, multiple high-loss candidate areas are determined from all energy monitoring sub-regions based on the energy consumption feature vector and environmental correlation data for each energy monitoring sub-region, including: The environmental sensitivity coefficient of each energy monitoring sub-region is determined based on the environmental correlation data. Based on a pre-trained deep learning classification model, multiple high-loss candidate areas are determined from all energy monitoring sub-regions by combining the energy consumption feature vector and environmental sensitivity coefficient of each energy monitoring sub-region.

[0012] Preferably, the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region is determined, including: A high-loss candidate region is selected as the selected candidate region, and the energy consumption change rate sequence of each time node in the selected candidate region is determined. Multiple energy consumption fluctuation points are determined from all time points based on the energy consumption change rate sequence of each time point; Determine the temporal distribution entropy of all energy consumption fluctuation points in the selected candidate region; Continue to select the remaining high-loss candidate regions as selected candidate regions, and determine the temporal distribution entropy of all energy consumption fluctuation points in the remaining high-loss candidate regions.

[0013] Preferably, the energy loss type label of each high-loss candidate region is determined based on the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region and the equipment operating characteristics, including: A time window for detecting energy consumption anomalies in each high-loss candidate region is preset; Obtain the equipment operation feature matrix corresponding to each high-loss candidate region; The energy consumption anomaly characteristics of each high-loss candidate region are determined based on the time window and the equipment operation characteristic matrix. The loss risk score of each high-loss candidate region is determined based on the temporal distribution entropy and energy consumption anomaly characteristics of all energy consumption fluctuation points in each high-loss candidate region. The energy loss type label for each high-loss candidate region is determined based on the loss risk score of each high-loss candidate region.

[0014] A deep learning-based property energy consumption assessment system, applicable to the aforementioned deep learning-based property energy consumption assessment methods, including: The data acquisition unit is used to acquire multi-source energy data and environmental correlation data of the target property, and to determine the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed in the multi-source energy data. An energy consumption coding unit is used to determine the temporal correlation degree between each energy monitoring sub-region and the basic energy consumption sequence based on the multi-source energy data, and to determine the energy consumption feature vector of each energy monitoring sub-region based on the temporal correlation degree and the equipment operation characteristics of each energy monitoring sub-region. The time series analysis unit is used to determine multiple high-loss candidate areas from all energy monitoring sub-regions based on the energy consumption feature vector and the environmental correlation data of each energy monitoring sub-region, and to determine the time series distribution entropy of all energy consumption fluctuation points in each high-loss candidate area; An energy consumption assessment unit is used to determine the energy loss type label of each high-loss candidate region based on the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region and the equipment operating characteristics.

[0015] Compared with the prior art, the beneficial effects of the present invention are: (1) Through the analysis of multi-source energy data, this invention can evaluate the energy consumption characteristics of different sub-regions in detail, making the energy loss assessment more accurate and helping property managers to identify which areas have energy waste or loss problems; and by combining deep learning and time series analysis, it can accurately judge and predict the changing trend of energy loss, provide early warning, and help property managers optimize energy use strategies. (2) By automatically extracting and analyzing energy data, including equipment operating status and environmental correlation data, this invention can intelligently identify areas with high energy consumption and provide targeted optimization solutions, thereby saving energy and reducing management costs; and by combining time series analysis and equipment operating characteristics, it can achieve real-time data feedback, making the identification of energy consumption more timely, thereby helping to respond quickly and adjust energy management strategies. (3) By calculating and analyzing the environmental sensitivity coefficient, this invention can assess energy loss under different environmental conditions and ensure high energy efficiency under various changing conditions. Furthermore, by combining regional division and energy consumption feature vector, it can provide personalized solutions for energy consumption optimization in different regions or equipment, thereby improving overall energy efficiency. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0017] In the diagram: 1. Data acquisition unit; 2. Energy consumption coding unit; 3. Time series analysis unit; 4. Energy consumption assessment unit. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1, please refer to Figure 1 This invention provides a technical solution: a deep learning-based method for assessing property energy loss, comprising: S1. Obtain multi-source energy data and environmental correlation data of the target property, and determine the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed in the multi-source energy data; S2. Determine the temporal correlation degree of each energy monitoring sub-region and the basic energy consumption sequence based on multi-source energy data, and determine the energy consumption feature vector of each energy monitoring sub-region based on the temporal correlation degree and the equipment operation characteristics of each energy monitoring sub-region. S3. Based on the energy consumption feature vector and the environmental correlation data of each energy monitoring sub-region, determine multiple high-loss candidate regions from all energy monitoring sub-regions, and determine the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region; S4. Determine the energy loss type label for each high-loss candidate region based on the temporal distribution entropy of all energy consumption fluctuation points and the equipment operation characteristics.

