An energy consumption usage data management system and method based on edge computing

Abstract

The application discloses an energy consumption use data management system and method based on edge computing, and relates to the technical field of data management.The method comprises the following steps: collecting energy consumption data by using an edge device, binding geographical coordinates and time stamps for each piece of data, superimposing a device-level feature code, and generating a unique digital fingerprint; when it is detected that the energy consumption value deviates from an adaptive threshold and the feature matching degree is lower than a feature matching degree threshold, triggering a high-confidence abnormality marking mechanism automatically, and generating a hot spot event record; managing the collected data according to a three-level storage strategy, wherein the three-level storage comprises device-end storage, edge-end storage and cloud-end storage; based on space-time coding, quickly positioning geographical adjacent edge nodes, aggregating the running states of geographical adjacent devices, generating a local optimization instruction through a preset strategy rule library, generating an optimization strategy package by the cloud based on multi-dimensional information, and after virtual verification, directing the optimization strategy package to the target device end.

Description

Technical Field

[0001] This invention relates to the field of data management technology, specifically to an energy consumption data management system and method based on edge computing. Background Technology

[0002] With the acceleration of global digital transformation, the number of IoT devices deployed in energy management systems is growing exponentially. Traditional centralized architectures face problems such as high data transmission bandwidth pressure, weak real-time response capabilities, and lagging global decision-making.

[0003] Energy consumption data generated by different regions and different types of equipment are characterized by uneven spatial and temporal distribution, large differences in features, and rapid dynamic changes. Traditional anomaly detection methods based on fixed thresholds are difficult to adapt to complex and ever-changing energy consumption scenarios, easily leading to false alarms or missed alarms. Existing systems often use a single-level data storage approach, which cannot effectively distinguish the value of data, resulting in a waste of cloud storage resources and affecting the rapid retrieval and analysis of historical data.

[0004] The complexity of energy systems is also reflected in the difficulty of multi-source data fusion, the lack of unified feature encoding and correlation analysis mechanisms for multi-dimensional data, and the difficulty in uncovering deep energy consumption patterns. In terms of optimization and control, there is a lack of collaborative decision-making mechanisms between the edge and cloud, resulting in delayed responses to local anomalies and a lack of precision in global policy deployment, failing to meet the real-time requirements of energy dispatching scenarios. Simultaneously, edge nodes have limited computing resources, and when faced with sudden traffic surges or complex computing tasks, the lack of effective load balancing and task migration mechanisms can easily lead to system crashes or service interruptions. Summary of the Invention

[0005] The purpose of this invention is to provide an energy consumption data management system and method based on edge computing to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for managing energy consumption data based on edge computing, the method comprising:

[0007] Energy consumption data is collected using edge devices. Each data point is bound to geographic coordinates and timestamps, and device-level feature codes are overlaid to generate a unique digital fingerprint. The original data is then compressed using differential encoding.

[0008] The edge node performs multi-dimensional feature matching between the digital fingerprint and the historical digital fingerprint database. When the energy consumption value deviates from the adaptive threshold and the feature matching degree is lower than the feature matching degree threshold, the high confidence anomaly marking mechanism is automatically triggered and hot event records are generated.

[0009] The collected data is managed according to a three-tier storage strategy, which includes device-side storage, edge storage, and cloud storage. The device-side storage uses a circular caching mechanism to temporarily cache the original data. The edge storage stores details of hot events and aggregated statistical indicators, and selects high-value data to the cloud based on the data value assessment results. The cloud storage permanently stores cross-regional correlation information and integrates external parameters to build a global analysis model.

[0010] Based on spatiotemporal coding, the edge nodes of geographical proximity are quickly located, the operating status of geographically nearby devices is aggregated, local optimization instructions are generated through a preset policy rule base, and the cloud integrates multi-dimensional information to generate an optimization policy package, which is then sent to the target device after virtual verification.

[0011] During the strategy operation, the running status of each node is monitored in real time, and the strategy execution effect is fed back to the cloud. When the resource utilization data of any edge node exceeds the preset warning threshold, the task migration process based on geographical location priority is automatically started, and the data transmission mode is dynamically switched according to the real-time channel quality.

[0012] According to the above scheme, the energy consumption data includes power data, water supply data, and environmental data; the power data includes voltage data, current data, active power data, and reactive power data; the water supply data includes water flow rate data and water pressure data; and the environmental data includes temperature data and humidity data.

[0013] The system obtains the precise geographic location of edge data, converts the longitude and latitude data of the geographic location into GeoHash encoding, and adds a timestamp to each energy consumption data by real-time synchronization with the network time protocol.

[0014] By analyzing waveform variation characteristics using voltage and current data, and calculating voltage fluctuation frequency characteristics, current harmonic component characteristics, active power change rate characteristics, and reactive power change rate characteristics, the characteristics of power data are analyzed. Water supply data characteristics are analyzed based on the variation gradient of water flow velocity data and the fluctuation amplitude of water pressure data, combined with the correlation between water flow velocity and water pressure. Environmental data characteristics are analyzed using the rate of temperature rise and fall and the range of humidity variation.

[0015] The characteristics of power data, water supply data, and environmental data are normalized and integrated to generate an equipment-level feature code; geographic coordinate data, timestamp data, and the equipment-level feature code are combined using a cryptographic hash algorithm to generate a fixed-length digital fingerprint.

[0016] First-order difference is used to calculate the difference between adjacent data points, and the difference sequence is compressed by run-length encoding to reduce data transmission volume and storage usage.

[0017] According to the above scheme, the current digital fingerprint is matched and compared with historical digital fingerprints with the same geographical coordinates stored in the historical digital fingerprint database. The matching and comparison includes periodic features based on time series analysis, geographical relevance features based on GeoHash encoding, and operating status features based on feature parameters. The periodic similarity of the time dimension is calculated by dynamic time warping algorithm, the geographical relevance of the spatial dimension is evaluated by GeoHash prefix matching degree, and the operating status difference of the device dimension is calculated by Euclidean distance based on feature parameters.

