CN117591942B - Power load data anomaly detection method, system, medium and equipment - Google Patents

Power load data anomaly detection method, system, medium and equipment Download PDF

Info

Publication number
CN117591942B
CN117591942B CN202410069428.5A CN202410069428A CN117591942B CN 117591942 B CN117591942 B CN 117591942B CN 202410069428 A CN202410069428 A CN 202410069428A CN 117591942 B CN117591942 B CN 117591942B
Authority
CN
China
Prior art keywords
data
training
load data
layer
load
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202410069428.5A
Other languages
Chinese (zh)
Other versions
CN117591942A (en
Inventor
李萌
朱峰
陈云龙
刘继彦
王者龙
吴雪霞
张雪梅
刘昳娟
石雨帆
许帅
于相洁
王倩
李静
徐美玲
侯燕文
王若晗
高玉华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Marketing Service Center of State Grid Shandong Electric Power Co Ltd
Original Assignee
Marketing Service Center of State Grid Shandong Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Marketing Service Center of State Grid Shandong Electric Power Co Ltd filed Critical Marketing Service Center of State Grid Shandong Electric Power Co Ltd
Priority to CN202410069428.5A priority Critical patent/CN117591942B/en
Publication of CN117591942A publication Critical patent/CN117591942A/en
Application granted granted Critical
Publication of CN117591942B publication Critical patent/CN117591942B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computing Systems (AREA)
  • Economics (AREA)
  • Water Supply & Treatment (AREA)
  • Databases & Information Systems (AREA)
  • Public Health (AREA)
  • Fuzzy Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Testing Electric Properties And Detecting Electric Faults (AREA)

Abstract

The invention belongs to the technical field of data detection, and provides a method, a system, a medium and equipment for detecting power load data abnormality, wherein the technical scheme is as follows: training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training to obtain a fitting value, otherwise, continuing the training to obtain a trained load data reconstruction model; and predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is abnormal data according to the predicted value, if so, replacing the abnormal value with the predicted value, otherwise, judging the data to be normal data. The long-term dependency relationship between the power utilization sequences is captured through a bidirectional LSTM model, and the self-encoder is utilized for data reconstruction and feature extraction, so that the identification of abnormal data is realized.