[0020] It should be noted that the collection includes various energy data within the property (such as energy consumption data for electricity, air conditioning, and lighting systems), as well as related environmental data (such as temperature and humidity). The basic energy consumption sequence refers to the record of changes in these energy data over a period of time. Energy monitoring sub-areas are different parts of the property divided according to function or area; the energy consumption of each sub-area may differ. For example, one area might have high air conditioning consumption, while another might have high lighting consumption. For instance, suppose a large commercial office building collects energy consumption data for various areas (such as air conditioning areas, lighting areas, elevator areas, etc.). It also collects... Environmental data (such as temperature and humidity) from different regions were collected; time-series correlation analysis examines the relationship between each energy monitoring sub-region and the overall basic energy consumption sequence; for example, whether the energy consumption changes in the air-conditioned area are closely related to temperature changes; by calculating the time-series correlation between each sub-region and the basic energy consumption sequence, the energy consumption characteristics of each region can be obtained (for example, the energy consumption of the air-conditioned area is highly correlated with temperature changes, and the energy consumption of the elevator area is highly correlated with usage frequency); for example, analyzing the time-series relationship between the energy consumption of the air-conditioned area and changes in external temperature reveals that the energy consumption fluctuations in the air-conditioned area are significantly affected by temperature changes, thus allowing the calculation of the energy consumption characteristic vector of the air-conditioned area; Based on the energy consumption feature vectors and environmental data of each sub-region, it is determined which areas have potential energy loss problems. By calculating the temporal distribution entropy of energy consumption fluctuation points, it is further identified which areas have abnormally large energy consumption fluctuations, potentially indicating a high risk of energy loss. Temporal distribution entropy is used to measure the randomness or complexity of energy consumption fluctuations; areas with large fluctuations may indicate equipment instability or low energy efficiency. For example, based on the analysis of energy consumption characteristics and temperature data in the air-conditioned area, it was found that the air-conditioned system experienced large energy consumption fluctuations at certain times, unrelated to changes in external temperature, which may indicate that the air-conditioned system has low energy efficiency under certain conditions. Therefore, the air-conditioned area was marked as such. Each region is designated as a high-loss candidate region. For each high-loss candidate region, its loss type is further analyzed by combining the temporal distribution entropy and the operating characteristics of the equipment (such as runtime, workload, etc.). For example, the energy efficiency of an air conditioning system may decrease due to equipment aging or improper maintenance, while the energy consumption of an elevator system may be high due to frequent use. Each region is assigned an energy loss type label, such as high loss, normal loss, low loss, etc. For example, by analyzing the energy consumption fluctuations and operating characteristics of the air conditioning system, combined with temperature data, the loss type of the air conditioning system is ultimately labeled as "high loss," indicating that the energy use efficiency of this region is low and may need optimization.

[0021] In an optional embodiment, determining the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed from multi-source energy data includes: Extract raw energy consumption data from multiple energy metering nodes of the target property from multi-source energy data; The energy consumption stability coefficient of each energy metering node is determined based on the raw energy consumption data of multiple energy metering nodes. Based on all energy consumption stability coefficients and the spatiotemporal distribution characteristics of multi-source energy data, a baseline area corresponding to the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed are determined in the target property.

[0022] It should be noted that multi-source energy data typically includes multiple sources, such as electricity, water, and gas, each of which may be metered in different areas or on different devices. Each metering node represents a specific energy consumption record point, which may correspond to a certain area, device, or system. The raw energy consumption data of multiple energy metering nodes in a target property refers to the actual energy consumption data collected from these nodes. This data may be recorded in a time series, including data on electricity, water, and gas consumption within a certain period. For example, suppose there is a commercial building that includes multiple energy metering nodes, such as: electricity metering nodes: including electricity consumption for air conditioning, lighting, computer equipment, etc.; water metering nodes: including water consumption for toilets, kitchens, etc.; gas metering nodes: Natural gas consumption of kitchen equipment; collecting raw energy consumption data from these nodes, such as hourly electricity consumption and monthly water consumption; determining the energy consumption stability coefficient of each energy metering node based on the raw energy consumption data from multiple energy metering nodes: the energy consumption stability coefficient is determined by analyzing the stability of the energy consumption data of each metering node; for example, energy consumption in some areas may fluctuate drastically with changes in the external environment, while in other areas it may be relatively stable; the stability coefficient can be obtained by calculating the degree of energy consumption fluctuation at each node, the smaller the fluctuation, the higher the stability coefficient; usually, stability can be measured by statistical methods such as standard deviation and coefficient of variation; if the energy consumption fluctuation of a certain node is large, it may mean that the equipment in that area is malfunctioning. Problems have been identified (such as equipment failure or insufficient maintenance), requiring further analysis. For example, assuming we are analyzing the power metering nodes of an air conditioning system, in summer, the energy consumption of air conditioning may fluctuate significantly with changes in outdoor temperature, while in winter, the energy consumption may remain relatively stable. By calculating the energy consumption volatility of the air conditioning power metering nodes, we can obtain their stability coefficients. If the energy consumption of the air conditioning fluctuates significantly, it indicates that its stability coefficient is low, and system optimization may be necessary. Based on all energy consumption stability coefficients and the spatiotemporal distribution characteristics of multi-source energy data, we determine the benchmark area corresponding to the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed in the target property. In this step, we analyze the energy consumption stability coefficients of all energy metering nodes, combined with the temporal distribution of energy data. Based on the spatial distribution characteristics (e.g., the energy consumption variation pattern of a certain area over different time periods), we determine which areas serve as baseline areas and which areas are energy monitoring sub-areas requiring further analysis. Baseline areas are those with relatively stable energy consumption that meet expected energy efficiency standards, typically serving as a reference area for the overall property energy efficiency assessment. Energy monitoring sub-areas are those requiring further analysis; these areas may exhibit higher energy consumption fluctuations or poor stability, potentially requiring optimization measures. For example, assuming we are analyzing multiple areas (such as offices, kitchens, and meeting rooms) within a building, the energy consumption fluctuations at the electricity metering nodes in the office area are relatively small and not significantly affected by external ambient temperature, thus being identified as the baseline area.Energy consumption in the kitchen area fluctuates significantly, especially during peak dining hours, when electricity and gas consumption rise markedly, exhibiting a low energy stability coefficient. Therefore, the kitchen area has been identified as an energy monitoring sub-area, potentially requiring further optimization of equipment or control strategies.

