An energy-saving method and system based on power big data

Abstract

This invention relates to an energy-saving method and system based on big data in the power industry, belonging to the field of energy-saving optimization data processing technology. The method includes: accessing data from smart meters and external sources; constructing a data model based on spatiotemporal fusion of multi-source data in the cloud; establishing a dynamic benchmark for the operating scenario based on the data model to group nodes; then, based on the deviation of group evaluation indicators, obtaining a multi-dimensional comprehensive ranking and generating a list of energy-saving renovation nodes; and completing energy-saving optimization through bidirectional parsing of the node list, including upward aggregation of nodes to generate scheduling strategies and downward decomposition of load to identify equipment to optimize user energy consumption. This invention constructs a unified data model through real-time acquisition and encrypted transmission of smart meters and multi-source data; utilizes machine learning clustering to optimize the operating benchmark; performs node grouping and feature extraction; calculates node deviations in real time and generates an energy-saving potential list; and achieves node resource scheduling and optimization of high-energy-consuming equipment through bidirectional parsing.

Description

Technical Field

[0001] This invention belongs to the field of energy-saving optimization data processing technology, specifically relating to an energy-saving method and system based on power big data. Background Technology

[0002] Energy conservation and emission reduction in power systems have become a core issue concerning both economic efficiency and sustainable environmental development. The widespread adoption of smart grids and the maturity of Internet of Things (IoT) technology have led to the deployment of numerous smart sensing devices in power systems, generating massive amounts of multi-source, heterogeneous big data on power operations.

[0003] However, traditional energy-saving methods mainly rely on periodic manual inspections, assessments based on single energy efficiency indicators (such as power factor), or static benchmark comparisons based on simple historical averages. These methods are often characterized by lag, one-sidedness, and extensiveness, making them difficult to adapt to the dynamic and complex characteristics of modern power systems with varying loads and complex interference factors. At the same time, some data-driven energy-saving technologies typically only target a single link in the energy-saving chain, lacking an end-to-end closed-loop solution from data perception and potential identification to dispatch control, and failing to organically connect macro-grid dispatch with micro-equipment management. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, this invention provides an energy-saving method and system based on power big data. It constructs a unified data model through real-time acquisition and encrypted transmission of smart meters and multi-source data, utilizes machine learning clustering to optimize operating benchmarks, performs node grouping and feature extraction, calculates node deviations in real time and generates an energy-saving potential list, and achieves node resource scheduling and optimization of high-energy-consuming equipment through bidirectional analysis.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] The first aspect of this disclosure provides an energy-saving method based on big data on electricity, comprising the following steps:

[0007] S1. Multi-source heterogeneous data perception: Through smart meters and external data access, after edge preprocessing and RTP encrypted transmission, a data model is constructed based on spatiotemporal fusion of multi-source data in the cloud;

[0008] S2. Node energy-saving potential identification: Based on the data model, establish a dynamic benchmark for the operation scenario to divide the nodes into groups, and then obtain a multi-dimensional comprehensive ranking and generate a list of energy-saving renovation nodes based on the deviation of the group evaluation index.

[0009] S3, Energy-saving scheduling and measure analysis: Energy-saving optimization is completed through bidirectional analysis of the node list, including upward aggregation of nodes to generate scheduling strategies and downward decomposition of load to identify equipment in order to optimize user energy consumption.

[0010] Furthermore, the multi-source heterogeneous data sensing includes the following steps:

[0011] S11. Data Acquisition: By deploying high-frequency sampling smart meters at the bus entry point, power data is collected, including active power, reactive power, voltage, current, power factor and harmonics, and meteorological data, electricity price data and calendar information are accessed based on timestamps and the location coordinates of the power consumption unit.

[0012] S12. Preprocessing: Data is cleaned, filtered, and cached through the edge computing gateway, and then encrypted and sent to the cloud platform using RTP streaming transmission;

[0013] S13. Data Fusion: The cloud platform merges incoming streaming data with batch historical data. The fusion engine uses timestamps and location coordinates as keys to associate and align power data, meteorological data, electricity price data, and calendar information at the same time and location to build a unified data model.

[0014] Furthermore, the RTP streaming includes:

[0015] All data to be transmitted is classified, and a unique RTP payload type value is defined for each type of data. Different serialization encapsulation strategies are adopted according to the data type: for high-frequency power sampling data, the numerical value, measurement point, and quality code are encapsulated according to a preset binary structure; for meteorological data, electricity price data, and calendar information, they are converted into binary streams through serialization, and all data packets are embedded with high-precision timestamps and location coordinate tags during encapsulation.

[0016] The serialized data payload is encapsulated into an RTP packet. The corresponding payload type value is set in the RTP standard header, and the chunk sequence number, total number of chunks, and location metadata are written in the extended header. The transmission priority is marked according to the data real-time requirements, and a differentiated transmission strategy is executed according to the data priority.

[0017] Furthermore, the identification of the node's energy-saving potential includes the following steps:

[0018] S21. Dynamic scenario-based benchmark value acquisition: Construct feature vectors for each time point through multi-dimensional data from the data model, and use K-means clustering to cluster all data points into operating scenario clusters. Within each scenario cluster, select high-quality time points by sorting the time points with the highest energy efficiency, and set the average value of each operating parameter of the high-quality time points as the dynamic benchmark value of each parameter under the operating scenario.

[0019] S22. Node Clustering and Feature Extraction: Calculate the statistical features of the node time series data for a preset time period for each node and construct an initial feature matrix; after standardizing the initial matrix, input it into the PCA algorithm for analysis to obtain a set of principal components of the energy efficiency state dimension, including comprehensive load and energy intensity, reactive power and voltage quality, harmonic pollution and power quality, and load volatility.

[0020] K-means is used to cluster the scores on the principal components of all nodes, and the node populations are output, including high-energy-consuming, low-power-factor, harmonic-polluting, and stable populations.

