CN-121980478-A - Multi-dimensional edge matching load identification method based on power utilization acquisition and breadth search

CN121980478ACN 121980478 ACN121980478 ACN 121980478ACN-121980478-A

Abstract

The invention discloses a multidimensional edge matching load identification method based on power acquisition and breadth search, which relates to the technical field of load identification, and aims at the core problem of serious deficiency of power event detection and load state identification accuracy in the prior art, the method comprises the steps of defining load types, distinguishing power ramp and mutation by first-order and second-order difference, extracting multi-dimensional edge characteristics, designing a dynamic threshold adjustment mechanism, realizing multi-dimensional breadth-first search matching of rising edges and falling edges, extracting multi-dimensional load characteristics to distinguish similar loads, and planning and distributing load electricity consumption in real time according to breadth-first search results to form a set of full-flow technical scheme from power signal detection to load identification and energy consumption metering. The method realizes the accurate detection of the power event and the effective identification of the full-load working state, improves the overall accuracy of non-invasive load identification from a core link, and lays a foundation for the follow-up load decomposition and the accurate measurement of energy consumption.

Inventors

CUI MENGHAN
XIA YUBAO
CHEN YOUFENG
LIU GUOPENG
SUN YAN
SHI JIAFENG
SHI MENGYUN
LI QUAN
WANG YIYU
ZHANG XUAN
WANG JING
HE CHENGJIAN
WANG GUOYU