[0023] In one optional embodiment, extracting raw energy consumption data from multiple energy metering nodes of the target property from multi-source energy data includes: Multi-source energy data is preprocessed to obtain a standardized energy dataset; Determine the instantaneous and cumulative energy consumption values ​​for each energy metering node in the standardized energy dataset; The reliability of energy consumption data collection for each energy metering node is determined based on instantaneous and cumulative energy consumption values. Based on the reliability of energy consumption data collection at each energy metering node, the raw energy consumption data of multiple energy metering nodes for the target property are extracted from all energy metering nodes.

[0024] It should be noted that, since the measurement units, metering methods, and data ranges for each energy source may differ, preprocessing is to convert all data into a unified standard format for subsequent analysis. Standardization eliminates differences between different energy data sources and typically includes unit conversion, data imputation, and outlier detection, allowing values ​​for different energy sources to be compared within the same framework. For example, in a building, there are three energy metering nodes: electricity, water, and natural gas. Electricity is measured in kilowatt-hours (kWh), water in cubic meters (m³), and natural gas in cubic meters (m³). Through preprocessing, the energy consumption data for water and natural gas are converted into a unified standard format. A standard unit (e.g., kilowatt-hours of total energy consumption) is used to fill in missing data, enabling the values ​​of the three energy sources to be compared and analyzed under the same standard. Instantaneous energy consumption refers to energy consumption data at a specific moment, usually expressed as the instantaneous rate of energy consumption; for example, the instantaneous energy consumption value of an electricity meter could be the amount of electricity consumed per minute. Cumulative energy consumption refers to the cumulative energy consumption over a period of time, which could be the sum of energy consumption per day, month, or period. Cumulative energy consumption helps to assess long-term energy efficiency; for example, for an electricity metering node: the instantaneous energy consumption value might be the amount of electricity consumed at a specific moment (e.g., the amount of electricity consumed in a certain hour: 1.2 kWh); the cumulative energy consumption value is the total energy consumption over a certain period of time. The total electricity consumption of a device or area since a certain point in time (e.g., the device's total electricity consumption in January: 150 kWh); assessing the reliability of energy consumption data for each energy metering node; typically, reliability is assessed based on indicators such as data completeness, volatility, and stability; if the instantaneous energy consumption value of a metering node changes too much, or the cumulative energy consumption value is abnormal, it may indicate a problem with the collected data (e.g., equipment failure, sensor error, data transmission problems, etc.); nodes with low reliability may need to be recalibrated or ignored; for example: suppose the gas meter in the kitchen shows an abnormal data point, with the instantaneous energy consumption value suddenly increasing significantly at a certain moment. The error might occur during data collection, which would lower the reliability of that node's data collection. If the long-term data shows stable changes, the reliability is higher. Based on the reliability of each energy metering node's data collection, raw energy consumption data from multiple energy metering nodes of the target property is extracted from all energy metering nodes. By evaluating the reliability of each metering node, nodes with high reliability are selected as valid data sources. The aim is to extract trusted and verified energy consumption data, ignoring nodes with poor data quality or serious collection problems. Ultimately, the energy consumption analysis of the target property will be based on reliable and verified data to ensure the accuracy of the energy efficiency assessment.

[0025] In one optional embodiment, determining the energy consumption stability coefficient of each energy metering node based on the raw energy consumption data of multiple energy metering nodes includes: Construct a time-series network of energy consumption for the target property based on the raw energy consumption data from all energy metering nodes; Randomly select an energy metering node as the selected node, and determine the energy consumption fluctuation amplitude and frequency of the selected node within a preset time period; The energy consumption stability coefficient of the selected node is determined based on the energy consumption fluctuation amplitude and fluctuation frequency. The remaining energy metering nodes are selected as nodes, and the energy consumption stability coefficient of the remaining energy metering nodes is determined.