[0021] S23. Potential Assessment and Node Identification: A dynamic benchmark is obtained through real-time scene matching, and the index deviation is calculated based on the differences in node group labels. Finally, multi-dimensional indicators are integrated for comprehensive evaluation and global ranking to generate a list of nodes to be transformed.

[0022] Furthermore, the node time-series data includes voltage, current, active power, reactive power, power factor, and harmonic distortion rate; statistical characteristics include average load rate, peak load rate, load fluctuation index, total power consumption, average power factor, 95th percentile reactive power, 95th percentile voltage deviation, peak value of total harmonic distortion rate of current, severity of total harmonic distortion rate of current, and percentage of time when harmonics exceed the standard.

[0023] Furthermore, the generated list of modified nodes includes:

[0024] Real-time acquisition of current power data and additional information; matching the current data with the historical cluster centers of various scenarios to the corresponding operating scenario clusters; and mapping dynamic benchmark values ​​based on the operating scenario clusters.

[0025] The real-time operating value of each node is compared with the dynamic benchmark value one by one to calculate its index deviation, and a focusing strategy is executed according to the group classification of each node.

[0026] Based on the indicator deviation and focus results, a comprehensive evaluation vector is constructed for each node, including cost-saving benefits, security risks, and economic costs:

[0027] First, evaluation weights are assigned to different groups: high-energy-consuming nodes are given the highest weight for cost-saving benefits, harmonic pollution nodes are given the highest weight for safety risks, and low-power-factor nodes balance energy saving and safety. Finally, a multi-attribute decision algorithm is used to calculate a comprehensive score for each node and to rank them globally.

[0028] The nodes are sorted according to their comprehensive scores to obtain a list of nodes with energy-saving potential, and the node attribute data is then visualized.

[0029] Furthermore, the focusing strategy includes:

[0030] For nodes identified as high-energy-consuming, calculate the additional electricity cost resulting from the deviation of active power loss from the baseline;

[0031] For nodes with low power factor, calculate the equivalent energy loss caused by reactive power deviation;

[0032] For harmonic pollution nodes, harmonic data is used to assess the additional heat loss caused by harmonic currents.

[0033] The node attribute data includes the node's ID, geographical location, network line, group label, problem type, real-time deviation from the baseline, and adjustable potential estimate.

[0034] Furthermore, the analysis of the energy-saving scheduling and measures includes the following steps:

[0035] S31. Energy-saving scheduling strategy acquisition: By constructing a multi-dimensional tag aggregation load adjustment resource package and integrating external data and rule base for strategy mining, an ordered list of energy-saving scheduling strategies is generated.

[0036] S32. Identification of electricity consumption measures: By collecting load data from the node list at high frequency, the electrical equipment is identified through non-intrusive algorithms, and then an energy efficiency optimization report is generated through correlation analysis.

[0037] The energy-saving scheduling strategy acquisition includes the following steps:

[0038] S311. Constructing a load aggregation pool: By adding multi-dimensional feature labels to nodes and aggregating similar nodes based on the labels to form load adjustment resource packages, a load aggregation pool is constructed.

[0039] S312. Generate optimization strategies: By integrating external data and generating optimization tags based on domain rules, the rule base is queried and filtering conditions are set to finally generate an ordered list of optimization strategy resource packages.

[0040] The construction of the load aggregation pool includes the following steps:

[0041] Construct multi-dimensional feature labels: Add multi-dimensional feature labels to the nodes in the node list, including load nature, time characteristics, and spatial attributes. Load nature includes load reduction, load transfer, and load regulation; time characteristics include adjustable time period, response delay, duration, and minimum interval; spatial attributes include geographical location and grid location.

[0042] Obtaining load balancing resource packages: Based on the added multi-dimensional feature tags, nodes with similar characteristics are aggregated from four dimensions: function, time, space, and performance, thereby obtaining several load balancing resource packages;

[0043] The generation optimization strategy includes the following steps:

[0044] External data fusion: Acquire external knowledge data, including data streams from power grid dispatch centers, electricity markets, and environmental monitoring systems; set thresholds and rules based on domain knowledge; judge and label the data; and obtain optimized labels.

[0045] Optimization strategy mining: Based on the optimized tag query rule base, obtain the action type, set multiple filtering conditions, and obtain an ordered list of resource packages that meet the conditions;

[0046] The multiple screening criteria include:

[0047] Geographic matching: Identify all load conditioning resource packages whose geographic locations are associated with fault or anomaly locations specified in the optimization label;

[0048] Type matching: In the load adjustment resource package set output in the previous step, filter out resource packages whose adjustable type attributes match the action types specified in the rule base;

[0049] Capacity matching: For the currently selected load adjustment resource packages, verify whether their current status is within the available time period;

[0050] Priority matching: The load adjustment resource packages that ultimately meet the criteria are sorted in ascending order of response cost.

[0051] Furthermore, the electricity consumption management identification includes the following steps:

[0052] S321, Load monitoring data acquisition: Expand the data dimensions of the nodes on the node list, and continuously collect the voltage and current waveform data of the total load by enabling the high-frequency sampling mode of the smart meter.

[0053] S322, Load Decomposition and Equipment Identification: The collected high-frequency total load data stream is input into the non-intrusive load monitoring algorithm model. First, the event detection algorithm identifies the moment when the power changes significantly, thereby determining whether equipment is turned on or off. Then, the key feature fingerprints that characterize the equipment are extracted from the change events. Finally, the extracted feature fingerprints are compared and matched with the pre-established equipment feature database to identify the type and status of the energy-consuming equipment in operation and output the electricity consumption list.

[0054] S323. Obtain an energy efficiency optimization report: Perform spatiotemporal alignment and correlation analysis between node attribute data and electricity consumption list data. By cross-comparing the time period of the problem with the operating status of specific equipment, identify the key load equipment and its operating mode that cause the energy efficiency problem, and generate an energy efficiency optimization report.