[0018] An adaptive threshold range is dynamically generated based on historical data. The adaptive threshold range includes an energy consumption threshold, a feature matching degree threshold, and a fluctuation confidence interval.

[0019] When energy consumption data is detected to deviate from the adaptive threshold range and the feature matching degree is lower than the feature matching degree threshold, a high confidence anomaly marking mechanism is automatically triggered. The anomaly marking mechanism includes spatiotemporal location marking, deviation degree quantification, feature difference description and anomaly type classification.

[0020] Generate hotspot event records that include an anomaly marking mechanism, and assign a unique identifier code to each hotspot event record.

[0021] According to the above scheme, the dynamically generated adaptive threshold range includes:

[0022] Extract historical energy consumption data and corresponding historical digital fingerprints of the same device identifier from the historical digital fingerprint database. The historical energy consumption data covers different time periods and operating conditions of the device during normal operation.

[0023] Based on the historical energy consumption data, the mean, variance, and extreme values ​​of energy consumption values ​​over time are calculated, a basic statistical model of energy consumption values ​​is constructed, and the initial range of energy consumption value thresholds is determined.

[0024] The periodicity of the historical digital fingerprint in the time dimension, the geographical relevance in the spatial dimension, and the operating status in the device dimension are analyzed. The matching similarity distribution of each dimension is calculated, and the basic value of the feature matching degree threshold is set according to the distribution.

[0025] Based on the fluctuations in the matching degree between energy consumption values ​​and characteristics in historical energy consumption data, statistical analysis methods are used to determine the confidence interval of fluctuations.

[0026] The statistical analysis methods include:

[0027] Historical energy consumption data is segmented according to time series, and the segmentation is based on equipment operating cycle, operating condition change nodes, change cycle of external environmental influencing factors, and similarity of data characteristics.

[0028] Based on each data segment, the sample mean, sample standard deviation, skewness, and kurtosis of energy consumption values ​​and feature matching degree are calculated. The sample mean and sample standard deviation are used to characterize the central tendency and dispersion of the data, while skewness and kurtosis are used to assess the degree of deviation of the data distribution from the normal distribution.

[0029] Based on sample statistics, a parametric estimation method is used to determine the confidence level. When the historical energy consumption data distribution conforms to the normal distribution assumption, the quantile function of the normal distribution is used to calculate the confidence interval. When the historical energy consumption data distribution deviates from the normal distribution, a resampling method is used for nonparametric estimation. Sample sets are generated through multiple samplings with replacement, and the statistics of each sample set are calculated. The confidence interval is determined based on the distribution of the statistics. When the historical energy consumption data has obvious periodic or seasonal characteristics, a time series analysis method is used to remove trend and seasonal components, and the above method is applied to the residual series to calculate the confidence interval.

[0030] The confidence intervals for each time period are weighted and fused. The weights are dynamically determined based on the representativeness of the time period, the size of the data, the similarity of the current working conditions, and the data quality assessment results to generate fluctuating confidence intervals.

[0031] By combining the current equipment operating environment parameters and peak and off-peak electricity consumption periods, the initial range of the energy consumption threshold is dynamically adjusted to generate an adaptive threshold range that includes the threshold ranges of power parameters, water supply parameters, and environmental parameters.

[0032] According to the above scheme, the device-side storage adopts a circular caching mechanism, storing the original energy consumption data, geographic coordinate data, timestamp data, device-level feature code and digital fingerprint in the order of data collection time; when the cache space reaches the preset capacity limit, it automatically overwrites the earliest collected data;

[0033] The edge storage establishes a hotspot event database and a statistical feature database. The hotspot event database stores hotspot event records, and the statistical feature database stores statistical values ​​of electricity data, water supply data, and environmental data aggregated according to a preset time granularity.

[0034] The edge node executes a data value assessment algorithm, which calculates a data value score based on data access frequency, information entropy value, and timeliness coefficient using a weighted summation formula, and scores and sorts hot event records and statistical feature data; the preset threshold is the industry benchmark value;

[0035] Based on the data value scoring results, hot event records and statistical feature data with scores higher than the preset threshold are synchronized to cloud storage, while data with scores lower than the preset threshold are stored at the edge.

[0036] Cloud storage constructs a cross-regional knowledge base, which integrates hot event records and statistical feature data from different edge nodes, and stores them in association with external electricity price data, meteorological data, and equipment ledger data;

[0037] Based on the associated data in the cross-regional knowledge base, the cloud-based system trains a global energy consumption analysis model, which includes a power load prediction model, a water supply anomaly detection model, and an environmental correlation analysis model.

[0038] According to the above scheme, the edge node locates the neighboring edge node within a preset radius using the GeoHash encoded data in the hot event record through a prefix matching algorithm, and obtains the real-time operating status data of the devices connected to the neighboring edge node. The real-time operating status data includes the device's geographical location, real-time energy consumption value, digital fingerprint, and abnormal marking status.

[0039] The real-time power data, water supply data, and environmental data collected from the geographically adjacent edge nodes are aggregated to form a regional performance consumption status dataset. Based on the matching results between the regional performance consumption status dataset and the policy rule base, local optimization instructions are immediately generated. The local optimization instructions include power control parameters, water supply regulation parameters, and environmental optimization parameters.

[0040] The cloud receives hot event records, statistical feature data and local optimization instruction execution feedback data uploaded by each edge node, and combines them with external parameters and historical energy consumption data in the cross-regional knowledge base to generate a global optimization strategy package based on time and space dimensions.

[0041] A virtual energy consumption system is constructed using digital twin technology. The generated optimization strategy package is then simulated and verified in multiple scenarios. The simulation and verification include the predicted energy consumption value after strategy execution, the device load balancing degree, and the occurrence rate of abnormal events. When the energy consumption prediction deviation value of the simulation and verification exceeds the allowable range, a strategy regeneration mechanism is triggered, and the feature data of the failed scenario is recorded for optimization of the digital twin model. After successful verification, the optimization strategy package is distributed to the target edge node and device according to the GeoHash encoding of the target device.