Assignees

南京米特科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. The multidimensional edge matching load identification method based on power acquisition and breadth search is characterized by comprising the following steps of: Step S10, based on the current and voltage original data acquired by the power utilization information acquisition system in real time, calculating total active power, total reactive power and power factor, and forming a continuous power sequence as input of a non-invasive load identification algorithm through averaging and noise reduction treatment; step S20, dividing the electric loads into three basic models according to the load power change characteristics in advance, aiming at different types of electric loads, distinguishing power ramp and power mutation by calculating first-order and second-order differences on continuous power sequences, and filtering false power change signals caused by noise; Step S30, extracting multidimensional features of corresponding edges of the detected power mutation points, wherein the multidimensional features comprise a time dimension, a power dimension and a difference dimension, constructing a normalized feature vector, and dynamically adjusting a matching tolerance threshold of the multidimensional edge features according to the real-time working condition, the load superposition degree and the load type of the power grid; step S40, taking the multi-dimensional edge characteristics as a core, adopting a breadth-first search strategy, matching the power rising and falling edges according to the principle of highest similarity and smallest matching cluster, extracting load steady-state and dynamic characteristics from a successfully matched load operation section, constructing a load characteristic library, and distinguishing similar loads; step S50, combining load type definition, power mutation detection, edge matching and load characteristic recognition results, recognizing a load full-working state, splitting a total power sequence according to the recognition results according to time dimension, and planning and distributing the power consumption of various types of loads in real time; And step S60, integrating the processing results of all the steps, and outputting a visualized and standardized non-invasive load identification and energy consumption measurement result, wherein the visualized and standardized non-invasive load identification and energy consumption measurement result comprises load basic information, full-working state information, energy consumption measurement information and a power decomposition curve.
2. The multi-dimensional edge matching load identification method based on power consumption collection and breadth search as claimed in claim 1, wherein in step S20, three kinds of power consumption loads are defined as follows: Single-state load, namely, only having two working states of on and off, wherein active power and reactive power are kept constant during operation, and no intermediate state exists; The multi-state load comprises a plurality of discrete working states, a plurality of working gears and modes, and the power is suddenly changed in a step mode when the working states are switched; continuously and smoothly regulating the power according to the working requirement, wherein the power is in a slowly-changing characteristic without discrete state.
3. The multi-dimensional edge matching load identification method based on electricity collection and breadth search according to claim 1, wherein in step S20, a first-order difference and a second-order difference are calculated for a second-level total active power sequence and a second-order total reactive power sequence respectively: The first-order difference represents the variation of power of adjacent time points and is used for capturing variation trend and speed, and the second-order difference pair calculates the variation and is used for precisely positioning inflection points and edges of a power sequence; The distribution of the second-order difference of the total active power sequence is counted by analyzing the historical data and the power sequence of the no-load variation period of the power system, and the threshold value is judged on the basis of 3 times of the standard deviation of the second-order difference sequence of the total active power; Judging power mutation through second-order difference, and judging power mutation when the absolute value of the second-order difference exceeds a basic judging threshold value, wherein the power mutation corresponds to start-stop of a single-state load and a gear or mode switching event of a multi-state load; Judging power ramp by the first-order difference, and judging the power ramp when the second-order difference does not exceed a basic judging threshold value but the first-order difference is continuously and smoothly non-zero, wherein the power ramp corresponds to a power adjusting process of a continuously-changing load; and introducing reactive power difference characteristics to perform cross verification, and judging that the effective power changes only when the gradual change and abrupt change signals detected by the active power difference are consistent with the trend of the reactive power difference characteristics.
4. The method for identifying multi-dimensional edge matching load based on power acquisition and breadth search according to claim 1, wherein in step S30, for the detected effective power mutation point, a time window is extended forward and backward with respect to the point as a center, the window is defined as an edge interval to be analyzed, multi-dimensional features of corresponding rising edge and falling edge are extracted from the window, and the multi-dimensional edge features include: the time dimension is edge start time, edge end time and edge duration; the power dimension is the active power variation amplitude, the reactive power variation amplitude and the power factor variation value; the difference dimension is a first-order difference peak value, a second-order difference peak value and a difference change rate; And carrying out normalization processing on the multi-dimensional edge characteristics to construct multi-dimensional feature vectors of each rising edge and each falling edge, wherein the multi-dimensional feature vectors are used as core basis for the subsequent breadth-first search matching.
5. The multi-dimensional edge matching load identification method based on power consumption acquisition and breadth search as claimed in claim 1, wherein in step S30, the matching tolerance threshold of the multi-dimensional edge feature is dynamically adjusted according to the real-time working condition, the load superposition degree and the load type of the power grid, and the detailed steps are as follows: Setting a judgment index for dynamic adjustment of a threshold value, wherein the judgment index comprises power grid noise intensity obtained by calculation of fluctuation variance of a differential sequence, load superposition degree and power consumption load type, which are judged by the number and density of power mutation points; replacing the fixed similarity matching threshold value with a dynamic threshold value, wherein the value of the fixed similarity matching threshold value is determined by the judging index and is expressed as a weighted sum of a basic similarity threshold value, power grid noise intensity, load superposition degree and load type adjustment items; When the noise intensity of the power grid is larger than a dynamic threshold value and power edges are overlapped due to superposition of multiple devices, the matching tolerance threshold values of the active power variation amplitude, the reactive power variation amplitude and the edge duration time characteristics are enlarged, meanwhile, the threshold values of the power factor variation value and the differential variation rate are kept unchanged, and leakage matching caused by complex working conditions is avoided; When the noise intensity of the power grid is smaller than the dynamic threshold value and no load is added, the matching tolerance threshold value of all the characteristics is reduced, and false matching is avoided; And aiming at the gear and mode switching of the multi-state load, a matching threshold interval of the power variation amplitude is independently set, and the characteristic rule of step mutation is adapted.
6. The multi-dimensional edge matching load identification method based on power acquisition and breadth search according to claim 1, wherein in step S40, based on the dynamically adjusted matching tolerance threshold, the accurate matching of the power mutation edge is realized through the multi-dimensional breadth first search matching algorithm of the rising and falling edges, comprising the following detailed steps: Setting a core rule of multi-dimensional breadth-first search matching: time logic rules that the matched falling edge time must be after the corresponding rising edge; The feature similarity rule is that the similarity of the multidimensional feature vectors of the edge pairs to be matched is higher than a similarity threshold set by a dynamic threshold; Calculating the feature comprehensive similarity of all the candidate rising and falling edges by taking the multi-dimensional edge feature vector as a core, and constructing an edge matching cost function by adopting a calculation method of combining Euclidean distance and cosine similarity; and solving the cost function by adopting a breadth-first search strategy, wherein the edge pair with highest comprehensive similarity of the prior matching features is adopted, so that the number of edges contained in the constructed matching cluster is minimized, and meanwhile, intermediate calculation results are recorded and multiplexed through an algorithm, thereby avoiding repeated matching attempts.
7. The multi-dimensional edge matching load identification method based on power consumption collection and breadth search according to claim 1, wherein in step S40, for each load operation section corresponding to each successfully matched edge cluster, multi-dimensional load steady-state characteristics and dynamic characteristics are extracted, a complete load characteristic library is constructed, and comprehensive matching of multi-dimensional load characteristics is realized, and the method comprises the following detailed steps: Extracting load steady-state characteristics, namely an active power average value, a reactive power average value, a rated power interval, a power factor steady value and equipment operation efficiency; Extracting load dynamic characteristics, namely power mutation rate, state transition time, power maintenance time of each working state and power gradual change adjusting range; The method comprises the steps of constructing a standardized load template library in advance, storing the multi-dimensional load characteristic reference values and tolerance ranges of various common electric equipment in the library, carrying out comprehensive similarity calculation on the extracted multi-dimensional load characteristics to be identified and the characteristic reference values in the load template library, selecting the equipment type corresponding to the template with the highest similarity as an identification result, and realizing the accurate distinguishing of similar loads through the cross verification of the multi-dimensional load characteristics.
8. The multi-dimensional edge matching load identification method based on power consumption collection and breadth search according to claim 1, wherein in step S50, the full-operation states of different types of loads are identified by combining load type definition, power mutation detection, edge matching and load feature identification results: for single-state load, identifying the start-stop state of a switch and a corresponding time node; Aiming at the multi-state load, not only the start-stop state of the multi-state load is identified, but also gear switching, mode adjustment and non-start-stop type working state transition in the running process of the multi-state load are identified, and meanwhile, the switching time and the corresponding power value of each state are recorded; Aiming at the continuous variable load, tracking the power ramp process, and identifying the starting time, the ending time and the power change range of power regulation of the load, so as to realize the whole-course identification of the ramp power state; aiming at a long-term stable operation type load, the characteristic anchoring is carried out through the characteristic characteristics of constant low power and stable power factor, so that the signal is stripped from the power change of the equipment, and the omission is avoided.
9. The multi-dimensional edge matching load identification method based on power consumption collection and breadth search according to claim 1, wherein in step S50, the total power sequence is split segment by segment according to the edge multi-dimensional breadth priority search result and the load full-working state identification result, and the power consumption of each load is planned and distributed in real time: aiming at a single-state load, multiplying the operation time length of the single-state load in the on state by corresponding constant active power, and accurately calculating the electricity consumption of the single-state load; for multi-state load, multiplying the operation time length of different working states by the active power of the corresponding state, respectively calculating the power consumption of each state, and accumulating to obtain the total power consumption of the load; for continuously-changing load, performing time integration on an active power sequence in a power ramp process by an integration method, and calculating actual power consumption in an adjustment process; Aiming at the superposition operation time period of multiple devices, the matched power decomposition result is preferentially searched according to the multi-dimensional breadth, and the total power consumption is accurately split and distributed in real time according to the real-time power contribution duty ratio of each load.
10. The multi-dimensional edge matching load identification method based on electricity collection and breadth search according to claim 1, wherein in step S60, the processing results of all the above steps are integrated, and the visualized and standardized non-invasive load identification and energy consumption measurement results are output, and specific output contents include: Basic information of each load, namely load type and identification confidence; The full working state information of each load, namely starting and stopping time, gear position, mode switching time and power change range; the energy consumption metering information of each load comprises the electricity consumption, the total electricity consumption and the electricity consumption in unit time of each working state; The power decomposition curve is a total power change curve, each independent load power change curve and a continuous change load power gradual change curve.