Description

Power load data anomaly detection method, system, medium and equipment
Technical Field
The invention belongs to the technical field of data detection, and particularly relates to a method, a system, a medium and equipment for detecting power consumption load data abnormality.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
The power load data is one of the data which is widely applied, and plays an irreplaceable role in the aspects of load regulation, power grid operation and the like. However, load data may be abnormal for various reasons, which causes the cleaning of power load data to be a common problem faced in the fields of load management and the like. In addition, the load data has a timing characteristic, and it is difficult for the conventional method to efficiently extract the timing characteristic thereof.
In recent years, with the rapid development of artificial intelligence, data-driven anomaly detection methods are rapidly developed on the basis of mass data. Currently, various techniques have been applied to solve the problem of anomaly detection. For example, chouderet et al have built an automatic monitoring model to capture loss anomalies by analyzing fault signals and current-to-voltage ratios. The data-driven modeling method is divided into three strategies of regression, classification and clustering. Fan et al propose a classification-based system for anomaly detection of building electricity data that uses an automatic encoder to classify the data. In addition, zhang et al propose a linear regression anomaly detector that considers the effect of temperature on household power consumption and creates a different linear model that compares the predicted outcome of the model with the actual power consumption data. If the actual power consumption is significantly lower than the reference value, it is marked as abnormal. Furthermore, wang et al applied different clustering algorithms to detect 10kV non-technical losses and analyzed and compared performance.
In practical application, by adopting the method, commercial load data anomaly detection faces the following technical problems:
(1) Due to the complexity and variability of the load data, conventional algorithms may not accurately identify true outliers. For example, some algorithms may be too sensitive, misjudging normal fluctuations as anomalies, or too conservative to capture true anomalies;
(2) Conventional anomaly detection algorithms typically require a significant amount of computing resources and time to process large-scale load data, resulting in limited real-time;
(3) Commercial load data is often characterized by high dimensionality and large scale, such as various sensor data in electrical power systems. Conventional anomaly detection algorithms may be difficult to process such large-scale data sets because they may be limited by computation and storage;
(4) Conventional anomaly detection algorithms are typically built based on static models or assumptions and cannot accommodate changes in data distribution and new anomaly patterns.
Disclosure of Invention
In order to solve at least one technical problem in the background technology, the invention provides a method, a system, a medium and equipment for detecting the abnormality of electricity load data, which combine a bidirectional LSTM and a self-encoder to construct a new algorithm for detecting the abnormality of commercial electricity load data, capture the long-term dependency relationship between electricity sequences through a bidirectional LSTM model, and utilize the self-encoder to reconstruct data and extract characteristics so as to realize the identification of the abnormal data.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
The first aspect of the present invention provides a method for detecting abnormality of electrical load data, comprising the steps of:
obtaining electricity consumption data;
Training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training to obtain a fitting value, otherwise, continuing the training to obtain a trained load data reconstruction model;
And predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is abnormal data according to the predicted value, if so, replacing the abnormal value with the predicted value, otherwise, judging the data to be normal data.
Further, the construction process of the load data reconstruction model is as follows: combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, obtaining a potential electricity utilization rule of a user load curve based on forward and backward dependency relation of bidirectional LSTM learning electricity utilization load data, and extracting electricity utilization peak characteristics; and (3) encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristics, decoding at a characteristic layer, and carrying out data reconstruction on the extracted power consumption peak characteristics at the characteristic layer to obtain the reconstructed power consumption data.
Further, the learning forward and backward dependency relationship of the electrical load data based on the bidirectional LSTM includes:
In the Forward layer, forward computation is carried out from time 1 to time t, the output of each Forward hidden layer is stored, in the Backward layer, reverse computation is carried out from time t to time 1, the output of each Backward hidden layer is stored, and in each time, the results of the output of the Forward layer and the output of the Backward layer at corresponding time are combined, so that the final output is calculated.
Further, the adaptive coding AE includes an encoder and a decoder, and the reconstruction error is obtained by the encoding of the encoder and the decoding of the decoder, including: the encoder maps the input vector to a non-linear hidden state representation by affine transformation and the decoder maps the hidden state representation back to the original input space by inverse transformation of the encoder, the difference between the two being called reconstruction error.
Further, when the load data reconstruction model is trained, the method comprises the following steps: the power consumption data with noise is input into an LSTM coder for training, the coding result of the coder is put into the LSTM decoder for decoding, the decoding output is activated through a ReLU activation layer, the nonlinear relation among the neural networks is enhanced, and then the data size of the LSTM input is changed through a full connection layer, so that the data size identical to the data size of the input is obtained.
Further, in the training process of the load data reconstruction model, an Adam optimizer is used for optimizing and adjusting parameters, and after each round of training is finished, the learning rate is finely adjusted.
Further, in the training process of the load data reconstruction model, the method further comprises the following steps: acquiring electricity consumption data, carrying out normalization processing on the electricity consumption data, and distributing the normalized electricity consumption data to generate a test set and a training set;
Or, taking the data which are obviously outliers as error data, deleting the abnormal data, taking the rest data as normal electricity consumption data, and then marking the rest data partially, wherein the rest data comprises the normal electricity consumption data and the abnormal electricity consumption data, and marking the rest data partially to obtain the data tag.