[0026] It should be noted that an "energy consumption time series network" is established by collecting raw energy consumption data from various energy metering nodes (such as electricity, water, gas, etc.). This network can be viewed as a data structure containing all nodes and their energy consumption relationships, typically used to analyze energy consumption trends, correlations, and other dynamic characteristics. An energy metering node is randomly selected as the chosen node, and its energy consumption fluctuation amplitude and frequency within a preset time period are determined. A node (e.g., an electricity metering node) is randomly selected from all energy metering nodes. Within the selected time period (e.g., one hour, one day, one week, etc.), the node's energy consumption fluctuation amplitude (i.e., the difference between the maximum and minimum energy consumption) and fluctuation frequency (i.e., the frequency of energy consumption changes) need to be measured. The fluctuation amplitude can be obtained by calculating the difference between the node's maximum and minimum energy consumption. The frequency of energy consumption fluctuations is calculated per unit of time, usually by counting the number of times energy consumption changes reach a certain threshold. The energy consumption stability coefficient is used to assess the degree of energy consumption fluctuation at a node. Generally, nodes with larger fluctuation amplitudes or higher fluctuation frequencies will have lower stability coefficients. The stability coefficient is a comprehensive indicator based on fluctuation amplitude and fluctuation frequency. It can be calculated using some mathematical formulas, such as: stability coefficient = 1 / (fluctuation amplitude × fluctuation frequency), indicating that the more stable the energy consumption change, the larger the coefficient. The remaining energy metering nodes are selected as nodes, and the energy consumption stability coefficient of the remaining energy metering nodes is determined. After calculating the stability coefficient of one node, the same calculation is performed on all other energy metering nodes to evaluate the stability coefficient of each node, and finally the stability coefficient of all energy metering nodes in the property is obtained.

[0027] In an optional embodiment, based on all energy consumption stability coefficients and the spatiotemporal distribution characteristics of multi-source energy data, a baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property, including: Obtain the spatial location information and energy consumption collection timestamp of each energy metering node; A spatiotemporal distribution feature matrix of multi-source energy data is constructed based on the spatial location information of each energy metering node and the energy consumption collection timestamp. Obtain all energy metering nodes of the target property, and determine the spatial correlation index of energy consumption for each energy metering node based on the spatiotemporal distribution feature matrix; Randomly select an energy metering node and obtain the raw energy consumption data and spatial location information corresponding to the energy metering node; Obtain the energy consumption stability coefficient and energy consumption spatial correlation index corresponding to the energy metering node; The regional division weights of energy metering nodes are determined based on the spatial correlation index and energy stability coefficient of energy consumption. Continue selecting the remaining energy metering nodes and determining the regional division weights of the remaining energy metering nodes; Based on the regional division weights of all energy metering nodes, the baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property.

[0028] It should be noted that each energy metering node not only has energy consumption data, but also geographical location information (such as floor, area, etc.) and a timestamp (recording the specific time of energy consumption data collection). This information helps to analyze energy consumption in both space and time. A spatiotemporal distribution feature matrix is ​​constructed by combining the spatial location information and energy consumption timestamps of all energy metering nodes. This matrix can display the energy consumption characteristics of different regions and time periods, which helps to analyze the spatial distribution and temporal changes of energy use. All energy metering nodes of the target property are acquired, and the spatial correlation index of energy consumption for each energy metering node is determined based on the spatiotemporal distribution feature matrix. Based on the above matrix, the spatial correlation index of energy consumption for each node is determined, that is, the correlation between the energy consumption of that node and the energy consumption of the surrounding area and other nodes. For example, there may be some correlation between the electricity consumption of a certain floor and the water consumption of the same floor. An energy metering node (such as an electricity metering node) is randomly selected, and its data is collected. The system collects energy consumption data and spatial location information, such as the floor and area where the node is located. Based on the node's historical data, it calculates its energy consumption stability coefficient (as mentioned above, a comprehensive index of fluctuation amplitude and frequency) and energy consumption spatial correlation index to assess the node's energy consumption stability and its correlation with other nodes. Using the energy consumption stability coefficient and spatial correlation index, it determines the node's regional division weight. This indicates the degree of influence of the node's energy consumption characteristics on the overall energy consumption analysis. If a node has small energy consumption fluctuations and high correlation with other nodes, its weight in regional division will be relatively large. Similar calculations are performed on each remaining energy metering node in the target property to determine their respective regional division weights. Based on the regional division weights of all nodes, a benchmark area (e.g., an area with typical energy consumption performance) and several energy monitoring sub-areas to be analyzed are determined in the property. These areas can be used for more refined energy efficiency management and optimization.

[0029] In an optional embodiment, determining the energy consumption characteristic vector of each energy monitoring sub-region based on time-series correlation and equipment operating characteristics of each energy monitoring sub-region includes: Obtain a standardized energy dataset from multi-source energy data and transform the standardized energy dataset into a time-series feature matrix; Extract the time-series sub-matrix corresponding to each energy monitoring sub-region from the time-series feature matrix; Determine the device operating status parameters for each time series submatrix; Determine the equipment characteristic factors for each energy monitoring sub-region based on the equipment operating status parameters; Energy consumption feature vectors for each energy monitoring sub-region are constructed based on equipment characteristic factors and temporal correlation.

