CN-121996916-A - Industrial and commercial user load data cleaning method and device for energy storage safety

CN121996916ACN 121996916 ACN121996916 ACN 121996916ACN-121996916-A

Abstract

The invention discloses an energy storage safety-oriented industrial and commercial user load data cleaning method and device, and belongs to the technical field of electric power data cleaning. The method comprises the steps of obtaining original load data, preprocessing an original load sequence to obtain a preprocessed load sequence, carrying out time domain decomposition on the preprocessed load sequence to obtain a trend component, a periodic component and a remainder component, carrying out frequency domain decomposition on the remainder component to obtain natural mode components with different frequencies, wherein the natural mode components comprise a low-frequency component and a high-frequency component, carrying out abnormal date confirmation based on the high-frequency component, correcting the remainder component of the abnormal date based on a preset PSO-BP prediction model, and superposing the trend component, the periodic component and the corrected remainder component to obtain corrected load data. The invention can obviously improve the accuracy and reliability of cleaning the power data of the users of the industrial and commercial energy storage system.

Inventors

LV YIFAN
ZENG AIDONG
LI JINGJIAO
Du Zhimen
WEI QIANG

Assignees

南京工程学院

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (9)

1. The energy storage safety-oriented industrial and commercial user load data cleaning method is characterized by comprising the following steps of: Acquiring original load data in a target power grid system, and preprocessing the original load data to obtain a preprocessing load sequence; Performing time domain decomposition on the preprocessing load sequence to obtain a trend component, a periodic component and a remainder component; Carrying out frequency domain decomposition on the remainder component to obtain natural mode components with different frequencies, wherein the natural mode components comprise a low-frequency component and a high-frequency component; Confirming an abnormal date based on the high-frequency component, and correcting a residual component of the abnormal date based on a preset PSO-BP prediction model, wherein the input of the preset PSO-BP prediction model is the residual component corresponding to the abnormal date, the date type and the air-condition factor, and the input is the corrected residual component; and superposing the trend component, the periodic component and the corrected remainder component to obtain corrected load data.
2. The energy storage security oriented industrial and commercial user load data cleaning method according to claim 1, wherein the method for performing time domain decomposition on the preprocessing load sequence comprises the following steps: Performing iterative decomposition of the inner loop and the outer loop which are nested with each other on the preprocessing load sequence until the number of times of the outer loop reaches a preset value, and outputting corresponding trend components, period components and remainder components; wherein, the internal circulation includes the following steps: The method comprises the steps of carrying out a pretreatment load sequence, carrying out a trending to obtain a trending sequence, carrying out a LOESS regression smoothing process to the trending sequence to obtain a periodic component of a current iteration round, carrying out a cycle removing sequence to the pretreatment load sequence, and carrying out a LOESS regression smoothing process to the cycle removing sequence to obtain a trending component of the current iteration round; the outer loop comprises the following steps: Based on the periodic component and the trend component obtained by the inner loop, calculating the remainder component of the current iteration round, calculating Lu Bangquan weight of each time point based on the remainder component, and feeding back the robust weight to the inner loop to participate in the LOESS regression smoothing process.
3. The energy storage security oriented business user load data cleaning method according to claim 1, wherein the method for performing frequency domain decomposition on the remainder component comprises: And decomposing the residual component by an EMD method, wherein the formula is as follows: Wherein, the The component of the residual term is represented, Representing the i-th natural modal component, The residual function is represented, and n represents the number of decomposed natural mode components.
4. The energy storage security oriented business user load data cleaning method according to claim 1, wherein the method for confirming abnormal date based on the high frequency component comprises: And grouping the high-frequency components according to the date type, processing the grouped high-frequency components in a feature space based on a density clustering algorithm, and marking the date corresponding to the data point falling into the preset low-density area as an abnormal date.
5. The energy storage safety-oriented business user load data cleaning method according to claim 1, further comprising the steps of determining a correlation of weather factors and loads and a correlation of date types and loads before correcting the remainder component of the abnormal date based on a preset PSO-BP prediction model; If the weather factor is related to the load, inputting the weather factor into a preset PSO-BP prediction model, otherwise, setting the weather factor to zero, and if the date type is related to the load, inputting the date type into the preset PSO-BP prediction model, otherwise, setting the date type to zero.
6. The energy storage safety-oriented industrial and commercial user load data cleaning method according to claim 5, wherein the method for determining the correlation between the meteorological factors and the loads is characterized in that the pearson correlation coefficient of the load 7-day moving average sequence and the daily average air temperature sequence corresponding to the low-frequency components is calculated based on the low-frequency components, and a calculation formula is as follows: Wherein, the Representing the correlation coefficient; A daily average air temperature sequence is shown; representing a 7-day moving average sequence of load; The standard deviation of the daily average air temperature sequence and the load 7-day moving average sequence is represented, cov (DEG) represents a covariance function, if the correlation coefficient is larger than a preset weather threshold value, the load is related to weather factors, otherwise, the load is not related; The method for determining the correlation between the date type and the load comprises the steps of taking the abnormal date as a day to be detected, and carrying out correlation analysis on the abnormal date and the load based on a probability distribution function, wherein the probability distribution function is as follows: Wherein, the A probability value indicating that the daily load level to be detected falls within a normal range; Representing the actual load of the day to be detected; A power consumption expected value representing the date of the same type; And if the probability value is smaller than a preset probability threshold, indicating that the load is related to the date type, otherwise, not related.
7. The energy storage security oriented business user load data cleaning method of claim 6, wherein if the date type is related to the load, then performing a secondary adjustment on the modified load data based on the date type.
8. The energy storage safety-oriented industrial and commercial user load data cleaning method according to claim 1 is characterized in that the preset PSO-BP prediction model is formed by coupling a convolutional neural network and a counter-propagating neural network, deep structural features of input data are extracted through two groups of convolutional layers and pooling layers, nonlinear mapping is performed through the BP network, the extracted deep structural features are mapped to an output target, wherein the nonlinear mapping uses mean square error of a minimum training set as an adaptability function, and initial weights and thresholds of the PSO-BP prediction model are subjected to iterative optimization in a global parameter space by means of a particle swarm algorithm.
9. An energy storage security oriented business user load data cleaning device for implementing the energy storage security oriented business user load data cleaning method of any one of claims 1 to 8, the device comprising: the data preprocessing module is used for acquiring original load data in a target power grid system, and processing random power failure and significant outliers in the original load data to obtain a preprocessing load sequence; The data decomposition module is used for carrying out time domain decomposition on the preprocessing load sequence to obtain a trend component, a periodic component and a remainder component, and carrying out frequency domain decomposition on the remainder component to obtain natural mode components with different frequencies, wherein the natural mode components comprise a low-frequency component and a high-frequency component; The data correction module is used for constructing a preset PSO-BP prediction model, confirming an abnormal date based on the high-frequency component and correcting the remainder component of the abnormal date based on the preset PSO-BP prediction model; and the data output module is used for superposing the trend component, the periodic component and the corrected remainder component to obtain corrected load data.