A second aspect of the present invention provides an electrical load data abnormality detection system including:
the data acquisition module is used for acquiring electricity consumption data;
The load data reconstruction model training module is used for training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing the feature layers, judging whether the change of the reconstruction error is smaller than a set threshold value, ending the training when the reconstruction error tends to be stable if the change of the reconstruction error is smaller than the set threshold value, obtaining a fitting value, otherwise continuing the training, and obtaining a trained load data reconstruction model;
The anomaly detection module is used for predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is anomaly data according to the predicted value, if so, replacing the anomaly value with the predicted value, otherwise, obtaining normal data.
A third aspect of the present invention provides a computer-readable storage medium.
A computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in a method for detecting an electrical load data abnormality as described in the first aspect:
obtaining electricity consumption data;
Training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training to obtain a fitting value, otherwise, continuing the training to obtain a trained load data reconstruction model;
And predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is abnormal data according to the predicted value, if so, replacing the abnormal value with the predicted value, otherwise, judging the data to be normal data.
Further, the construction process of the load data reconstruction model is as follows: combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, obtaining a potential electricity utilization rule of a user load curve based on forward and backward dependency relation of bidirectional LSTM learning electricity utilization load data, and extracting electricity utilization peak characteristics; and (3) encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristics, decoding at a characteristic layer, and carrying out data reconstruction on the extracted power consumption peak characteristics at the characteristic layer to obtain the reconstructed power consumption data.
Further, the learning forward and backward dependency relationship of the electrical load data based on the bidirectional LSTM includes:
In the Forward layer, forward computation is carried out from time 1 to time t, the output of each Forward hidden layer is stored, in the Backward layer, reverse computation is carried out from time t to time 1, the output of each Backward hidden layer is stored, and in each time, the results of the output of the Forward layer and the output of the Backward layer at corresponding time are combined, so that the final output is calculated.
Further, the adaptive coding AE includes an encoder and a decoder, and the reconstruction error is obtained by the encoding of the encoder and the decoding of the decoder, including: the encoder maps the input vector to a non-linear hidden state representation by affine transformation and the decoder maps the hidden state representation back to the original input space by inverse transformation of the encoder, the difference between the two being called reconstruction error.
Further, when the load data reconstruction model is trained, the method comprises the following steps: the power consumption data with noise is input into an LSTM coder for training, the coding result of the coder is put into the LSTM decoder for decoding, the decoding output is activated through a ReLU activation layer, the nonlinear relation among the neural networks is enhanced, and then the data size of the LSTM input is changed through a full connection layer, so that the data size identical to the data size of the input is obtained.
Further, in the training process of the load data reconstruction model, an Adam optimizer is used for optimizing and adjusting parameters, and after each round of training is finished, the learning rate is finely adjusted.
Further, in the training process of the load data reconstruction model, the method further comprises the following steps: acquiring electricity consumption data, carrying out normalization processing on the electricity consumption data, and distributing the normalized electricity consumption data to generate a test set and a training set;
Or, taking the data which are obviously outliers as error data, deleting the abnormal data, taking the rest data as normal electricity consumption data, and then marking the rest data partially, wherein the rest data comprises the normal electricity consumption data and the abnormal electricity consumption data, and marking the rest data partially to obtain the data tag.
A fourth aspect of the invention provides a computer device.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in a method of detecting electrical load data anomalies as described in the first aspect when the program is executed:
obtaining electricity consumption data;
Training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training to obtain a fitting value, otherwise, continuing the training to obtain a trained load data reconstruction model;
And predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is abnormal data according to the predicted value, if so, replacing the abnormal value with the predicted value, otherwise, judging the data to be normal data.
Further, the construction process of the load data reconstruction model is as follows: combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, obtaining a potential electricity utilization rule of a user load curve based on forward and backward dependency relation of bidirectional LSTM learning electricity utilization load data, and extracting electricity utilization peak characteristics; and (3) encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristics, decoding at a characteristic layer, and carrying out data reconstruction on the extracted power consumption peak characteristics at the characteristic layer to obtain the reconstructed power consumption data.
Further, the learning forward and backward dependency relationship of the electrical load data based on the bidirectional LSTM includes:
In the Forward layer, forward computation is carried out from time 1 to time t, the output of each Forward hidden layer is stored, in the Backward layer, reverse computation is carried out from time t to time 1, the output of each Backward hidden layer is stored, and in each time, the results of the output of the Forward layer and the output of the Backward layer at corresponding time are combined, so that the final output is calculated.
Further, the adaptive coding AE includes an encoder and a decoder, and the reconstruction error is obtained by the encoding of the encoder and the decoding of the decoder, including: the encoder maps the input vector to a non-linear hidden state representation by affine transformation and the decoder maps the hidden state representation back to the original input space by inverse transformation of the encoder, the difference between the two being called reconstruction error.