[0030] It should be noted that standardized datasets typically have a time dimension, with energy consumption data at each point in time constituting a data record. These data are integrated into a time-series feature matrix, where rows represent different time points and columns represent different energy metering nodes or energy consumption types. The target property may be divided into multiple sub-regions, such as different floors, areas, or groupings of areas. Each sub-region may consist of multiple energy metering nodes. A sub-matrix of data from each energy monitoring sub-region is extracted, representing the time-series data of each energy node within that region. Equipment operating status parameters are usually related to the equipment's operating status (on / off, load, etc.) and can be obtained through historical data analysis, equipment communication protocols, etc. For example, equipment may experience high load or shutdown at a certain time, affecting its energy consumption pattern. These status parameters typically include the equipment's load rate, Operating time, on / off status, efficiency, etc.; equipment characteristic factors reflect the performance and influence of equipment; based on the operating status, load, efficiency, etc. of the equipment, the equipment characteristic factors of each sub-area are calculated; these factors can help assess the contribution of equipment to the energy consumption of the sub-area; for example, the equipment characteristic factor of the air conditioning system may be proportional to the load rate; for example, for the second floor, the equipment characteristic factor of the air conditioning system may be calculated as follows: Equipment 1 (air conditioning) characteristic factor = air conditioning load rate (0.8) * equipment efficiency (0.9) = 0.72; the energy consumption characteristic vector is a vector constructed based on equipment characteristic factors, time series correlation (time correlation, such as the similarity of energy consumption patterns in different time periods) and other possible factors (such as weather, weekdays, etc.); this vector contains information such as energy efficiency, load, equipment status, and time series data of all energy equipment in the area.

[0031] In an optional embodiment, multiple high-loss candidate areas are determined from all energy monitoring sub-regions based on energy consumption feature vectors and environmental correlation data for each energy monitoring sub-region, including: The environmental sensitivity coefficient of each energy monitoring sub-region is determined based on environmental correlation data; Based on a pre-trained deep learning classification model, multiple high-loss candidate areas are identified from all energy monitoring sub-regions by combining the energy consumption feature vector and environmental sensitivity coefficient of each energy monitoring sub-region.

[0032] It should be noted that environmentally related data typically refers to external data related to factors such as climate, weather, season, outdoor temperature, humidity, and air quality. These factors can significantly impact energy use (such as electricity and air conditioning energy consumption). For example, air conditioning load increases when the weather temperature is high. The environmental sensitivity coefficient is used to measure the degree of influence of environmental factors on energy consumption in a certain area. Each energy monitoring sub-area will have a corresponding environmental sensitivity coefficient, reflecting the intensity of the area's energy consumption response to environmental factors (such as temperature and humidity). For example, suppose there are two floors, the first floor and the second floor. The first floor is close to the exterior wall, and the temperature fluctuates greatly, resulting in a large fluctuation in air conditioning load. The first floor is more affected by external temperature; while the second floor, located inside the building, is less affected by ambient temperature. By analyzing environmental data, different environmental sensitivity coefficients can be calculated for the first and second floors. The environmental sensitivity coefficient for the first floor is 1.5 (indicating a greater impact from external temperature); the environmental sensitivity coefficient for the second floor is 0.8 (indicating a lesser impact from external temperature). Deep learning classification models (such as neural networks, decision trees, etc.) are typically used for classification or prediction based on large amounts of feature data. In this scenario, the model has been trained to identify areas that may have high energy consumption risks. Energy consumption feature vectors (the vector mentioned in the previous step that includes device feature factors and temporal correlation) and environmental sensitivity... The coefficients (reflecting the sensitivity of regional energy consumption to environmental changes) will be used as input to the model. Through model prediction, combined with these input data, multiple candidate regions with potential high energy consumption problems can be identified. For example, assuming a deep learning model has been trained, it can determine whether a region belongs to a high-energy-consumption area based on the input feature vectors (including energy consumption data, equipment load, environmental sensitivity coefficients, etc.). Input: Energy consumption feature vector for floor 1: [0.8, 0.7, 0.75, 1.5] (representing the feature factors of electricity, air conditioning, and other equipment, and the environmental sensitivity coefficient); Energy consumption feature vector for floor 2: [0.6, 0.5, 0.65, 0.8] (…). (This refers to the characteristic factors and environmental sensitivity coefficients of equipment such as electricity and air conditioning). Through model calculations, the first floor, due to its high environmental sensitivity coefficient and high energy consumption characteristic vector, may be identified as a high-loss area, while the second floor may be identified as a normal energy consumption area. High-loss candidate areas refer to those areas with abnormally high energy consumption according to the model prediction results. By combining the results of the deep learning model and the environmental sensitivity coefficient, it is possible to accurately identify which areas have problematic energy usage patterns and are worth further optimization. These candidate areas can be used for energy efficiency assessment, and further optimization measures can be taken, such as adjusting equipment settings, improving insulation effects, and optimizing air conditioning load.