[0042] According to the above scheme, the policy rule base includes:

[0043] A multi-dimensional rule index structure is constructed based on energy consumption data and corresponding digital fingerprints.

[0044] A preset set of basic rules is provided, which includes power load balancing rules, water supply pressure regulation rules, and environmental parameter optimization rules.

[0045] Based on historical execution effect feedback, the state is the regional energy consumption status dataset, the action is the adjustment of rule parameters, and the reward is the energy consumption optimization rate after strategy execution. The rule parameter weights are adjusted through reinforcement learning algorithm to generate a strategy rule base.

[0046] When a new abnormal pattern or device type is detected, a corresponding control rule template is automatically generated. The system also periodically integrates global optimization strategies distributed from the cloud, synchronously updates the rule index structure and parameter weights, and updates the strategy rule library. Based on strategy execution effect scoring data, environmental change sensitivity data, and device adaptability data, the system dynamically updates the rule weights using a gradient descent algorithm.

[0047] According to the above scheme, during the strategy execution process, the edge node collects energy consumption data and strategy execution effect data from the device in real time and uploads them to the cloud. The strategy execution effect data includes energy consumption change rate data, device response time data, and anomaly recovery data.

[0048] Based on a global energy consumption analysis model, the cloud performs real-time analysis on energy consumption data and strategy execution effect data uploaded by each edge node, constructs a multi-dimensional evaluation index system, and quantitatively evaluates the strategy execution effect by comparing the changes in indicators before and after strategy execution.

[0049] When the resource utilization data of any edge node exceeds the preset warning threshold, the task migration process based on geographical location priority is automatically initiated.

[0050] The system monitors the signal strength, packet loss rate, and bit error rate of the data transmission link in real time. When the channel quality is lower than the preset threshold, it activates the multi-mode switching mechanism, selects the optimal transmission mode based on the real-time performance parameters of the available communication methods, dynamically segments the transmitted data, and adjusts the segment size and redundancy according to the channel quality.

[0051] Based on the evaluation results of the strategy execution effect, the cloud incrementally updates the global energy consumption analysis model and adjusts the parameter weights in the strategy rule base.

[0052] According to the above scheme, the task migration process includes:

[0053] Based on GeoHash encoded data and network topology connection data of edge nodes, a set of k nearest neighbors centered on the target node is constructed. Based on the available memory space, remaining storage capacity, network latency, and historical task success rate data of candidate nodes, a comprehensive resource score is calculated. Non-core computing tasks of the target node are migrated to candidate nodes in descending order of comprehensive resource score. During the migration process, the task execution progress and resource usage changes are monitored in real time. When the deviation of CPU or memory utilization exceeds 20%, a secondary migration is triggered.

[0054] An energy consumption data management system based on edge computing, the system includes: a data processing module, an edge analysis module, a storage management module, an optimization control module, and a resource scheduling module;

[0055] The data processing module includes a data acquisition module, a feature encoding module, and a data compression module. The data acquisition module is used to acquire power data, water supply data, and environmental data. The feature encoding module is used to bind geographic coordinates, synchronize timestamps, extract device features, and generate digital fingerprints. The data compression module uses a differential encoding compression algorithm to reduce data transmission volume.

[0056] The edge analysis module includes a real-time monitoring module, an anomaly detection module, and a local optimization module. The real-time monitoring module compares the current data with the historical digital fingerprint database to perform multi-dimensional feature matching. The anomaly detection module uses dynamic thresholds and feature matching degrees to mark high-confidence anomalies and generate hotspot event records. The local optimization module generates local optimization instructions based on a preset strategy rule base.

[0057] The storage management module includes a device-side storage module, an edge-side storage module, and a cloud-side storage module. The device-side storage module uses a circular caching mechanism to temporarily store raw data. The edge-side storage module manages the hot event database and the statistical feature database to complete data value assessment. The cloud-side storage module builds a cross-regional knowledge base and integrates multi-dimensional data for global analysis.

[0058] The optimization control module includes a region aggregation module, a strategy generation module, and a strategy verification module. The region aggregation module aggregates the states of neighboring nodes based on GeoHash encoding. The strategy generation module generates an optimization strategy package by integrating cross-regional data and external parameters. The strategy verification module uses digital twins to perform multi-scenario simulation verification.

[0059] The resource scheduling module includes a status monitoring module, a task migration module, and a transmission optimization module; the status monitoring module collects node resource utilization and channel quality indicators in real time; the task migration module dynamically schedules computing tasks based on geographical location priority; and the transmission optimization module dynamically switches data transmission modes according to channel quality.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] 1. This invention is based on multi-dimensional feature matching and adaptive threshold mechanism, combined with statistical analysis and time series modeling, to achieve high-confidence detection of energy consumption anomalies, reduce false alarm rate, and provide early warning of potential energy waste or equipment failure.

[0062] 2. This invention adopts a three-level storage strategy, dynamically allocating storage resources according to data value. This ensures both the short-term traceability of original data and long-term cloud storage and cross-regional correlation analysis, thereby achieving hierarchical data management and value transfer, and reducing cloud storage pressure.

[0063] 3. This invention uses GeoHash encoding prefix matching to quickly locate geographically adjacent edge nodes, aggregates the operating status of regional devices in real time, and generates local optimization instructions based on the policy rule base, thereby improving energy efficiency and device operating stability. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating the steps of an energy consumption data management method based on edge computing according to the present invention.

[0065] Figure 2 This is a schematic diagram of the structure of an energy consumption data management system based on edge computing according to the present invention. Detailed Implementation

[0066] 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.

[0067] Example: Figures 1-2 As shown, the present invention provides a technical solution, a method for managing energy consumption data based on edge computing, the method comprising the following steps:

[0068] S1. Collect energy consumption data using edge devices, bind geographical coordinates and timestamps to each data point, overlay device-level feature codes to generate a unique digital fingerprint, and compress the original data through differential encoding.