Further, when the load data reconstruction model is trained, the method comprises the following steps: the power consumption data with noise is input into an LSTM coder for training, the coding result of the coder is put into the LSTM decoder for decoding, the decoding output is activated through a ReLU activation layer, the nonlinear relation among the neural networks is enhanced, and then the data size of the LSTM input is changed through a full connection layer, so that the data size identical to the data size of the input is obtained.
Further, in the training process of the load data reconstruction model, an Adam optimizer is used for optimizing and adjusting parameters, and after each round of training is finished, the learning rate is finely adjusted.
Further, in the training process of the load data reconstruction model, the method further comprises the following steps: acquiring electricity consumption data, carrying out normalization processing on the electricity consumption data, and distributing the normalized electricity consumption data to generate a test set and a training set;
Or, taking the data which are obviously outliers as error data, deleting the abnormal data, taking the rest data as normal electricity consumption data, and then marking the rest data partially, wherein the rest data comprises the normal electricity consumption data and the abnormal electricity consumption data, and marking the rest data partially to obtain the data tag.
Compared with the prior art, the invention has the beneficial effects that:
1. The present invention combines bi-directional LSTM with a self-encoder: the bidirectional LSTM and the self-encoder are combined to construct a new algorithm for detecting the abnormality of commercial load data, the long-term dependency relationship between power utilization sequences is captured through the bidirectional LSTM model, and the self-encoder is utilized for data reconstruction and feature extraction, so that the identification of the abnormal data is realized.
2. Consider the timing characteristics: aiming at the problem that commercial load data has time sequence characteristics, the algorithm can perform time sequence modeling and analysis on the user load data, and better understand information before and after different moments, so that the accuracy of anomaly detection is improved.
3. And (3) self-adaptive feature extraction: the algorithm does not need a large number of electricity stealing samples for training, adaptively extracts characteristics according to a user history load sequence, and judges the abnormal moment of data through reconstruction errors. The self-adaptive feature extraction mode can be better adapted to the load features of different users, and the flexibility and the accuracy of anomaly detection are improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
Fig. 1 is a flowchart of an overall method for detecting an abnormality of electrical load data according to an embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Aiming at the technical problems of accuracy, real-time performance, expandability, self-adaptability and the like faced by the abnormal detection of commercial load data.
The whole idea of the invention is as follows:
(1) Combining bi-directional LSTM and self-encoder: a novel algorithm is constructed for anomaly detection of commercial load data by combining a bidirectional LSTM with a self-encoder. The long-term dependency relationship between the power utilization sequences is captured through a bidirectional LSTM model, and the self-encoder is utilized for data reconstruction and feature extraction, so that the identification of abnormal data is realized.
(2) Consider the timing characteristics: aiming at the problem that commercial load data has time sequence characteristics, the algorithm can perform time sequence modeling and analysis on the user load data, and better understand information before and after different moments, so that the accuracy of anomaly detection is improved.
(3) And (3) self-adaptive feature extraction: the algorithm does not need a large number of electricity stealing samples for training, adaptively extracts characteristics according to a user history load sequence, and judges the abnormal moment of data through reconstruction errors. The self-adaptive feature extraction mode can be better adapted to the load features of different users, and the flexibility and the accuracy of anomaly detection are improved.
The load data reconstruction model constructed by the invention is based on forward and backward dependency relationship of bidirectional LSTM learning power consumption load data by combining the characteristics of bidirectional LSTM and self-adaptive coding AE, obtains potential power consumption rules of a user load curve and extracts power consumption peak characteristics; and (3) encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristic, decoding at a characteristic layer, and carrying out data reconstruction on the extracted power consumption peak characteristic at the characteristic layer to obtain reconstructed power consumption data, wherein the reconstructed power consumption data is recorded as a Bi-LSTM-AE model in the embodiment of the invention.
Example 1
Referring to fig. 1, the embodiment provides a method for detecting abnormality of power load data, which includes the following steps:
step 1: obtaining electricity consumption data;
step 2: carrying out normalization processing on the electricity consumption data;
Step 3: distributing the normalized power consumption data to generate a test set and a training set;
Step 4: training the constructed load data reconstruction model and fitting the data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training when the reconstruction error tends to be stable, obtaining a fitting value, otherwise, continuing the training;
Step 5: and predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is abnormal data according to the predicted value, if so, replacing the abnormal value with the predicted value, otherwise, judging the data to be normal data.
In this embodiment, in step 1, the electricity consumption data includes unpublished historical electricity consumption data, which are derived from a certain electric power grid company. The primary load data is daily electricity data for a mall and commercial building, collected every 15 minutes, and the total data set time ranges from 2021, 1, to 2022, 10.
In step 2, considering the sensitivity of the Bi-LSTM-AE model to different data, the power consumption data needs to be normalized, where the data is normalized first, the maximum max and the minimum min in the time-series power consumption data are obtained, then the difference between max and min is divided by min subtracted from each data, that is, (data-min)/(max-min), thereby normalizing the data to interval [0,1] for the processing of the subsequent model, then dividing the data into data blocks in 7 days, for example, and encoding the weather condition, season, holiday information using a Transformer network, encoding into vector form, then splicing according to channel dimension (for example, encoding weather, season, holiday into threeThen stitch intoAs additional anomaly detection data. And marking normal electricity consumption data and partial abnormal electricity consumption data, and taking the marked data as training samples.