[0055] The second aspect of this disclosure provides an energy-saving system based on power big data, applied to an energy-saving method based on power big data as described above, including a data acquisition and sensing unit, an energy-saving potential identification unit, and an optimization strategy generation unit;

[0056] The data acquisition and sensing unit is responsible for the comprehensive acquisition of multi-source heterogeneous data, edge preprocessing, secure encrypted transmission, and cloud fusion modeling to generate a data model.

[0057] The energy-saving potential identification unit automatically identifies and evaluates the energy-saving potential of each power consumption node based on the acquired data model, and generates a list of nodes to be prioritized for renovation.

[0058] The optimization strategy generation unit is used to receive the node list, generate macro-level energy-saving scheduling strategies, parse out equipment-level energy efficiency optimization measures, and form an executable energy-saving scheme.

[0059] The beneficial effects of this invention are as follows:

[0060] This invention first involves the real-time acquisition and encrypted transmission of data from smart meters and external multi-source sources, followed by high-precision spatiotemporal fusion in the cloud to construct a unified data model. Then, using this data model, optimal operating benchmarks for different scenarios are extracted from historical data through machine learning clustering technology, and all electricity-consuming nodes are grouped and their features extracted. Based on this, the current operating scenario is matched in real-time, the differential deviation between each node and the optimal benchmark is calculated, and multi-dimensional indicators are integrated for comprehensive evaluation and global ranking, generating a list of nodes with energy-saving potential. Finally, the node list is bidirectionally analyzed: upwards, similar nodes are aggregated to form schedulable resource packages to respond to the macro-level needs of the power grid; downwards, non-intrusive load decomposition technology is used to identify specific high-energy-consuming devices, providing users with accurate optimization reports, thus forming a data-driven, self-optimizing energy-saving closed loop. Attached Figure Description

[0061] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0062] Figure 1 A schematic diagram illustrating the steps of an energy-saving method based on big data of electricity, provided in an embodiment of the present invention;

[0063] Figure 2 This is a schematic diagram of the structure of an energy-saving system based on power big data, provided as an embodiment of the present invention. Detailed Implementation

[0064] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0065] This embodiment provides an energy-saving method based on big data on electricity, such as... Figure 1 As shown, it includes the following steps:

[0066] S1. Multi-source heterogeneous data sensing: Through smart meters and external data access, after edge preprocessing and RTP encrypted transmission, a data model is constructed in the cloud based on spatiotemporal fusion of multi-source data, including the following steps:

[0067] S11. Data Acquisition: By deploying high-frequency sampling smart meters at the bus entry point, power data is collected, including active power, reactive power, voltage, current, power factor, and harmonics. Additional information such as meteorological data (temperature, humidity, and sunlight), electricity price data, and calendar information (weekdays / holidays) is also accessed based on timestamps and the location coordinates of the electricity consumption unit.

[0068] Understandably, when accessing meteorological and electricity price data, the meteorological grid data and electricity price data are precisely interpolated and matched to the geographical location of each electricity-consuming unit using timestamps and the latitude and longitude coordinates of the power station. For example, the system can obtain the irradiance, wind speed, and temperature curves of the park's coordinates for the next 24 hours at the second level, achieving high-precision spatiotemporal alignment.

[0069] S12. Preprocessing: Data is cleaned, filtered, and cached through an edge computing gateway, and then encrypted and sent to the cloud platform using RTP streaming transmission.

[0070] RTP streaming includes:

[0071] All data to be transmitted is categorized, and a unique RTP load type value is defined for each category. For example, active / reactive power sampling data is defined as type 99, meteorological data is defined as type 100, electricity price data is defined as type 101, and calendar information is defined as type 102.

[0072] Different serialization and encapsulation strategies are adopted according to the data type: For high-frequency power sampling data, the values, measurement points and quality codes are encapsulated in a preset binary structure; For meteorological data, electricity price data and calendar information, serialization is performed using formats such as MessagePack or JSON, and the data is converted into a binary stream. At the same time, all data packets are embedded with high-precision timestamps and location coordinate tags during encapsulation.

[0073] The serialized data payload is encapsulated into an RTP packet. The corresponding payload type value is set in the RTP standard header, and metadata such as the block sequence number, total number of blocks (for waveform data), and location are written in the extended header. At the same time, the transmission priority is marked according to the real-time requirements of the data: power data is marked as high priority, meteorological data as medium priority, and electricity price and calendar data as low priority.

[0074] Differentiated transmission strategies are implemented based on data priority. High-priority power data is sent immediately, allowing for a very low degree of packet loss; medium-priority meteorological data is sent when network connectivity is guaranteed, aiming for a low packet loss rate; low-priority electricity price and calendar data adopt a reliable transmission mode, triggering a retransmission mechanism when necessary to ensure data integrity and accuracy.

[0075] It should be noted that RTP streaming has been optimized at the protocol layer to address the time-series and high-throughput characteristics of power data. It achieves protocol-level traffic splitting based on load type, meets different data quality requirements with differentiated strategies, and uses spatiotemporal tags as a unified index for multi-source data fusion, ultimately providing a solid data foundation for advanced energy data analysis.

[0076] S13. Data Fusion: The cloud platform merges incoming streaming data with batch historical data. The fusion engine uses timestamps and location coordinates as keys to associate and align power data, meteorological data, electricity price data, and calendar information at the same time and location to build a unified data model.

[0077] It should be noted that the uploaded data includes timestamps and location coordinates. The data model constructed by merging these coordinates enables traceability and provides a foundation for subsequent mapping of potential node objects.