[0069] Specifically, energy consumption data includes electricity data, water supply data, and environmental data. Electricity data includes voltage, current, active power, and reactive power data; water supply data includes water flow rate and water pressure data; and environmental data includes temperature and humidity data. For example, edge devices are deployed in each branch, including smart meters, smart water meters, and environmental sensors. Data collected at branch A includes: electricity data: voltage 228V, current 35A, active power 8kW; water supply data: flow rate 2.5L / s, water pressure 0.25MPa; and environmental data: temperature 26℃, humidity 65%.

[0070] Furthermore, the precise geographic location of the edge data is obtained, and the longitude and latitude data are converted into GeoHash codes. Real-time synchronization with the network time protocol is performed to add a timestamp to each energy consumption data point. Voltage and current data are used to analyze waveform variation characteristics. By calculating voltage fluctuation frequency characteristics, current harmonic component characteristics, active power change rate characteristics, and reactive power change rate characteristics, the characteristics of the power data are analyzed. Based on the gradient of water flow velocity data and the fluctuation amplitude of water pressure data, combined with the correlation between water flow velocity and water pressure, the characteristics of the water supply data are analyzed. The rate of temperature rise and fall and the range of humidity changes are used to further analyze... Analyze environmental data characteristics; normalize the characteristics of power data, water supply data, and environmental data, and integrate them to generate equipment-level feature codes; generate fixed-length digital fingerprints by combining geographic coordinate data, timestamp data, and equipment-level feature codes using a cryptographic hash algorithm; for example, convert the longitude and latitude data of the geographical location of branch A into GeoHash code X, add the timestamp 2025-08-2010:00:00, and the equipment-level feature code is a current harmonic distortion rate of 7.2% and a power factor of 0.92; generate fixed-length digital fingerprints by combining geographic coordinate data, timestamp data, and equipment-level feature codes using a cryptographic hash algorithm.

[0071] Furthermore, the difference between the original data of adjacent time series is encoded using a difference coding compression algorithm, thereby reducing the amount of data transmitted and the storage footprint.

[0072] S2. Edge nodes perform multi-dimensional feature matching between digital fingerprints and historical digital fingerprint databases. When the energy consumption value deviates from the adaptive threshold and the feature matching degree is lower than the feature matching degree threshold, a high confidence anomaly marking mechanism is automatically triggered, and hot event records are generated.

[0073] Specifically, the current digital fingerprint is matched against historical digital fingerprints with the same geographical coordinates stored in the historical digital fingerprint database. The matching comparison includes periodic features based on time series analysis, geographical relevance features based on GeoHash encoding, and operational status features based on feature parameters. The time dimension uses a dynamic time warping algorithm to calculate periodic similarity, the spatial dimension uses GeoHash prefix matching to assess geographical relevance, and the device dimension uses Euclidean distance based on feature parameters to calculate operational status differences. An adaptive threshold range is dynamically generated based on historical data, including an energy consumption threshold, a feature matching threshold, and a fluctuation confidence interval. When energy consumption is detected... When data deviates from the adaptive threshold range and the feature matching degree is lower than the feature matching degree threshold, a high-confidence anomaly marking mechanism is automatically triggered. The anomaly marking mechanism includes spatiotemporal location marking, deviation degree quantification, feature difference description, and anomaly type classification. Hotspot event records containing the anomaly marking mechanism are generated, and a unique identifier code is assigned to each hotspot event record. For example, comparing the power curve of branch A from 14:00 to 15:00 today with historical data for the same period, the dynamic time warping algorithm found that the similarity was only 65% ​​(threshold ≥ 80%), which was judged as an abnormal fluctuation. The air conditioning load of 5 branches within a 3-kilometer radius all increased by 10% during the same period, but this branch suddenly increased by 30%, and the GeoHash prefix matching degree showed a high geographical correlation. When the real-time power reaches 11kW and the matching degree is 62%, the anomaly marking is triggered, and a hotspot event record is generated: Spatiotemporal location: Branch A, Chaoyang District, Beijing, GeoHash:X, 2025-08-20 10:00:00; Deviation degree: Power exceeds the threshold by 22%; Feature difference: Current harmonics increase by 9%, temperature rises by 4℃.

[0074] Furthermore, historical energy consumption data and corresponding historical digital fingerprints with the same device identifiers are extracted from the historical digital fingerprint database. The historical energy consumption data covers different time periods and operating conditions of the equipment's normal operation. For example, the air conditioning operation data of branch A over the past year is retrieved, covering scenarios such as high temperatures in summer, low temperatures in winter, and high customer traffic during holidays. Based on the historical energy consumption data, the mean, variance, and extreme values ​​of energy consumption values ​​in the time dimension are calculated to construct a basic statistical model of energy consumption values ​​and determine the initial range of energy consumption value thresholds. For example, the average power of air conditioning in summer is 8kW, with a variance of 0.5, so the threshold is set to 7-9kW, and the feature matching degree threshold is set to 75%. The periodicity characteristics of historical digital fingerprints in the time dimension, the geographical correlation characteristics in the spatial dimension, and the operating status characteristics in the device dimension are analyzed to calculate the matching similarity distribution of features in each dimension. Based on the distribution, the basic value of the feature matching degree threshold is set. Based on the fluctuation of energy consumption values ​​and feature matching degree in historical energy consumption data, statistical analysis methods are used to determine the fluctuation confidence interval.