In the embodiment, the data is normalized by adopting a max-min scaling method, and the method is as follows:
In the above, the ratio of/> Represents the maximum in the dataset, represents/>Minimum in the dataset.
In step 4, training the constructed load data reconstruction model based on the electricity consumption data, and automatically extracting features and reconstructing a feature layer, wherein the method comprises the following steps: inputting the preprocessed historical electricity data into a Bi-LSTM-AE model;
The construction process of the Bi-LSTM-AE model comprises the following steps: and combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, learning the potential electricity utilization rule of the user load curve based on the bidirectional LSTM, automatically extracting the electricity consumption peak characteristic, and carrying out data reconstruction on the characteristic layer by utilizing the extracted electricity consumption peak characteristic to obtain the reconstructed electricity utilization data.
First, the use of bi-directional LSTM helps to capture forward and backward dependencies in time series data (electricity usage load data) and improve the ability to extract potential law features, learn the potential electricity usage laws of the user load curve, and automatically extract electricity usage spike features, such as customer electricity usage behavior.
And then, the self-Adaptive Encoder (AE) is utilized to allow the data to be encoded, the peak characteristics are reserved, the characteristic layer is decoded to be close to the original data, the characteristics of noise and unnecessary information are obviously reduced, the extracted power consumption peak characteristics are subjected to data reconstruction at the characteristic layer, and the reconstructed power consumption data are obtained.
Since the bidirectional LSTM is a cyclic neural network capable of well processing time-series data, long-term dependency and sequence patterns in the time-series data can be captured. The adaptive coding AE can learn potential representations of the input data so that features in the time series can be better characterized.
Among them, long short-term memory neural network (LSTM) was first proposed by Hochreiter and Schmidhuber as improved versions of recurrent neural network (Recurrent Neural Network, RNN). LSTM significantly alleviates the gradient explosion and gradient vanishing problems in RNNs and is able to maintain long-term memory information during training. LSTM has a gating structure, and can selectively add and delete information of cell states, so that the problems of gradient disappearance and gradient explosion are avoided. The LSTM nerve unit includes three "gate" structures, namely a forgetting gate, an input gate, and an output gate. The specific structure is as follows: the calculation expression at time t is as follows:
The input feature vector is an input feature vector at time t in the same time window, and represents a hidden state vector and a neuron state vector of input and output at time t. At that time, no existence exists. These values are passed in the same layer and are output as LSTM layers. The weight matrix and the weight matrix respectively corresponding to the forgetting gate, the input gate, the hidden state vector and the output gate are combined into a weight matrix for the sake of simplicity. Is a Sigmoid function.
Bidirectional LSTM is an improvement over LSTM in that its hidden layer is composed of two parts, the forward LSTM cell state and the reverse LSTM cell state. The bidirectional LSTM, i.e. one LSTM unit processes forward input and the other one processes reverse input, so that both past information and future information can be retained, and the network can better understand the information before and after a certain moment.
The principle of bi-directional LSTM is as follows:
In Forward layer, forward computation is performed from time 1to time t, and the output of each time Forward hidden layer is saved. In the Backward layer, the Backward calculation is performed from time t to time 1, and the output of the Backward hidden layer of each time is saved.
Finally, at each moment, combining the results output by the Forward layer and the Backward layer at the corresponding moment, and calculating the final output, wherein the mathematical expression is as follows:
Principle analysis of significant effect of load data reconstruction model:
The self-encoder (Autoencoder, AE) uses an unsupervised learning approach, the encoder maps the input vector to a non-linear hidden state representation by affine transformation, the decoder maps the hidden state representation back to the original input space by inverse transformation of the encoder, the difference between the two is called reconstruction error, and is typically calculated using a norm. The learning objective of the self-encoder is to minimize this reconstruction error. The self-encoder based anomaly detection is a bias-based anomaly detection method that uses reconstruction errors as a standard, with data points of high reconstruction errors being considered anomalies.
Nonlinear feature extraction: both the bidirectional LSTM and the adaptive coding AE have strong nonlinear feature extraction capability. The bidirectional LSTM can learn complex sequence modes and features through a gating mechanism, and the self-adaptive coding AE can learn high-order feature representation of data through a multi-layer coding and decoding process.
Data reconstruction and feature learning: the self-adaptive coding AE has the capability of learning the internal structure and characteristics of the data, the bidirectional LSTM can learn the complex characteristics of the time sequence data, and the combined model can realize the reconstruction of the data and the learning of the characteristics through the self-encoder, so that the information in the time sequence data can be effectively extracted and represented.
Therefore, the construction process of the load data reconstruction model specifically comprises the following steps:
Firstly inputting noisy data into an LSTM coder for training, putting the coding result of the coder into the LSTM decoder for decoding, activating the decoding output through a ReLU activating layer, enhancing the nonlinear relation between the neural networks, and then changing the data size of the LSTM input through a full connection layer so as to obtain the data size identical to the input data.
In the training process of the network, a Adam (AdaptiveMomentEstimation) optimizer is used for optimizing and adjusting parameters, after each round of training is finished, the learning rate is finely adjusted, and after all training rounds are exhausted, the LSTM-AE training is finished. The model structure utilizes the characteristics of the self-encoder, namely the dimension reduction encoding can be carried out on the data, and then the reconstruction is carried out, so that the relation between different characteristic data at the same moment is explored.
In addition, by analyzing the error value after reconstruction, the abnormal data is preliminarily screened, and the data determined to be abnormal is repaired.
In the training process, the error value after reconstruction is analyzed, the abnormal data is primarily screened, the data judged to be abnormal is repaired, and when the reconstruction error tends to be stable, the training is ended, so that the fitting value is obtained.