[0078] S2. Node Energy-Saving Potential Identification: Based on the data model, a dynamic benchmark for the operating scenario is established to divide the nodes into groups. Then, based on the deviation of the group evaluation indicators, a multi-dimensional comprehensive ranking is obtained, and a list of energy-saving retrofit nodes is generated, including the following steps:

[0079] S21. Dynamic Scenario-Based Benchmark Value Acquisition: Construct feature vectors for each time point using multi-dimensional data from the data model. Use K-means clustering to cluster all data points into operating scenario clusters. Within each scenario cluster, sort and filter (e.g., Top 5%) the time points with the highest energy efficiency (lowest total system active power loss). Set the average values ​​of various operating parameters (including not only the total value, but also the voltage of each node, the current of each branch, the power factor of each power consumption unit, etc.) of the high-quality time points as the dynamic benchmark values ​​of each parameter under the operating scenario.

[0080] The multi-dimensional data includes total active power, total reactive power, environmental data, and date type. As you can understand, total active power represents the overall load level, total reactive power represents the overall reactive power demand, and environmental data includes parameters such as ambient temperature and solar irradiance. Ambient temperature directly affects air conditioning load and equipment efficiency, while solar irradiance can accurately predict the output trend of distributed photovoltaic systems. The specific environmental data included will be adjusted according to the actual application scenario. The date type is a categorical variable (e.g., 0-weekday, 1-weekend, 2-holiday), which needs to be uniquely coded before being input along with other continuous variables.

[0081] S22. Node Clustering and Feature Extraction: Calculate the statistical features of node time-series data for a preset time period (e.g., one month) for each node and construct an initial feature matrix; after standardizing the initial matrix, input it into the PCA (Principal Component Analysis) algorithm for analysis to obtain a set of principal components of energy efficiency state dimensions, including comprehensive load and energy intensity, reactive power and voltage quality, harmonic pollution and power quality, and load volatility.

[0082] The node time-series data includes voltage, current, active power, reactive power, power factor, and harmonic distortion rate; the statistical characteristics include average load rate, peak load rate, load fluctuation index, total power consumption, average power factor, 95th percentile reactive power, 95th percentile voltage deviation, peak value of total harmonic distortion rate of current, severity of total harmonic distortion rate of current, and percentage of time when harmonics exceed the standard.

[0083] K-means is used to cluster the scores on the principal components of all nodes, and the node populations are output, including high-energy-consumption, low-power-factor, harmonic-polluting, and stable populations.

[0084] Understandably, node clustering involves classifying all monitored nodes or power-consuming units to identify groups with common problem characteristics. In the above process, PCA is used to reduce dimensionality and extract comprehensive energy efficiency status indicators. Statistical features such as mean, standard deviation, and 95th percentile are calculated to construct an initial feature matrix. Finally, PCA analysis is used to obtain the score of each principal component of the node, i.e., the comprehensive energy efficiency status indicator. Based on these scores, clustering is performed to accurately classify all nodes (monitoring points) into groups such as high energy consumption type, low power factor type, harmonic pollution type, and stable type.

[0085] S23. Potential Assessment and Node Identification: A dynamic benchmark is obtained through real-time scene matching, and the indicator deviation is calculated based on the differences in node group labels. Finally, multi-dimensional indicators are integrated for comprehensive evaluation and global ranking to generate a list of nodes to be renovated, including:

[0086] Real-time acquisition of current power data and additional information; matching the current data with the historical cluster centers of various scenarios to the corresponding operating scenario clusters; and mapping dynamic benchmark values ​​based on the operating scenario clusters.

[0087] The real-time operating value of each node is compared with the dynamic benchmark value one by one to calculate its index deviation. At the same time, a focus strategy is executed according to the population classification of each node, and the focus strategy includes:

[0088] For nodes identified as high-energy-consuming, calculate the additional electricity cost resulting from the deviation of their active power loss from the baseline.

[0089] For low power factor nodes, calculate the equivalent energy loss caused by their reactive power deviation;

[0090] For harmonic pollution nodes, unique harmonic data is used to assess the additional heat loss caused by harmonic currents.

[0091] Understandably, the aforementioned deviation calculation and focusing strategy involves a full parameter comparison of all nodes, but for each node, only its most relevant abnormal indicators are interpreted based on its own group label, and a detailed quantitative potential is generated. In the focusing strategy, for example, the deviation of active power of a high-energy-consuming node is calculated, focusing on this deviation, and the "energy consumption-cost" model is activated. The deviation is multiplied by the operating time and electricity price to directly calculate the additional electricity cost, that is, the "potential" is quantified in RMB / year. At the same time, this node may also have a reactive power deviation, but for this group, this deviation is not the main problem at present, so it is not the core of the potential calculation.

[0092] Based on the indicator deviation and focus results, a comprehensive evaluation vector is constructed for each node, including cost-saving benefits, security risks, and economic costs:

[0093] First, evaluation weights are assigned to different groups: high-energy-consuming nodes are given the highest weight for cost-saving benefits, harmonic pollution nodes are given the highest weight for safety risks, and low-power-factor nodes balance energy saving and safety. Finally, a multi-attribute decision algorithm is used to calculate a comprehensive score for each node and perform global ranking based on this score, thereby identifying priority transformation targets that not only have great potential but also have high urgency for governance and reasonable return on investment.

[0094] For example, if node A calculates an annualized cost-saving potential of 50,000 yuan, it can be directly used as an input for energy-saving indicators. Similarly, the calculated voltage deviation and harmonic exceedance severity, after normalization or scoring, become inputs for safety indicators.

[0095] The nodes are sorted according to their comprehensive scores to obtain a list of nodes with energy-saving potential, and the node attribute data is then visualized.

[0096] It should be noted that during the potential assessment and node identification process, nodes with energy-saving potential are identified based on a comprehensive score. Simultaneously, based on node group attributes, such as whether the node is a source of harmonic pollution or a reactive power deficit point, the most targeted mitigation measures are automatically mapped, providing a foundation for node optimization and scheduling management optimization. Node attribute data includes the node's ID, geographical location, associated line, group label, problem type (e.g., excessive reactive power, large load fluctuations), real-time deviation from the baseline, and adjustable potential estimates.