[0075] Furthermore, the statistical analysis methods include: segmenting historical energy consumption data according to time series, with segmentation based on equipment operating cycles, nodes of change in operating conditions, cycles of change in external environmental influencing factors, and the similarity of data characteristics; calculating the sample mean, sample standard deviation, skewness, and kurtosis of energy consumption values ​​and feature matching degree for each segment; the sample mean and sample standard deviation are used to characterize the central tendency and dispersion of the data, while skewness and kurtosis are used to assess the degree of deviation of the data distribution from the normal distribution; and determining the confidence level based on the sample statistics using parameter estimation methods. If the historical energy consumption data distribution conforms to the normal distribution assumption, the confidence level is determined using the normal distribution assumption. Confidence intervals are calculated using quantile functions. When the distribution of historical energy consumption data deviates from a normal distribution, a resampling method is used for nonparametric estimation. Sample sets are generated through multiple samplings with replacement, and the statistics of each sample set are calculated. Confidence intervals are determined based on the distribution of the statistics. When historical energy consumption data has obvious periodic or seasonal characteristics, time series analysis is used to remove trend and seasonal components. The above method is then applied to the residual series to calculate confidence intervals. The confidence intervals for each time period are weighted and fused. The weights are dynamically determined based on the representativeness of the time period, the size of the data, the similarity of the current operating conditions, and the data quality assessment results to generate fluctuation confidence intervals.

[0076] Furthermore, by combining the current equipment operating environment parameters and peak and off-peak electricity consumption periods, the initial range of energy consumption threshold values ​​is dynamically adjusted to generate an adaptive threshold range that includes the threshold ranges of power parameters, water supply parameters, and environmental parameters.

[0077] S3. The collected data is managed according to a three-level storage strategy, which includes device-side storage, edge storage, and cloud storage. Device-side storage uses a circular caching mechanism to temporarily cache the raw data. Edge storage stores hot event details and aggregated statistical indicators, and selects high-value data to the cloud based on the data value assessment results. Cloud storage permanently stores cross-regional correlation information and integrates external parameters to build a global analysis model.

[0078] Specifically, the device-side storage adopts a circular caching mechanism, storing raw energy consumption data, geographic coordinate data, timestamp data, device-level feature code, and digital fingerprint in the order of data collection time. When the cache space reaches the preset capacity limit, it automatically overwrites the earliest collected data. For example, the raw data of the most recent 2 hours is cached in a circular manner to quickly trace back the raw waveforms at abnormal moments.

[0079] Furthermore, edge storage establishes a hotspot event database and a statistical feature database. The hotspot event database stores hotspot event records, while the statistical feature database stores statistical values ​​of electricity, water supply, and environmental data aggregated according to a preset time granularity. Edge nodes execute a data value assessment algorithm. This algorithm calculates a data value score based on data access frequency, information entropy value, and timeliness coefficient using a weighted summation formula. The algorithm then scores and ranks the hotspot event records and statistical feature data. For example: Access frequency: Headquarters queries branch anomaly records daily, weight 40%; Timeliness: Real-time anomaly data has a timeliness coefficient of 0.9, weight 30%; Information entropy: This anomaly pattern is newly discovered, with a high entropy value, weight 30%; A score of 0.85, higher than the preset threshold of 0.7, is synchronized to the cloud. The preset threshold is an industry benchmark value. Based on the data value score results, hotspot event records and statistical feature data with scores higher than the preset threshold are synchronized to cloud storage, while data with scores lower than the preset threshold are retained in edge storage.

[0080] Furthermore, cloud storage constructs a cross-regional knowledge base, which integrates hot event records and statistical feature data from different edge nodes and stores them in association with external electricity price data, meteorological data, and equipment ledger data. Based on the associated data in the cross-regional knowledge base, the cloud trains a global energy consumption analysis model, which includes a power load prediction model, a water supply anomaly detection model, and an environmental correlation analysis model.

[0081] S4. Based on spatiotemporal coding, quickly locate geographically adjacent edge nodes, aggregate the operating status of geographically adjacent devices, generate local optimization instructions through a preset policy rule base, generate optimization policy packages by integrating multi-dimensional information in the cloud, and send them to the target device after virtual verification.

[0082] Specifically, edge nodes use GeoHash encoded data from hotspot event records to locate neighboring edge nodes within a preset radius using a prefix matching algorithm. They also acquire real-time operational status data of devices connected to these neighboring edge nodes, including device geographic location, real-time energy consumption values, digital fingerprints, and anomaly markers. Real-time power, water, and environmental data collected from geographically neighboring edge nodes are aggregated to form a regional performance consumption status dataset. For example, locating 10 branches within a 5km radius of branch A, it is found that 8 of them have normal air conditioning loads, ruling out regional faults and determining the issue to be a single-store equipment problem. Based on the matching results between the regional performance consumption status dataset and the policy rule base, local optimization instructions are immediately generated. These instructions include power control parameters, water supply adjustment parameters, and environmental optimization parameters. For example, the basic rule base triggers the device overload protection rule, automatically increasing the set temperature of branch A's air conditioner from 18℃ to 24℃, reducing the compressor load. The generated local instructions are then sent to the air conditioning controller. The system adjusts the inverter parameters, reducing the power to 8.5kW (this is just an example and not a limitation). The cloud receives hotspot event records, statistical feature data, and local optimization command execution feedback data uploaded by each edge node. Combining these with external parameters and historical energy consumption data from a cross-regional knowledge base, it generates a global optimization strategy package based on time and space dimensions. A virtual energy consumption system is built using digital twin technology to simulate and verify the generated optimization strategy package in multiple scenarios. The simulation verification includes the predicted energy consumption value after strategy execution, equipment load balancing, and abnormal event occurrence rate. When the energy consumption prediction deviation exceeds the allowable range, a strategy regeneration mechanism is triggered, and the failure scenario feature data is recorded for digital twin model optimization. After successful verification, the optimization strategy package is distributed to the target edge node and device according to the GeoHash encoding of the target device. For example, by simulating the impact of raising the temperature on customer experience using digital twin, a predicted decrease in customer traffic of ≤3% is generated. After successful verification, the command is issued.