[0033] In an optional embodiment, determining the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region includes: Select a high-loss candidate region as the selected candidate region, and determine the energy consumption change rate sequence for each time node in the selected candidate region; Multiple energy consumption fluctuation points are determined from all time points based on the energy consumption change rate sequence at each time point; Determine the temporal distribution entropy of all energy consumption fluctuation points in the selected candidate region; Continue to select the remaining high-loss candidate regions as selected candidate regions, and determine the temporal distribution entropy of all energy consumption fluctuation points in the remaining high-loss candidate regions.

[0034] It should be noted that, firstly, based on the preceding analysis, high-energy-consumption areas are identified, and then one of these areas is selected for further analysis; this area will become the selected candidate area. The energy consumption change rate represents the magnitude of change in energy consumption at a given time point compared to the previous time point; it reflects energy consumption fluctuations and helps identify abnormal energy consumption changes. The energy consumption change rate sequence is a sequence of changes calculated based on the energy consumption data for each time point. For example, suppose the selected candidate area is an office area, and hourly energy consumption data is recorded: Hour 1 energy consumption: 100 kWh; Hour 2 energy consumption: 110 kWh; Hour 3 energy consumption: 120kWh; Energy consumption change rate: Change rate in the 2nd hour = (110-100) / 100 = 0.1 (a 10% increase); Change rate in the 3rd hour = (120-110) / 110 = 0.0909 (an approximately 9% increase); These change rates constitute the energy consumption change rate sequence: [0.1, 0.0909]; Energy consumption fluctuation points refer to points in the energy consumption change rate sequence where significant changes occur, which may indicate abnormal energy consumption fluctuations during that period; A threshold can be set to identify points where the change rate exceeds a certain standard as fluctuation points; For example, if the energy consumption change rate exceeds 10% (i.e., 0.1), that time point is considered the energy consumption fluctuation point. Fluctuation point; Suppose the observed energy consumption change rate from hour 1 to hour 2 is 10%, which meets the criteria for a fluctuation point, therefore hour 2 is an energy consumption fluctuation point; Time series distribution entropy is an indicator that measures the uncertainty or complexity in a time series; it can be used to quantify the distribution pattern of energy consumption fluctuation points in a time series, helping to determine whether there are regular or abnormal fluctuations; the higher the time series distribution entropy, the more complex the changes in energy consumption fluctuations, which may indicate that the energy consumption fluctuations in this region have no clear pattern and there is room for optimization; for example, if the distribution time of energy consumption fluctuation points in the selected candidate area is relatively regular and the time series distribution entropy is low, it may indicate that the energy consumption fluctuations in this region are relatively stable; If the temporal distribution of the energy consumption fluctuation points is very irregular and the temporal distribution entropy is high, it indicates that the energy consumption fluctuations in this area are exceptionally complex and require further optimization. Once the first candidate area is selected and its energy consumption fluctuation points and temporal distribution entropy are analyzed, the remaining high-loss candidate areas are selected, and the above steps are repeated to analyze the energy consumption fluctuation points and temporal distribution entropy of each candidate area. For example, assuming the first selected candidate area is the 1st floor, after analysis, its temporal distribution entropy of energy consumption fluctuation points is found to be 0.85. Next, the 2nd floor is selected as the next candidate area, and its energy consumption change rate sequence, fluctuation points, and temporal distribution entropy are repeatedly calculated. Finally, the temporal distribution entropy of the 2nd floor is found to be 0.60.

[0035] In an optional embodiment, the energy loss type label for each high-loss candidate region is determined based on the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region and the equipment operating characteristics, including: Pre-set the time window for detecting energy consumption anomalies in each high-loss candidate region; Obtain the equipment operation feature matrix corresponding to each high-loss candidate region; The energy consumption anomaly characteristics of each high-loss candidate region are determined based on the time window and the equipment operation characteristic matrix. The loss risk score of each high-loss candidate region is determined based on the temporal distribution entropy and energy consumption anomaly characteristics of all energy consumption fluctuation points in each high-loss candidate region. The energy loss type label for each high-loss candidate region is determined based on the loss risk score of each high-loss candidate region.