In step 5, data is input into a Bi-LSTM-AE network model, where i=1, 2, …, n, and a predicted value is obtained.
Once the implementation of the first two-part technique is completed, a predicted value can be calculatedAnd actual value/>Absolute error/>. Then, a certain threshold/>, can be selected(E.g./>)If the error exceeds the threshold, it is considered as outlier data), the outlier data is replaced with the predicted value of the Bi-LSTM-AE model.
The abnormal data refers to the abnormal data which is obviously outlier data in the normalized data obtained in the same industry/industry, and the abnormal data is deleted, so that the rest data is normal power consumption data as much as possible, and the process can be regarded as a simple data preprocessing process. And then, marking the rest data partially, wherein the rest data comprises normal electricity consumption data and abnormal electricity consumption data, and marking the data partially to obtain data labels (abnormal labels and normal labels).
Effect verification
The invention adopts the simulation data to simulate the electricity stealing of the user to verify the feasibility of the proposed method, and then adopts the actual data to prove the reliability of the proposed method. The Bi-LSTM-AE model used was configured as shown in Table 1.
TABLE 1 configuration Table
The experiment uses load data of four commercial buildings of 2021, 2 office buildings and 2 markets for verification. All data sets are real data sets with manual labeling, the load data sampling interval is 15 minutes, and the sampling number per day is 96 points.
Precision, recall and F1-score are used to evaluate the algorithm. The algorithm is compared as shown in table 2. The precision is the ratio of correctly predicted samples to the total number of samples; the recall is the ratio of the number of samples correctly predicted as positive samples to the number of all actual positive samples; the F1 score is a harmonic mean of precision and recall, both of which can be considered comprehensively.
The accuracy was evaluated using an F1-score, the formula of which was calculated as:
In the above, the ratio of/> For the accuracy/>Is the recall rate.
The results indicate that Bi-LSTM-AE is able to generate the best overall dataset. Although some anomalies can be accurately identified using AE, the accuracy is not high. In contrast, the method using Bi-LSTM-AE can accurately identify most anomalies.
Table 2 algorithm result comparison
As shown in Table 2, the Bi-LSTM-AE model of the invention has an average 50% improvement over the conventional AE model in terms of abnormality detection accuracy (prediction accuracy) and an average 437% improvement over Boxplot model; the recall rate (detection accuracy) exceeds 108% of AE, boxplot; f1-score (the average of both reconcilations) exceeded 128% of AE, box-plot 498%. This shows that the model is far superior to the conventional algorithm in anomaly detection and data prediction.
Example two
The embodiment provides a power consumption load data abnormality detection system, including:
the data acquisition module is used for acquiring electricity consumption data;
The data preprocessing module is used for acquiring power consumption data, carrying out normalization processing on the power consumption data, and distributing the normalized power consumption data to generate a test set and a training set;
The load data reconstruction model training module is used for training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing the feature layers, judging whether the change of the reconstruction error is smaller than a set threshold value, ending the training when the reconstruction error tends to be stable if the change of the reconstruction error is smaller than the set threshold value, obtaining a fitting value, otherwise continuing the training, and obtaining a trained load data reconstruction model;
The anomaly detection module is used for predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is anomaly data according to the predicted value, if so, replacing the anomaly value with the predicted value, otherwise, obtaining normal data.
The data preprocessing module is further used for regarding the data which are obviously outliers as error data, deleting the abnormal data, taking the rest data as normal electricity consumption data, and then marking the rest data partially, wherein the rest data comprises the normal electricity consumption data and the abnormal electricity consumption data, and marking the rest data partially to obtain the data tag.
In the load data reconstruction model training module, the construction process of the load data reconstruction model is as follows: combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, obtaining a potential electricity utilization rule of a user load curve based on forward and backward dependency relation of bidirectional LSTM learning electricity utilization load data, and extracting electricity utilization peak characteristics; and (3) encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristics, decoding at a characteristic layer, and carrying out data reconstruction on the extracted power consumption peak characteristics at the characteristic layer to obtain the reconstructed power consumption data.
The forward and backward dependency relationship based on the bidirectional LSTM learning power consumption load data comprises the following steps:
In the Forward layer, forward computation is carried out from time 1 to time t, the output of each Forward hidden layer is stored, in the Backward layer, reverse computation is carried out from time t to time 1, the output of each Backward hidden layer is stored, and in each time, the results of the output of the Forward layer and the output of the Backward layer at corresponding time are combined, so that the final output is calculated.
The adaptive coding AE includes an encoder and a decoder, and the reconstruction error is obtained by the encoding of the encoder and the decoding of the decoder, including: the encoder maps the input vector to a non-linear hidden state representation by affine transformation and the decoder maps the hidden state representation back to the original input space by inverse transformation of the encoder, the difference between the two being called reconstruction error.
The load data reconstruction model training comprises the following steps: the power consumption data with noise is input into an LSTM coder for training, the coding result of the coder is put into the LSTM decoder for decoding, the decoding output is activated through a ReLU activation layer, the nonlinear relation among the neural networks is enhanced, and then the data size of the LSTM input is changed through a full connection layer, so that the data size identical to the data size of the input is obtained.
In the training process of the load data reconstruction model, an Adam optimizer is used for optimizing and adjusting parameters, and after each round of training is finished, the learning rate is subjected to fine adjustment.
Example III
The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in a power load data abnormality detection method according to the first embodiment, including:
obtaining electricity consumption data;