[0083] Furthermore, a multi-dimensional rule index structure is constructed based on energy consumption data and corresponding digital fingerprints; a basic rule set is preset, including power load balancing rules, water supply pressure regulation rules, and environmental parameter optimization rules; based on historical execution effect feedback, the status is the regional energy consumption status dataset, the action is the rule parameter adjustment, and the reward is the energy consumption optimization rate after strategy execution; the rule parameter weights are adjusted through reinforcement learning algorithms to generate a strategy rule library; when a new abnormal pattern or equipment type is detected, a corresponding control rule template is automatically generated, and the global optimization strategy issued by the cloud is periodically integrated to synchronously update the rule index structure and parameter weights, and update the strategy rule library; based on strategy execution effect scoring data, environmental change sensitivity data, and equipment adaptability data, the rule weights are dynamically updated through gradient descent algorithm.

[0084] S5. During the strategy operation, monitor the running status of each node in real time and feed back the strategy execution effect to the cloud; when the resource utilization data of any edge node exceeds the preset warning threshold, automatically start the task migration process based on geographical location priority and dynamically switch the data transmission mode according to the real-time channel quality.

[0085] Specifically, during strategy execution, edge nodes collect energy consumption data and strategy execution effect data from the devices in real time and upload them to the cloud. The strategy execution effect data includes energy consumption change rate data, device response time data, and anomaly recovery data. Based on a global energy consumption analysis model, the cloud performs real-time analysis on the energy consumption data and strategy execution effect data uploaded by each edge node, constructs a multi-dimensional evaluation index system, and quantitatively evaluates the strategy execution effect by comparing the changes in the indexes before and after strategy execution.

[0086] Furthermore, when the resource utilization data of any edge node exceeds a preset warning threshold, a task migration process based on geographic location priority is automatically initiated. The signal strength, packet loss rate, and bit error rate of the data transmission link are monitored in real time. When the channel quality is lower than a preset threshold, a multi-mode switching mechanism is activated. Based on the real-time performance parameters of available communication methods, the optimal transmission mode is selected, and the transmitted data is dynamically fragmented. The fragment size and redundancy are adjusted according to the channel quality. For example, if the edge node resource utilization of branch A is 90%, a task migration is initiated: three neighboring nodes are selected, with memory ≥ 2GB, latency ≤ 10ms, historical success rate ≥ 95%, and scores of 0.85, 0.78, and 0.72 respectively. The non-real-time data aggregation task is migrated to the node with the highest score, reducing resource utilization to 70%.

[0087] Furthermore, based on the GeoHash encoded data and network topology connection data of the edge nodes, a set of k nearest neighbors centered on the target node is constructed. Based on the available memory space data, remaining storage capacity data, network latency data, and historical task success rate data of the candidate nodes, a comprehensive resource score is calculated. Non-core computing tasks of the target node are migrated to the candidate nodes in descending order of comprehensive resource score. During the migration process, the task execution progress and resource usage changes are monitored in real time. When the deviation of CPU or memory utilization exceeds 20%, a secondary migration is triggered.

[0088] This invention provides another technical solution: an energy consumption data management system based on edge computing, which includes: a data processing module, an edge analysis module, a storage management module, an optimization control module, and a resource scheduling module.

[0089] The data processing module includes a data acquisition module, a feature encoding module, and a data compression module. The data acquisition module is used to collect power data, water supply data, and environmental data. The feature encoding module is used to bind geographic coordinates, synchronize timestamps, extract equipment features, and generate digital fingerprints. The data compression module uses a differential encoding compression algorithm to reduce data transmission volume.

[0090] The edge analysis module includes a real-time monitoring module, an anomaly detection module, and a local optimization module. The real-time monitoring module compares the current data with the historical digital fingerprint database to perform multi-dimensional feature matching. The anomaly detection module uses dynamic thresholds and feature matching degrees to mark high-confidence anomalies and generate hotspot event records. The local optimization module generates local optimization instructions based on a preset strategy rule base.

[0091] The storage management module includes a device-side storage module, an edge-side storage module, and a cloud-side storage module. The device-side storage module uses a circular caching mechanism to temporarily store raw data. The edge-side storage module manages the hot event database and the statistical feature database to complete data value assessment. The cloud-side storage module builds a cross-regional knowledge base and integrates multi-dimensional data for global analysis.

[0092] The optimization control module includes a regional aggregation module, a policy generation module, and a policy verification module. The regional aggregation module aggregates the states of neighboring nodes based on GeoHash encoding. The policy generation module generates optimization policy packages by integrating cross-regional data and external parameters. The policy verification module uses digital twins to perform multi-scenario simulation verification.

[0093] The resource scheduling module includes a status monitoring module, a task migration module, and a transmission optimization module. The status monitoring module collects node resource utilization and channel quality indicators in real time. The task migration module dynamically schedules computing tasks based on geographical location priority. The transmission optimization module dynamically switches data transmission modes according to channel quality.

[0094] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for managing energy consumption data based on edge computing, characterized in that: The method includes: Energy consumption data is collected using edge devices. Each data point is bound to geographic coordinates and timestamps, and device-level feature codes are overlaid to generate a unique digital fingerprint. The original data is then compressed using differential encoding. The edge node performs multi-dimensional feature matching between the digital fingerprint and the historical digital fingerprint database. When the energy consumption value deviates from the adaptive threshold and the feature matching degree is lower than the feature matching degree threshold, the high confidence anomaly marking mechanism is automatically triggered and hot event records are generated. The collected data is managed according to a three-tier storage strategy, which includes device-side storage, edge storage, and cloud storage. The device-side storage uses a circular caching mechanism to temporarily cache the original data. The edge storage stores details of hot events and aggregated statistical indicators, and selects high-value data to the cloud based on the data value assessment results. The cloud storage permanently stores cross-regional correlation information and integrates external parameters to build a global analysis model. Based on spatiotemporal coding, the edge nodes of geographical proximity are quickly located, the operating status of geographically nearby devices is aggregated, local optimization instructions are generated through a preset policy rule base, and the cloud integrates multi-dimensional information to generate an optimization policy package, which is then sent to the target device after virtual verification. During the strategy operation, the running status of each node is monitored in real time, and the strategy execution effect is fed back to the cloud. When the resource utilization data of any edge node exceeds the preset warning threshold, the task migration process based on geographical location priority is automatically started, and the data transmission mode is dynamically switched according to the real-time channel quality.