[0097] S3. Energy-saving scheduling and measure analysis: Energy-saving optimization is achieved through bidirectional analysis of the node list, including upward aggregation of nodes to generate scheduling strategies and downward decomposition of loads to identify equipment in order to optimize user energy consumption, including the following steps:

[0098] S31. Energy-saving scheduling strategy acquisition: By constructing a multi-dimensional tag-aggregated load adjustment resource package and integrating external data and rule bases for strategy mining, an ordered list of energy-saving scheduling strategies is generated, including the following steps:

[0099] S311. Constructing a load aggregation pool: By adding multi-dimensional feature labels to nodes and aggregating similar nodes based on the labels to form load adjustment resource packages, a load aggregation pool is constructed, including the following steps:

[0100] Construct multi-dimensional feature labels: Add multi-dimensional feature labels to the nodes in the node list, including load nature, time characteristics, and spatial attributes. Load nature includes load reduction type, load transfer type, and load regulation type; time characteristics include adjustable time period, response delay, duration, and minimum interval; spatial attributes include geographical location and grid location.

[0101] Understandably, each node in the energy-saving potential list is analyzed in depth, and a series of characteristic labels describing its adjustment capabilities are assigned. Load reduction type refers to loads that can be temporarily shut down or have their power reduced without affecting core functions. Examples include: some non-essential lighting systems, advertising signs, and idle equipment in office areas. Load transfer type refers to loads whose operating time can be flexibly arranged without changing the total energy consumption. Examples include: electric vehicle charging stations, laundry equipment, the charge / discharge cycle of energy storage systems, and pre-cooling / preheating of air conditioning systems utilizing building thermal inertia. Load regulation type refers to loads whose power can change continuously and smoothly. Examples include: water pumps, fans, and refrigeration compressors equipped with frequency converters. Adjustable period: Precisely indicating which days of the week and which time of day the node can accept adjustment commands. Response delay assesses the time required from the central control system issuing a command to the node's load starting to change. This determines the application scenarios (e.g., second / minute response is suitable for frequency regulation, while hourly response is suitable for peak shaving and valley filling). Duration determines the maximum time the node can continuously execute adjustment commands at one time. The minimum interval determines the minimum time interval required between two adjustment actions at this node to ensure that device lifespan or user experience is not affected.

[0102] Obtaining load balancing resource packages: Based on the added multi-dimensional feature tags, nodes with similar characteristics are aggregated from four dimensions: function, time, space, and performance, thereby obtaining several load balancing resource packages.

[0103] Specifically, aggregation by function combines nodes with the same "adjustable type," such as forming a "commercial building air conditioning load reduction package" or an "electric vehicle charging load transfer package"; aggregation by time combines nodes with highly overlapping "adjustable time periods," such as forming a "weekday midday peak adjustable resource package"; aggregation by space combines nodes located in the same substation or on the same feeder, such as forming a "XX Industrial Park A Feeder Resource Package," specifically designed to alleviate congestion pressure on that line; and aggregation by performance combines nodes with similar "response delays," such as forming a "fast response resource package," for participating in the ancillary service market with high speed requirements.

[0104] By acquiring load adjustment resource packages, the energy-saving potential of nodes can be transformed into macroscopically executable scheduling strategies.

[0105] S312. Generating Optimization Strategies: By integrating external data and generating optimization tags based on domain rules, the rule base is queried and filtering conditions are set to ultimately generate an ordered list of optimization strategy resource packages, including the following steps:

[0106] External data fusion: Acquire external knowledge data, including data streams from power grid dispatch centers, electricity markets, and environmental monitoring systems. Based on domain knowledge, set thresholds and rules to judge and label the data, and obtain optimized labels.

[0107] For example, when the load rate of a certain line exceeds 90%, its status is marked as "line overload warning"; when the real-time electricity price climbs to more than 1.5 yuan per kilowatt-hour, it is marked as "extremely high electricity price"; when a clear peak-shaving instruction is received from the power grid company, it is marked as "demand response event triggered". The external knowledge base not only contains the instantaneous values ​​of the data, but more importantly, it contains the underlying labels representing the power grid's safety, economy, and reliability status, providing a complete decision-making basis for subsequent strategy generation.

[0108] Optimization strategy mining: Based on the optimized tag query rule base, obtain the action type, set multiple filtering conditions, and obtain an ordered list of resource packages that meet the conditions.

[0109] The multiple screening criteria include:

[0110] Geographic matching: Identify all load conditioning resource packages whose geographic locations are associated with fault or anomaly locations specified in the optimization label;

[0111] Type matching: In the load adjustment resource package set output in the previous step, filter out resource packages whose adjustable type attributes match the action types specified in the rule base;

[0112] Capacity matching: For the currently selected load adjustment resource packages, verify whether their current status is within the available time period;

[0113] Priority matching: The load adjustment resource packages that ultimately meet the criteria are sorted in ascending order of response cost.

[0114] It should be noted that the query rule base stores logical judgment rules. It is a predefined set of decision rules containing the principle that "if (IF) a certain situation occurs, then (THEN) a certain action should be taken." These rules originate from the principles of power grid physics, power grid safety operation procedures, electricity market rules, and expert operational experience. For example, line overload leads to overheating, necessitating a reduction in active power; industry-mandated safety guidelines, such as equipment not exceeding its rated capacity; rules for economic game theory under different market signals; and strategies for experienced dispatchers to handle specific complex situations.

[0115] Energy-saving dispatch strategy acquisition is the first step in the bidirectional process of energy-saving dispatch and measure analysis: it maps the node list upwards to the dispatch level, thereby optimizing global and regional load balancing, energy efficiency improvement, and network loss reduction in the direction of energy dispatch. For example, the optimization label is: Line L1 overload warning, the trigger rule is: load reduction, and the matching logic includes: finding all resource packages whose geographical location is located on 'Line L1' or its downstream lines, filtering out resource packages with the adjustable type of "load reduction" or "load transfer" from the results of the previous step, filtering out resource packages that are currently in the callable time period, and finally sorting them from low to high response cost.