Training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training to obtain a fitting value, otherwise, continuing the training to obtain a trained load data reconstruction model;
And predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is abnormal data according to the predicted value, if so, replacing the abnormal value with the predicted value, otherwise, judging the data to be normal data.
The construction process of the load data reconstruction model comprises the following steps: combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, obtaining a potential electricity utilization rule of a user load curve based on forward and backward dependency relation of bidirectional LSTM learning electricity utilization load data, and extracting electricity utilization peak characteristics; and (3) encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristics, decoding at a characteristic layer, and carrying out data reconstruction on the extracted power consumption peak characteristics at the characteristic layer to obtain the reconstructed power consumption data.
Further, the learning forward and backward dependency relationship of the electrical load data based on the bidirectional LSTM includes:
In the Forward layer, forward computation is carried out from time 1 to time t, the output of each Forward hidden layer is stored, in the Backward layer, reverse computation is carried out from time t to time 1, the output of each Backward hidden layer is stored, and in each time, the results of the output of the Forward layer and the output of the Backward layer at corresponding time are combined, so that the final output is calculated.
The adaptive coding AE includes an encoder and a decoder, and the reconstruction error is obtained by the encoding of the encoder and the decoding of the decoder, including: the encoder maps the input vector to a non-linear hidden state representation by affine transformation and the decoder maps the hidden state representation back to the original input space by inverse transformation of the encoder, the difference between the two being called reconstruction error.
The load data reconstruction model training process comprises the following steps: the power consumption data with noise is input into an LSTM coder for training, the coding result of the coder is put into the LSTM decoder for decoding, the decoding output is activated through a ReLU activation layer, the nonlinear relation among the neural networks is enhanced, and then the data size of the LSTM input is changed through a full connection layer, so that the data size identical to the data size of the input is obtained.
In the training process of the load data reconstruction model, an Adam optimizer is used for optimizing and adjusting parameters, and after each round of training is finished, the learning rate is subjected to fine adjustment.
In the training process of the load data reconstruction model, the method further comprises the following steps: acquiring electricity consumption data, carrying out normalization processing on the electricity consumption data, and distributing the normalized electricity consumption data to generate a test set and a training set;
Or, taking the data which are obviously outliers as error data, deleting the abnormal data, taking the rest data as normal electricity consumption data, and then marking the rest data partially, wherein the rest data comprises the normal electricity consumption data and the abnormal electricity consumption data, and marking the rest data partially to obtain the data tag.
Example IV
The embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the steps in the electrical load data anomaly detection method according to the first embodiment, and the method includes:
obtaining electricity consumption data;
Training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training to obtain a fitting value, otherwise, continuing the training to obtain a trained load data reconstruction model;
And predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is abnormal data according to the predicted value, if so, replacing the abnormal value with the predicted value, otherwise, judging the data to be normal data.
The construction process of the load data reconstruction model comprises the following steps: combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, obtaining a potential electricity utilization rule of a user load curve based on forward and backward dependency relation of bidirectional LSTM learning electricity utilization load data, and extracting electricity utilization peak characteristics; and (3) encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristics, decoding at a characteristic layer, and carrying out data reconstruction on the extracted power consumption peak characteristics at the characteristic layer to obtain the reconstructed power consumption data.
Further, the learning forward and backward dependency relationship of the electrical load data based on the bidirectional LSTM includes:
In the Forward layer, forward computation is carried out from time 1 to time t, the output of each Forward hidden layer is stored, in the Backward layer, reverse computation is carried out from time t to time 1, the output of each Backward hidden layer is stored, and in each time, the results of the output of the Forward layer and the output of the Backward layer at corresponding time are combined, so that the final output is calculated.
The adaptive coding AE includes an encoder and a decoder, and the reconstruction error is obtained by the encoding of the encoder and the decoding of the decoder, including: the encoder maps the input vector to a non-linear hidden state representation by affine transformation and the decoder maps the hidden state representation back to the original input space by inverse transformation of the encoder, the difference between the two being called reconstruction error.
The load data reconstruction model training process comprises the following steps: the power consumption data with noise is input into an LSTM coder for training, the coding result of the coder is put into the LSTM decoder for decoding, the decoding output is activated through a ReLU activation layer, the nonlinear relation among the neural networks is enhanced, and then the data size of the LSTM input is changed through a full connection layer, so that the data size identical to the data size of the input is obtained.
In the training process of the load data reconstruction model, an Adam optimizer is used for optimizing and adjusting parameters, and after each round of training is finished, the learning rate is subjected to fine adjustment.
In the training process of the load data reconstruction model, the method further comprises the following steps: acquiring electricity consumption data, carrying out normalization processing on the electricity consumption data, and distributing the normalized electricity consumption data to generate a test set and a training set;
Or, taking the data which are obviously outliers as error data, deleting the abnormal data, taking the rest data as normal electricity consumption data, and then marking the rest data partially, wherein the rest data comprises the normal electricity consumption data and the abnormal electricity consumption data, and marking the rest data partially to obtain the data tag.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (6)