2. The energy consumption data management method based on edge computing according to claim 1, characterized in that: The energy consumption data includes electricity data, water supply data, and environmental data; the electricity data includes voltage data, current data, active power data, and reactive power data; the water supply data includes water flow rate data and water pressure data; and the environmental data includes temperature data and humidity data. The system obtains the precise geographic location of edge data, converts the longitude and latitude data of the geographic location into GeoHash encoding, and adds a timestamp to each energy consumption data by real-time synchronization with the network time protocol. By analyzing waveform variation characteristics using voltage and current data, and by calculating voltage fluctuation frequency characteristics, current harmonic component characteristics, active power change rate characteristics, and reactive power change rate characteristics, the characteristics of power data are analyzed. Based on the gradient of water flow velocity data and the fluctuation range of water pressure data, combined with the correlation between water flow velocity and water pressure, the characteristics of water supply data are analyzed; the characteristics of environmental data are analyzed by utilizing the rate of temperature rise and fall and the range of humidity variation. Electricity data features, water supply data features, and environmental data features are normalized and integrated to generate equipment-level feature codes; geographic coordinate data, timestamp data, and equipment-level feature codes are then combined using a cryptographic hash algorithm to generate a fixed-length digital fingerprint.

3. The energy consumption data management method based on edge computing according to claim 1, characterized in that: The current digital fingerprint is matched and compared with historical digital fingerprints with the same geographical coordinates stored in the historical digital fingerprint database. The matching and comparison includes the periodic features of the time dimension based on time series analysis, the geographical relevance features of the spatial dimension based on GeoHash encoding, and the operating status features of the device dimension based on feature parameters. An adaptive threshold range is dynamically generated based on historical data. The adaptive threshold range includes an energy consumption threshold, a feature matching degree threshold, and a fluctuation confidence interval. When energy consumption data is detected to deviate from the adaptive threshold range and the feature matching degree is lower than the feature matching degree threshold, a high confidence anomaly marking mechanism is automatically triggered. The anomaly marking mechanism includes spatiotemporal location marking, deviation degree quantification, feature difference description and anomaly type classification. Generate hotspot event records that include an anomaly marking mechanism, and assign a unique identifier code to each hotspot event record.

4. The energy consumption data management method based on edge computing according to claim 3, characterized in that: The dynamically generated adaptive threshold range includes: Extract historical energy consumption data and corresponding historical digital fingerprints of the same device identifier from the historical digital fingerprint database. The historical energy consumption data covers different time periods and operating conditions of the device during normal operation. Based on the historical energy consumption data, the mean, variance, and extreme values ​​of energy consumption values ​​over time are calculated, a basic statistical model of energy consumption values ​​is constructed, and the initial range of energy consumption value thresholds is determined. The periodicity of the historical digital fingerprint in the time dimension, the geographical relevance in the spatial dimension, and the operating status in the device dimension are analyzed. The matching similarity distribution of each dimension is calculated, and the basic value of the feature matching degree threshold is set according to the distribution. Based on the fluctuations in the matching degree between energy consumption values ​​and characteristics in historical energy consumption data, statistical analysis methods are used to determine the confidence interval of fluctuations. The statistical analysis methods include: Historical energy consumption data is segmented according to time series, and the segmentation is based on equipment operating cycle, operating condition change nodes, change cycle of external environmental influencing factors, and similarity of data characteristics. Based on each data segment, the sample mean, sample standard deviation, skewness, and kurtosis of energy consumption values ​​and feature matching degree are calculated. The sample mean and sample standard deviation are used to characterize the central tendency and dispersion of the data, while skewness and kurtosis are used to assess the degree of deviation of the data distribution from the normal distribution. Based on sample statistics, a parametric estimation method is used to determine the confidence level. When the historical energy consumption data distribution conforms to the normal distribution assumption, the quantile function of the normal distribution is used to calculate the confidence interval. When the historical energy consumption data distribution deviates from the normal distribution, a resampling method is used for nonparametric estimation. Sample sets are generated through multiple samplings with replacement, and the statistics of each sample set are calculated. The confidence interval is determined based on the distribution of the statistics. When the historical energy consumption data has obvious periodic or seasonal characteristics, a time series analysis method is used. After removing trend and seasonal components, a resampling method is used to perform nonparametric estimation on the residual series to calculate the confidence interval. The confidence intervals for each time period are weighted and fused. The weights in the weighting and fusion are dynamically determined based on the representativeness of the time period, the size of the data volume, the similarity of the current working conditions, and the data quality assessment results, to generate fluctuating confidence intervals. By combining the current equipment operating environment parameters and peak and off-peak electricity consumption periods, the initial range of the energy consumption threshold is dynamically adjusted to generate an adaptive threshold range that includes the threshold ranges of power parameters, water supply parameters, and environmental parameters.

5. The energy consumption data management method based on edge computing according to claim 1, characterized in that: The device-side storage adopts a circular caching mechanism, storing raw energy consumption data, geographic coordinate data, timestamp data, device-level feature code, and digital fingerprint in the order of data collection time; when the cache space reaches the preset capacity limit, it automatically overwrites the earliest collected data; The edge storage establishes a hotspot event database and a statistical feature database. The hotspot event database stores hotspot event records, and the statistical feature database stores statistical values ​​of electricity data, water supply data, and environmental data aggregated according to a preset time granularity. The edge node executes a data value assessment algorithm, which calculates a data value score based on data access frequency data, information entropy value, and timeliness coefficient, and scores and sorts hot event records and statistical feature data. Based on the data value scoring results, hot event records and statistical feature data with scores higher than the preset threshold are synchronized to cloud storage, while data with scores lower than the preset threshold are stored at the edge. Cloud storage constructs a cross-regional knowledge base, which integrates hot event records and statistical feature data from different edge nodes, and stores them in association with external electricity price data, meteorological data, and equipment ledger data; Based on the associated data in the cross-regional knowledge base, the cloud-based system trains a global energy consumption analysis model, which includes a power load prediction model, a water supply anomaly detection model, and an environmental correlation analysis model.