Description

Multi-dimensional edge matching load identification method based on power utilization acquisition and breadth search Technical Field The application relates to the technical field of load identification, in particular to a multidimensional edge matching load identification method based on power consumption acquisition and breadth search. Background The non-invasive load identification is a technology for identifying and decomposing the running states and the energy consumption of various electric equipment by analyzing the electricity consumption data such as current, voltage and the like at the total entrance of a user without installing a sensor on each electric appliance, and has become a high-level analysis function in an electricity consumption acquisition system, and is widely applied to electricity consumption analysis, abnormal electricity consumption monitoring and household demand response potential evaluation. Existing non-invasive load identification methods can be divided into the following three categories: The classical basic algorithm based on traditional event detection is that by carrying out time domain analysis on an acquired power sequence, setting a threshold value to detect power mutation points, extracting power variation amplitude, edge characteristics and the like of the mutation points, and combining with a simple matching rule, the method realizes preliminary decomposition and identification of loads, but has weak anti-interference capability, is easily influenced by power grid noise and measurement errors, and is easy to cause event false detection when multiple devices are overlapped; extracting steady-state shallow features of a load through event detection, inputting the features into a lightweight machine learning model to realize load type matching and recognition, wherein the method relies on the manually extracted shallow features, the similar loads are easy to recognize and confuse, and the accuracy rate is reduced when multiple devices are overlapped; The method is based on a lightweight deep learning algorithm, wherein simple feature coding is carried out on a power sequence after event detection, a lightweight deep learning model is input, time domain features of loads are automatically extracted to replace manual feature extraction, and end-to-end real-time load identification is realized. Disclosure of Invention Aiming at the technical problems in the prior art, the application provides a multidimensional edge matching load identification method based on power acquisition and breadth search, which aims at solving the core technical problems of serious deficiency of power event detection and load state identification in the existing non-invasive load identification technology, solves the technical pain points of the prior art that the power mutation detection has weak anti-interference capability, no effective identification means on power change of equipment non-start-stop working state transition and natural blind area detection of specific type load existence state, and specifically solves the problems as follows: The method solves the problems that the existing power event detection method is easy to be interfered by power grid transient noise and measurement errors, and power changes are overlapped and edges are overlapped due to simultaneous switching of multiple devices, so that event false detection and missing detection are caused, and finally load decomposition failure is caused; The method solves the problems that only focusing equipment start-stop event detection is carried out, power step or smooth fluctuation caused by non-start-stop load working state transition is ignored, and further state misjudgment (state transition is misjudged as new equipment start-stop) or state identification deletion (multi-state operation equipment is regarded as a single power state) is caused; The method solves the problems that state changes without obvious power mutation on continuous variable load cannot be captured, weak power signals of a permanently operated low-power load are easy to be covered by high-power equipment and are missed, power changes of different parts of the multi-state load are easy to be mistakenly identified as a plurality of independent equipment, and equipment state and quantity judgment are distorted in the prior art; The application provides a multidimensional edge matching load identification method based on electricity collection and breadth search, which realizes the accurate detection of power events and the effective identification of the full working state of loads, improves the overall accuracy of non-invasive load identification from a core link and lays a foundation for the accurate measurement of subsequent load decomposition and energy consumption, and comprises the following steps: Step S10, based on the current and voltage original data acquired by the power utilization information acquisition system in real time, calculating tota