Description

Industrial and commercial user load data cleaning method and device for energy storage safety Technical Field The invention belongs to the technical field of electric power data cleaning, and particularly relates to an energy storage safety-oriented industrial and commercial user load data cleaning method and device. Background With global energy transformation and the construction of a novel power system, industrial and commercial energy storage systems are used as core means for improving the flexibility of a power grid, realizing demand side response and reducing the electricity charge of the demand, and enter a large-scale application stage. The high-quality electricity load data is regarded as 'nutritional diet' of the intelligent 'brain' of the energy storage system, and the capacity configuration, the technical selection and the optimal control decision of all charging and discharging of the energy storage system are highly dependent on the accurate cognition of the load characteristics of the user side. The existing method for cleaning and restoring the load data of the power user is mainly based on a business rule primary screening or a traditional statistical evaluation model, partially researches that a single abnormal index judgment threshold value such as a missing value, a zero value, an instantaneous jump value and the like is set, logically filters an original acquisition sequence by utilizing a business experience library, and further researches that statistical characteristics of historical load distribution are analyzed by adopting a moving average method, a Z-score space distribution method or a simple clustering algorithm (such as K-means and DBSCAN), and static stripping is carried out on data points deviating from main distribution. The method has certain identification capability in the scene of processing single metering faults, terminal offline and other dominant bad numbers, and provides support for basic electricity data management. For industrial and commercial energy storage full life cycle operation and maintenance service oriented to 'source network charge storage' deep fusion, the prior art has the core defects of seriously threatening the safety and economic operation of an energy storage system, namely the energy storage configuration and selection risk caused by first and data reduction misalignment. The core power and capacity allocation of energy storage are highly dependent on the accurate analysis of the maximum demand and peak-valley characteristics of users, and if the traditional cleaning method can not accurately restore the false peak or the missing peak generated by ammeter faults or communication interruption, the excessive capacity allocation (investment waste) or insufficient capacity allocation (effective peak clipping) can be directly caused, and even the decision of selecting technical routes such as lithium batteries, flywheels or supercapacitors can be misled. Second, abnormal feature disturbances cause "learning distortion" of the energy storage control strategy. Modern energy storage control algorithms (e.g., reinforcement learning-based or model predictive control) require training strategies based on historical data. In the prior art, atypical modes such as holiday shutdown, equipment failure and the like are difficult to strip from basic rules, and if a control strategy 'absorbs' the pollution data, the algorithm is subjected to insufficient charging or overdischarging in normal working days and is seriously deviated from the economic expectations of peak Gu Jiacha arbitrage and demand conservation. Thirdly, high-frequency noise residues cause potential safety hazards of the energy storage system. For industrial and commercial energy storage involving frequency modulation or second-level response, the data cleaning index focuses on the surface point values, and deep structural features in the sequence evolution process cannot be extracted. The unwashed power data noise can accumulate into huge battery cell state of charge estimation error, which is very easy to cause battery overcharge and overdischarge, thereby causing serious safety accidents such as thermal runaway and the like. Therefore, it is needed to develop a data cleaning method capable of deeply decomposing load components, accurately identifying multidimensional anomalies and realizing high-fidelity reduction, and providing data support of high nutrition and high purity for safe operation and accurate decision of industrial and commercial energy storage systems. Disclosure of Invention The invention aims to provide an energy storage safety-oriented industrial and commercial user load data cleaning method and device, which are used for solving the problem that data recovery misalignment is easy to occur when large-scale data processing is carried out on electric load data cleaning in the prior art. The technical scheme of the invention for achieving the purpose is as follows: in a first a