6. The energy consumption data management method based on edge computing according to claim 1, characterized in that: The edge node locates neighboring edge nodes within a preset radius using a prefix matching algorithm based on the GeoHash encoded data in the hot event record, and obtains the real-time operating status data of the devices connected to the neighboring edge nodes. The real-time operating status data includes the device's geographical location, real-time energy consumption value, digital fingerprint, and abnormal marker status. The real-time power data, water supply data, and environmental data collected from the neighboring edge nodes are aggregated to form a regional performance consumption status dataset. Based on the matching results between the regional performance consumption status dataset and the policy rule base, local optimization instructions are immediately generated. The local optimization instructions include power control parameters, water supply regulation parameters, and environmental optimization parameters. The cloud receives hot event records, statistical feature data and local optimization instruction execution feedback data uploaded by each edge node, and combines them with external parameters and historical energy consumption data in the cross-regional knowledge base to generate a global optimization strategy package based on time and space dimensions. A virtual energy consumption system is constructed using digital twin technology. The generated optimization strategy package is then simulated and verified in multiple scenarios. The simulation and verification include the predicted energy consumption after strategy execution, the device load balancing degree, and the occurrence rate of abnormal events. After successful verification, the optimization strategy package is distributed to the target edge node and device according to the GeoHash encoding of the target device.

7. The energy consumption data management method based on edge computing according to claim 6, characterized in that: The policy rule base includes: A multi-dimensional rule index structure is constructed based on energy consumption data and corresponding digital fingerprints. A preset set of basic rules is provided, which includes power load balancing rules, water supply pressure regulation rules, and environmental parameter optimization rules. Based on historical performance feedback, the rule parameter weights are adjusted through reinforcement learning algorithms to generate a policy rule base. When a new abnormal pattern or device type is detected, a corresponding control rule template is automatically generated, and the global optimization strategy issued by the cloud is periodically integrated to synchronously update the rule index structure and parameter weights, and update the strategy rule library.

8. The energy consumption data management method based on edge computing according to claim 1, characterized in that: During strategy execution, edge nodes collect energy consumption data and strategy execution effect data from the device in real time and upload them to the cloud. The strategy execution effect data includes energy consumption change rate data, device response time data, and anomaly recovery data. Based on a global energy consumption analysis model, the cloud performs real-time analysis on energy consumption data and strategy execution effect data uploaded by each edge node, constructs a multi-dimensional evaluation index system, and quantitatively evaluates the strategy execution effect by comparing the changes in indicators before and after strategy execution. When the resource utilization data of any edge node exceeds the preset warning threshold, the task migration process based on geographical location priority is automatically initiated. The system monitors the signal strength, packet loss rate, and bit error rate of the data transmission link in real time. When the channel quality is lower than the preset threshold, it activates the multi-mode switching mechanism, selects the optimal transmission mode based on the real-time performance parameters of the available communication methods, dynamically segments the transmitted data, and adjusts the segment size and redundancy according to the channel quality. Based on the evaluation results of the strategy execution effect, the cloud incrementally updates the global energy consumption analysis model and adjusts the parameter weights in the strategy rule base.

9. The energy consumption data management method based on edge computing according to claim 8, characterized in that: The task migration process includes: Based on GeoHash encoded data and network topology connection data of edge nodes, a set of k nearest neighbors centered on the target node is constructed. Based on the available memory space, remaining storage capacity, network latency, and historical task success rate data of candidate nodes, a comprehensive resource score is calculated. Non-core computing tasks of the target node are migrated to candidate nodes in descending order of comprehensive resource score. During the migration process, task execution progress data and resource usage changes are monitored in real time. When node load imbalance data is detected, a secondary migration process is triggered.

10. An energy consumption data management system based on edge computing, characterized in that: The system includes: a data processing module, an edge analysis module, a storage management module, an optimization control module, and a resource scheduling module; The data processing module includes a data acquisition module, a feature encoding module, and a data compression module. The data acquisition module is used to acquire power data, water supply data, and environmental data. The feature encoding module is used to bind geographic coordinates, synchronize timestamps, extract device features, and generate digital fingerprints. The data compression module uses a differential encoding compression algorithm to reduce data transmission volume. The edge analysis module includes a real-time monitoring module, an anomaly detection module, and a local optimization module. The real-time monitoring module compares the current data with the historical digital fingerprint database to perform multi-dimensional feature matching. The anomaly detection module uses dynamic thresholds and feature matching degrees to mark high-confidence anomalies and generate hotspot event records. The local optimization module generates local optimization instructions based on a preset strategy rule base. The storage management module includes a device-side storage module, an edge-side storage module, and a cloud-side storage module. The device-side storage module uses a circular caching mechanism to temporarily store raw data. The edge-side storage module manages the hot event database and the statistical feature database to complete data value assessment. The cloud-side storage module builds a cross-regional knowledge base and integrates multi-dimensional data for global analysis. The optimization control module includes a region aggregation module, a strategy generation module, and a strategy verification module. The region aggregation module aggregates the states of neighboring nodes based on GeoHash encoding. The strategy generation module generates an optimization strategy package by integrating cross-regional data and external parameters. The strategy verification module uses digital twins to perform multi-scenario simulation verification. The resource scheduling module includes a status monitoring module, a task migration module, and a transmission optimization module; the status monitoring module collects node resource utilization and channel quality indicators in real time; the task migration module dynamically schedules computing tasks based on geographical location priority; and the transmission optimization module dynamically switches data transmission modes according to channel quality.

Images