[0116] S32. Electricity Consumption Management and Identification: By collecting load data from a high-frequency data acquisition node list, electrical equipment is identified through non-intrusive algorithms, and then an energy efficiency optimization report is generated through correlation analysis, including the following steps:

[0117] S321 Load Monitoring Data Acquisition: Expand the data dimensions of the nodes on the node list, and continuously collect the voltage and current waveform data of the total load by enabling the high-frequency sampling mode of smart meters; these high-frequency data streams accurately record the instantaneous changes of the load, including the unique "traces" left by the start-up, shutdown and switching of various electrical equipment, laying the data foundation for subsequent in-depth insights at the equipment level.

[0118] S322. Load Decomposition and Equipment Identification: The collected high-frequency total load data stream is input into a non-intrusive load monitoring algorithm model. First, the event detection algorithm accurately identifies the moments when significant power jumps occur, thus determining whether equipment is turned on or off. Then, key feature fingerprints characterizing the equipment are extracted from these jump events, such as changes in active and reactive power, transient current waveforms at startup, and unique harmonic components. Finally, these extracted features are compared and matched with a pre-established equipment feature database to accurately identify the types and status of the main energy-consuming equipment currently in operation, and output an electricity consumption list. The electricity consumption list includes key information such as equipment type, rated power, number and time of daily start-up and shutdown, running time, and daily power consumption.

[0119] S323. Obtain an energy efficiency optimization report: Perform spatiotemporal alignment and correlation analysis between node attribute data and electricity consumption list data. By cross-comparing the time period of the problem with the operating status of specific equipment, identify the key load equipment and its operating mode that cause the energy efficiency problem, and generate an energy efficiency optimization report.

[0120] Understandably, energy efficiency optimization reports not only accurately pinpoint the root causes of problems and correlate macroscopic phenomena with microscopic equipment responsibilities, but also provide optimization suggestions at different levels, ranging from operational behavior adjustments that require no investment to technological upgrades that require some investment, based on the user's specific business scenarios.

[0121] It should be noted that electricity consumption measure management identification is the second step in a two-way process of energy-saving dispatch and measure analysis: mapping the node list down to the user level, thereby optimizing user behavior and equipment management on both the user and equipment sides. The final generated energy-saving dispatch strategy and energy efficiency optimization report are sent to the dispatch center and the user side, respectively. The dispatch center can verify the feasibility of the energy-saving dispatch strategy through digital twin simulation before execution, while the user side can choose according to its preferences, which will not be elaborated further here.

[0122] This embodiment also provides an energy-saving system based on big data on electricity, such as... Figure 2 As shown, it includes a data acquisition and sensing unit, an energy-saving potential identification unit, and an optimization strategy generation unit.

[0123] The data acquisition and sensing unit is responsible for the comprehensive acquisition of multi-source heterogeneous data, edge preprocessing, secure encrypted transmission, and cloud fusion modeling to generate a data model.

[0124] Specifically, power data (such as power, voltage, current, and harmonics) and external data such as weather, electricity prices, and calendars are collected through smart meters and external interfaces. At the edge, the data is cleaned, filtered, and cached, and encrypted streaming is performed using RTP-based encryption. Differentiated real-time transmission strategies are implemented based on data categories to ensure data security and real-time performance. In the cloud, using high-precision timestamps and geographic coordinates as keys, data from all sources is spatiotemporally aligned and fused to construct a comprehensive data model for subsequent analysis and traceability.

[0125] The energy-saving potential identification unit automatically identifies and evaluates the energy-saving potential of each power consumption node based on the acquired data model, and generates a list of nodes to be prioritized for renovation.

[0126] Specifically, dynamic energy efficiency benchmarks are established for different operating scenarios (such as high-load weekdays and low-load holidays) using clustering algorithms (e.g., K-means). Energy consumption characteristics are extracted and nodes are grouped (e.g., high-energy-consumption, low-power-factor, harmonic-polluting) to accurately characterize the energy efficiency status of each node. The real-time operating parameters of each node are compared with the dynamic benchmarks of its respective scenario. Deviations are calculated based on the differences in its group characteristics, and a multi-attribute decision-making algorithm is used for comprehensive evaluation and global ranking. Finally, a priority list of nodes for modification, incorporating cost-saving benefits, governance urgency, and return on investment, is output.

[0127] The optimization strategy generation unit is used to receive the node list, generate macro-level energy-saving scheduling strategies, parse out equipment-level energy efficiency optimization measures, and form an executable energy-saving scheme.

[0128] Specifically, upward aggregation (scheduling strategy): Multi-dimensional tags such as function, time, and space are added to the nodes in the list, aggregating them into schedulable "load adjustment resource packages." External data from the power grid, market, and other sources are integrated with rule bases to generate ordered (e.g., sorted by cost) and executable optimized scheduling strategies (e.g., load reduction and transfer). Downward decomposition (equipment identification): High-frequency data acquisition is enabled for the nodes in the list, and non-intrusive load monitoring (NILM) technology is used to decompose the total load waveform, identifying key energy-consuming equipment and their operating modes (e.g., start-up and shutdown times and power consumption of air conditioning, lighting, etc.).

[0129] Ultimately, two types of outputs are generated: one is the aggregated scheduling strategy reported to the scheduling center; the other is the energy efficiency diagnostic report, which contains specific equipment optimization suggestions, and is sent to users.

[0130] This invention achieves deep fusion and high-precision traceability of multi-source heterogeneous data: through high-frequency sampling of smart meters and precise spatiotemporal interpolation matching of external data (weather, electricity price), and by utilizing an innovative differentiated RTP streaming protocol, it ensures the comprehensiveness, real-time performance and reliability of data collection.