[0036] It's important to note that, firstly, to monitor energy consumption anomalies, a time window is defined. Within this window, energy consumption fluctuations in that area are closely monitored. For example, the time window can be hourly, daily, or weekly, depending on the required level of analytical granularity. For instance, to analyze the energy consumption of an office building, an hourly time window could be chosen to monitor energy consumption fluctuations. Thus, hourly energy consumption data becomes an analytical unit. For each high-loss candidate area, characteristic data related to the operation of equipment within that area is collected. These characteristics typically include the equipment's operating status, load, operating time, start-up and shutdown times, etc. The feature matrix helps to understand the equipment's operating status over a specific time period. For example, for the first floor of an office building, assuming the floor has equipment such as air conditioning, lighting, and elevators, the hourly operating status of these devices (e.g., whether the air conditioning is on, whether the lighting is on, etc.) and their power consumption and running time can be recorded. Based on a set time window and equipment operating characteristic matrix, the energy consumption anomaly characteristics of each high-loss candidate area can be calculated. These characteristics may include abrupt changes in energy consumption (e.g., a sudden increase or decrease in energy consumption over a certain period of time), and abnormal relationships between equipment operating status and energy consumption. For example, assuming that on the first floor, within a certain time window, the air conditioning is running for a long time, while the lighting equipment is frequently switched on and off, and the elevator runs for a long time multiple times; based on the operating characteristics of these devices and their energy consumption data... This can identify energy consumption anomalies within a given time window; for example, if the air conditioner is left on for an hour, it can cause an abnormal increase in energy consumption during that period. The temporal distribution entropy of energy consumption fluctuation points measures the complexity of energy consumption fluctuations. A higher entropy value indicates more irregular energy consumption fluctuations in the area, which may mean improper system management or anomalies. If energy consumption changes abnormally frequently or equipment energy efficiency decreases in an area, it will increase the area's loss risk score. For example, assuming the temporal distribution entropy of energy consumption fluctuation points on the first floor is 0.85, it indicates that the energy consumption fluctuations in that area are relatively complex and irregular. In addition, the air conditioner's energy consumption is abnormally high, and the elevator frequently starts and stops at night. Combining these factors, a loss risk score for the area is calculated. If the score is high, This indicates that the energy consumption on the first floor is abnormally high. Based on the loss risk score, an energy loss type label can be assigned to each high-loss candidate area. The label helps identify the nature of the energy consumption problem in that area, such as: over-operation (e.g., equipment running for extended periods, resulting in excessive energy consumption); unreasonable load (equipment operating under excessive load, leading to low energy efficiency); poor management (equipment not being turned on or off at appropriate times); equipment failure (e.g., equipment running continuously but inefficiently). For example, assuming the loss risk score on the first floor is high, and the analysis shows that the air conditioning equipment has not been turned off for a long time and the elevator is frequently started, the first floor can be labeled as "over-operation" based on these issues. Other floors may have different labels, such as unreasonable load or equipment failure.

[0037] Example 2, please refer to Figure 2 This invention provides a technical solution: a deep learning-based property energy loss assessment system, applicable to the aforementioned deep learning-based property energy loss assessment method, comprising: Data acquisition unit 1 is used to acquire multi-source energy data and environmental correlation data of the target property, and to determine the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed in the multi-source energy data. Energy consumption coding unit 2 is used to determine the temporal correlation degree of each energy monitoring sub-region and the basic energy consumption sequence based on multi-source energy data, and to determine the energy consumption feature vector of each energy monitoring sub-region based on the temporal correlation degree and the equipment operation characteristics of each energy monitoring sub-region. The time series analysis unit 3 is used to determine multiple high-loss candidate areas from all energy monitoring sub-regions based on the energy consumption feature vector and the environmental correlation data of each energy monitoring sub-region, and to determine the time series distribution entropy of all energy consumption fluctuation points in each high-loss candidate area; Energy consumption assessment unit 4 is used to determine the energy loss type label of each high loss candidate area based on the temporal distribution entropy of all energy consumption fluctuation points in each high loss candidate area and the equipment operation characteristics.

[0038] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A deep learning-based method for assessing property energy loss, characterized in that, include: Acquire multi-source energy data and environmental correlation data of the target property, and determine the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed in the multi-source energy data; The temporal correlation degree between each energy monitoring sub-region and the basic energy consumption sequence is determined based on the multi-source energy data, and the energy consumption feature vector of each energy monitoring sub-region is determined based on the temporal correlation degree and the equipment operation characteristics of each energy monitoring sub-region. Based on the energy consumption feature vector and the environmental correlation data of each energy monitoring sub-region, multiple high-loss candidate regions are determined from all energy monitoring sub-regions, and the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region is determined. The energy loss type label for each high-loss candidate region is determined based on the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region and the equipment operating characteristics.

2. The deep learning-based property energy loss assessment method according to claim 1, characterized in that, Determine the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed from the multi-source energy data, including: Extract raw energy consumption data from multiple energy metering nodes of the target property from the multi-source energy data; The energy consumption stability coefficient of each energy metering node is determined based on the original energy consumption data of the multiple energy metering nodes. Based on all energy consumption stability coefficients and the spatiotemporal distribution characteristics of the multi-source energy data, a baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property.

3. The deep learning-based property energy loss assessment method according to claim 2, characterized in that, The raw energy consumption data of multiple energy metering nodes of the target property are extracted from the multi-source energy data, including: The multi-source energy data is preprocessed to obtain a standardized energy dataset; Determine the instantaneous energy consumption value and cumulative energy consumption value of each energy metering node in the standardized energy dataset; The reliability of energy consumption collection for each energy metering node is determined based on the instantaneous energy consumption value and the cumulative energy consumption value. Based on the reliability of energy consumption data collection at each energy metering node, the raw energy consumption data of multiple energy metering nodes for the target property are extracted from all energy metering nodes.

4. The deep learning-based property energy loss assessment method according to claim 3, characterized in that, The energy consumption stability coefficient of each energy metering node is determined based on the raw energy consumption data of the multiple energy metering nodes, including: Construct a time-series network of energy consumption for the target property based on the raw energy consumption data from all energy metering nodes; Randomly select an energy metering node as the selected node, and determine the energy consumption fluctuation amplitude and fluctuation frequency of the selected node within a preset time period; The energy consumption stability coefficient of the selected node is determined based on the energy consumption fluctuation amplitude and the fluctuation frequency. The remaining energy metering nodes are selected as nodes, and the energy consumption stability coefficient of the remaining energy metering nodes is determined.