[0131] Establish dynamic and scenario-based accurate evaluation benchmarks: Define different operating scenarios through K-means clustering, and extract dynamic benchmark values ​​from "high-quality scenarios" to make the evaluation criteria more scientific and more in line with the optimal operating state of the system, avoiding the one-sidedness of static benchmarks.

[0132] Achieving multi-dimensional and differentiated accurate profiling and ranking of node potential: Principal component analysis (PCA) and K-means clustering are used to group nodes from multiple energy efficiency dimensions (such as high energy consumption type and low power factor type); a focusing strategy is executed based on the cluster labels to calculate differentiated indicator deviations; and finally, a multi-attribute decision algorithm is used for comprehensive ranking to ensure that the identified priority transformation targets not only have great potential, but also have high governance urgency and reasonable return on investment.

[0133] A closed-loop energy-saving strategy with bidirectional analysis at both the macro and micro levels is formed: Upward aggregation transforms the micro-node potential into macro-grid-understandable and dispatchable "resources" by adding multi-dimensional feature tags to nodes and aggregating them into load regulation resource packages. The generated ordered optimization strategy list can be directly used for demand response or grid regulation. Downward decomposition uses non-intrusive load monitoring technology to decompose the total load of nodes to specific devices, generating energy efficiency optimization reports and providing users with specific and actionable energy-saving suggestions such as "when to shut down which devices".

[0134] This invention first involves the real-time acquisition and encrypted transmission of data from smart meters and external multi-source sources, followed by high-precision spatiotemporal fusion in the cloud to construct a unified data model. Then, using this model, optimal operating benchmarks for different scenarios are extracted from historical data through machine learning clustering technology, and all electricity-consuming nodes are grouped and their features extracted. Based on this, by real-time matching of the current operating scenario, the differential deviation between each node and the optimal benchmark is calculated, and multi-dimensional indicators are integrated for comprehensive evaluation and global ranking, automatically generating a list of energy-saving renovation nodes. Finally, this list is bidirectionally parsed: upwards, similar nodes are aggregated to form schedulable resource packages to respond to the macro-level needs of the power grid; downwards, non-intrusive load decomposition technology is used to identify specific high-energy-consuming equipment, providing users with accurate optimization reports, thus forming a data-driven, self-optimizing energy-saving closed loop.

[0135] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An energy saving method based on power big data, characterized in that: Includes the following steps: Multi-source heterogeneous data perception: Through smart meters and external data access, after edge preprocessing and RTP encrypted transmission, a data model is constructed by the cloud based on spatiotemporal fusion of multi-source data; Node energy-saving potential identification: Based on the data model, a dynamic benchmark for the operation scenario is established to divide the nodes into groups. Then, based on the deviation of the group evaluation index, a multi-dimensional comprehensive ranking is obtained and a list of energy-saving renovation nodes is generated. Energy-saving scheduling and measures analysis: Energy-saving optimization is achieved through bidirectional analysis of the node list, including upward aggregation of nodes to generate scheduling strategies and downward decomposition of loads to identify equipment in order to optimize user energy consumption; The multi-source heterogeneous data sensing includes the following steps: Data Acquisition: By deploying high-frequency sampling smart meters at the bus entry point, power data is collected, including active power, reactive power, voltage, current, power factor, and harmonics. Meteorological data, electricity price data, and calendar information are also accessed based on timestamps and the location coordinates of the electricity consumption units. Preprocessing: Data is cleaned, filtered, and cached through an edge computing gateway, and then encrypted and sent to the cloud platform using RTP streaming transmission; Data fusion: The cloud platform merges incoming streaming data with batch historical data. The fusion engine uses timestamps and location coordinates as keys to associate and align power data, meteorological data, electricity price data, and calendar information at the same time and location to build a unified data model. The identification of node energy-saving potential includes the following steps: Dynamic scenario-based benchmark value acquisition: The feature vector of each time point is constructed through the multi-dimensional data of the data model. K-means clustering is used to cluster all data points into running scenario clusters. Within each scenario cluster, the high-quality time points are selected by sorting the time points with the highest energy efficiency. The average value of each running parameter of the high-quality time points is set as the dynamic benchmark value of each parameter under the running scenario. Node clustering and feature extraction: Calculate the statistical features of the node time series data for a preset time period for each node and construct an initial feature matrix; after standardizing the initial matrix, input it into the PCA algorithm for analysis to obtain a set of principal components of the energy efficiency state dimension, including comprehensive load and energy intensity, reactive power and voltage quality, harmonic pollution and power quality, and load volatility; K-means is used to cluster the scores on the principal components of all nodes, and the node populations are output, including high-energy-consuming, low-power-factor, harmonic-polluting, and stable populations. Potential assessment and node identification: A dynamic benchmark is obtained through real-time scene matching, and the index deviation is calculated based on the differences in node group labels. Finally, multi-dimensional indicators are integrated for comprehensive evaluation and global ranking to generate a list of nodes to be transformed. The energy-saving scheduling and measures analysis includes the following steps: Energy-saving scheduling strategy acquisition: By constructing a multi-dimensional tag aggregation load adjustment resource package and integrating external data and rule base for strategy mining, an ordered list of energy-saving scheduling strategies is generated; Electricity consumption management identification: Load data from a high-frequency collection node list is used to decompose and identify electrical equipment using a non-intrusive algorithm, and then an energy efficiency optimization report is generated through correlation analysis. 2.The energy saving method based on power big data according to claim 1, characterized in that: The RTP streaming includes: All data to be transmitted is classified, and a unique RTP payload type value is defined for each type of data. Different serialization encapsulation strategies are adopted according to the data type: for high-frequency power sampling data, the numerical value, measurement point, and quality code are encapsulated according to a preset binary structure; for meteorological data, electricity price data, and calendar information, they are converted into binary streams through serialization, and all data packets are embedded with high-precision timestamps and location coordinate tags during encapsulation. The serialized data payload is encapsulated into an RTP packet. The corresponding payload type value is set in the RTP standard header, and the chunk sequence number, total number of chunks, and location metadata are written in the extended header. The transmission priority is marked according to the data real-time requirements, and a differentiated transmission strategy is executed according to the data priority. 3.The energy saving method based on power big data according to claim 1, characterized in that: The node time-series data includes voltage, current, active power, reactive power, power factor, and harmonic distortion rate; statistical features include average load rate, peak load rate, load fluctuation index, total power consumption, average power factor, 95th percentile reactive power, 95th percentile voltage deviation, peak value of total harmonic distortion rate of current, severity of total harmonic distortion rate of current, and percentage of time when harmonics exceed the standard.