5. The deep learning-based property energy loss assessment method according to claim 4, characterized in that, Based on all energy consumption stability coefficients and the spatiotemporal distribution characteristics of the multi-source energy data, a baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property, including: Obtain the spatial location information and energy consumption collection timestamp of each energy metering node; The spatiotemporal distribution feature matrix of the multi-source energy data is constructed based on the spatial location information of each energy metering node and the energy consumption collection timestamp. Obtain all energy metering nodes of the target property, and determine the spatial correlation index of energy consumption for each energy metering node based on the spatiotemporal distribution feature matrix; Randomly select an energy metering node and obtain the raw energy consumption data and spatial location information corresponding to the energy metering node; Obtain the energy consumption stability coefficient and energy consumption spatial correlation index corresponding to the energy metering node; The regional division weight of the energy metering node is determined based on the energy consumption spatial correlation index and energy consumption stability coefficient of the energy metering node. Continue selecting the remaining energy metering nodes and determine the regional division weights of the remaining energy metering nodes; Based on the regional division weights of all energy metering nodes, a baseline region corresponding to the basic energy consumption sequence and multiple energy monitoring sub-regions to be analyzed are determined in the target property.

6. The deep learning-based property energy loss assessment method according to claim 5, characterized in that, The energy consumption feature vector of each energy monitoring sub-region is determined based on the time-series correlation degree and the equipment operation characteristics of each energy monitoring sub-region, including: Obtain a standardized energy dataset from the multi-source energy data, and convert the standardized energy dataset into a time-series feature matrix; Extract the time-series sub-matrix corresponding to each energy monitoring sub-region from the time-series feature matrix; Determine the device operating status parameters for each time series submatrix; The equipment characteristic factors for each energy monitoring sub-region are determined based on the equipment operating status parameters. The energy consumption feature vector of each energy monitoring sub-region is constructed based on the device feature factors and the time-series correlation degree.

7. The deep learning-based property energy loss assessment method according to claim 6, characterized in that, Based on the energy consumption feature vector and environmental correlation data of each energy monitoring sub-region, multiple high-loss candidate areas are identified from all energy monitoring sub-regions, including: The environmental sensitivity coefficient of each energy monitoring sub-region is determined based on the environmental correlation data. Based on a pre-trained deep learning classification model, multiple high-loss candidate areas are determined from all energy monitoring sub-regions by combining the energy consumption feature vector and environmental sensitivity coefficient of each energy monitoring sub-region.

8. The deep learning-based property energy loss assessment method according to claim 7, characterized in that, Determine the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region, including: A high-loss candidate region is selected as the selected candidate region, and the energy consumption change rate sequence of each time node in the selected candidate region is determined. Multiple energy consumption fluctuation points are determined from all time points based on the energy consumption change rate sequence of each time point; Determine the temporal distribution entropy of all energy consumption fluctuation points in the selected candidate region; Continue to select the remaining high-loss candidate regions as selected candidate regions, and determine the temporal distribution entropy of all energy consumption fluctuation points in the remaining high-loss candidate regions.

9. The deep learning-based property energy loss assessment method according to claim 8, characterized in that, The energy loss type label for each high-loss candidate region is determined based on the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region and the equipment operating characteristics, including: A time window for detecting energy consumption anomalies in each high-loss candidate region is preset; Obtain the equipment operation feature matrix corresponding to each high-loss candidate region; The energy consumption anomaly characteristics of each high-loss candidate region are determined based on the time window and the equipment operation characteristic matrix. The loss risk score of each high-loss candidate region is determined based on the temporal distribution entropy and energy consumption anomaly characteristics of all energy consumption fluctuation points in each high-loss candidate region. The energy loss type label for each high-loss candidate region is determined based on the loss risk score of each high-loss candidate region.

10. A deep learning-based property energy loss assessment system, applicable to the deep learning-based property energy loss assessment method described in any one of claims 1-9, characterized in that, include: The data acquisition unit is used to acquire multi-source energy data and environmental correlation data of the target property, and to determine the basic energy consumption sequence and multiple energy monitoring sub-areas to be analyzed in the multi-source energy data. An energy consumption coding unit is used to determine the temporal correlation degree between each energy monitoring sub-region and the basic energy consumption sequence based on the multi-source energy data, and to determine the energy consumption feature vector of each energy monitoring sub-region based on the temporal correlation degree and the equipment operation characteristics of each energy monitoring sub-region. The time series analysis unit is used to determine multiple high-loss candidate areas from all energy monitoring sub-areas based on the energy consumption feature vector and the environmental correlation data of each energy monitoring sub-area, and to determine the time series distribution entropy of all energy consumption fluctuation points in each high-loss candidate area; An energy consumption assessment unit is used to determine the energy loss type label of each high-loss candidate region based on the temporal distribution entropy of all energy consumption fluctuation points in each high-loss candidate region and the equipment operating characteristics.

Images