4. The energy saving method based on power big data according to claim 3, characterized in that: The generated list of modified nodes includes: Real-time acquisition of current power data and additional information; matching the current data with the historical cluster centers of various scenarios to the corresponding operating scenario clusters; and mapping dynamic benchmark values ​​based on the operating scenario clusters. The real-time operating value of each node is compared with the dynamic benchmark value one by one to calculate its index deviation, and a focusing strategy is executed according to the group classification of each node. Based on the indicator deviation and focus results, a comprehensive evaluation vector is constructed for each node, including cost-saving benefits, security risks, and economic costs: First, evaluation weights are assigned to different groups: high-energy-consuming nodes are given the highest weight for cost-saving benefits, harmonic pollution nodes are given the highest weight for safety risks, and low-power-factor nodes balance energy saving and safety. Finally, a multi-attribute decision algorithm is used to calculate a comprehensive score for each node and to rank them globally. The nodes are sorted according to their comprehensive scores to obtain a list of nodes with energy-saving potential, and the node attribute data is then visualized.

5. The energy saving method based on power big data according to claim 4, characterized in that: The focusing strategy includes: For nodes identified as high-energy-consuming, calculate the additional electricity cost resulting from the deviation of active power loss from the baseline; For nodes with low power factor, calculate the equivalent energy loss caused by reactive power deviation; For harmonic pollution nodes, harmonic data is used to assess the additional heat loss caused by harmonic currents. The node attribute data includes the node's ID, geographical location, network line, group label, problem type, real-time deviation from the baseline, and adjustable potential estimate. 6.The energy saving method based on power big data according to claim 1, characterized in that: The energy-saving scheduling strategy acquisition includes the following steps: Constructing a load aggregation pool: By adding multi-dimensional feature labels to nodes and aggregating similar nodes based on the labels to form load adjustment resource packages, a load aggregation pool is constructed. Generate optimization strategies: By integrating external data and generating optimization tags based on domain rules, the rule base is queried and filtering conditions are set to finally generate an ordered list of optimization strategy resource packages; The construction of the load aggregation pool includes the following steps: Construct multi-dimensional feature labels: Add multi-dimensional feature labels to the nodes in the node list, including load nature, time characteristics, and spatial attributes. Load nature includes load reduction, load transfer, and load regulation; time characteristics include adjustable time period, response delay, duration, and minimum interval; spatial attributes include geographical location and grid location. Obtaining load balancing resource packages: Based on the added multi-dimensional feature tags, nodes with similar characteristics are aggregated from four dimensions: function, time, space, and performance, thereby obtaining several load balancing resource packages; The generation optimization strategy includes the following steps: External data fusion: Acquire external knowledge data, including data streams from power grid dispatch centers, electricity markets, and environmental monitoring systems; set thresholds and rules based on domain knowledge; judge and label the data; and obtain optimized labels. Optimization strategy mining: Based on the optimized tag query rule base, obtain the action type, set multiple filtering conditions, and obtain an ordered list of resource packages that meet the conditions; The multiple screening criteria include: Geographic matching: Identify all load conditioning resource packages whose geographic locations are associated with fault or anomaly locations specified in the optimization label; Type matching: In the load adjustment resource package set output in the previous step, filter out resource packages whose adjustable type attributes match the action types specified in the rule base; Capacity matching: For the currently selected load adjustment resource packages, verify whether their current status is within the available time period; Priority matching: The load adjustment resource packages that ultimately meet the criteria are sorted in ascending order of response cost.

7. The energy saving method based on power big data according to claim 6, characterized in that: The electricity consumption management identification includes the following steps: Load monitoring data acquisition: Expand the data dimensions of the nodes on the node list, and continuously collect the voltage and current waveform data of the total load by enabling the high-frequency sampling mode of smart meters; Load decomposition and equipment identification: The high-frequency total load data stream is input into the non-intrusive load monitoring algorithm model. First, the event detection algorithm identifies the moment when the power changes significantly, thereby determining whether equipment is turned on or off. Then, the key feature fingerprints that identify the equipment are extracted from the change events. Finally, the extracted feature fingerprints are compared and matched with the pre-established equipment feature database to identify the type and status of the energy-consuming equipment in operation and output the electricity consumption list. Obtain an energy efficiency optimization report: Perform spatiotemporal alignment and correlation analysis between node attribute data and electricity consumption list data. By cross-comparing the time period of the problem with the operating status of specific equipment, identify the key load equipment and its operating mode that cause the energy efficiency problem, and generate an energy efficiency optimization report.

8. An energy-saving system based on power big data, executing an energy-saving method based on power big data as described in any one of claims 1-7, characterized in that: It includes a data acquisition and sensing unit, an energy-saving potential identification unit, and an optimization strategy generation unit; The data acquisition and sensing unit is responsible for the comprehensive acquisition of multi-source heterogeneous data, edge preprocessing, secure encrypted transmission, and cloud fusion modeling to generate a data model. The energy-saving potential identification unit automatically identifies and evaluates the energy-saving potential of each power consumption node based on the acquired data model, and generates a list of nodes to be prioritized for renovation. The optimization strategy generation unit is used to receive the node list, generate macro-level energy-saving scheduling strategies, parse out equipment-level energy efficiency optimization measures, and form an executable energy-saving scheme.

Images