1. The power load data anomaly detection method is characterized by comprising the following steps:
obtaining electricity consumption data;
Training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing a feature layer, judging whether the change of a reconstruction error is smaller than a set threshold value, if so, ending the training to obtain a fitting value, otherwise, continuing the training to obtain a trained load data reconstruction model;
Predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the predicted value is abnormal data or not according to the predicted value, if so, replacing the abnormal data with the predicted value, otherwise, judging the abnormal data as normal data;
The construction process of the load data reconstruction model comprises the following steps: combining the characteristics of the bidirectional LSTM and the self-adaptive coding AE, obtaining a potential electricity utilization rule of a user load curve based on forward and backward dependency relation of bidirectional LSTM learning electricity utilization load data, and extracting electricity utilization peak characteristics; encoding the power consumption load data based on the self-adaptive encoding AE, reserving the power consumption peak characteristics, decoding at a characteristic layer, and reconstructing the data of the extracted power consumption peak characteristics at the characteristic layer to obtain reconstructed power consumption data;
The forward and backward dependency relationship based on the bidirectional LSTM learning power consumption load data comprises the following steps:
In the Forward layer, forward calculation is carried out from time 1 to time t, the output of each Forward hidden layer is stored, in the Backward layer, reverse calculation is carried out from time t to time 1, the output of each Backward hidden layer is stored, and at each time, the results of the output of the Forward layer and the output of the Backward layer at the corresponding time are combined, and the final output is calculated;
The adaptive coding AE includes an encoder and a decoder, and the reconstruction error is obtained by the encoding of the encoder and the decoding of the decoder, including: the encoder maps the input vector to a nonlinear hidden state representation through affine transformation, and the decoder maps the hidden state representation back to the original input space through inverse transformation of the encoder, wherein the difference between the two is called reconstruction error;
The load data reconstruction model training comprises the following steps: the power consumption data with noise is input into an LSTM coder for training, the coding result of the coder is put into the LSTM decoder for decoding, the decoding output is activated through a ReLU activation layer, the nonlinear relation among the neural networks is enhanced, and then the data size of the LSTM input is changed through a full connection layer, so that the data size identical to the data size of the input is obtained.
2. The method for detecting abnormal electricity load data according to claim 1, wherein in the training process of the load data reconstruction model, an Adam optimizer is used for optimizing and adjusting parameters, and after each round of training is finished, the learning rate is finely adjusted.
3. The method for detecting anomalies in electrical load data as recited in claim 1, further comprising, during training of the load data reconstruction model: acquiring electricity consumption data, carrying out normalization processing on the electricity consumption data, and distributing the normalized electricity consumption data to generate a test set and a training set;
Or, taking the data which are obviously outliers as error data, deleting the abnormal data, taking the rest data as normal electricity consumption data, and then marking the rest data partially, wherein the rest data comprises the normal electricity consumption data and the abnormal electricity consumption data, and marking the rest data partially to obtain the data tag.
4. An electrical load data abnormality detection system based on an electrical load data abnormality detection method according to any one of claims 1 to 3, characterized by comprising:
the data acquisition module is used for acquiring electricity consumption data;
The load data reconstruction model training module is used for training the constructed load data reconstruction model based on the power consumption data, automatically extracting features and reconstructing the feature layers, judging whether the change of the reconstruction error is smaller than a set threshold value, ending the training when the reconstruction error tends to be stable if the change of the reconstruction error is smaller than the set threshold value, obtaining a fitting value, otherwise continuing the training, and obtaining a trained load data reconstruction model;
The anomaly detection module is used for predicting based on the trained load data reconstruction model to obtain a predicted value, judging whether the data is anomaly data according to the predicted value, if so, replacing the anomaly data with the predicted value, otherwise, obtaining normal data.
5. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of a method for detecting an abnormality of electrical load data according to any one of claims 1 to 3.
6. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of a method for detecting an electrical load data anomaly as claimed in any one of claims 1 to 3 when the program is executed.
CN202410069428.5A 2024-01-18 2024-01-18 Power load data anomaly detection method, system, medium and equipment Active CN117591942B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410069428.5A CN117591942B (en) 2024-01-18 2024-01-18 Power load data anomaly detection method, system, medium and equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410069428.5A CN117591942B (en) 2024-01-18 2024-01-18 Power load data anomaly detection method, system, medium and equipment

Publications (2)

Publication Number Publication Date
CN117591942A CN117591942A (en) 2024-02-23
CN117591942B true CN117591942B (en) 2024-04-19

Family

ID=89916910

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410069428.5A Active CN117591942B (en) 2024-01-18 2024-01-18 Power load data anomaly detection method, system, medium and equipment

Country Status (1)

Country Link
CN (1) CN117591942B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114707698A (en) * 2022-03-03 2022-07-05 深圳智芯微电子科技有限公司 Long-term power load prediction method based on comprehensive shape error and time error
CN116108885A (en) * 2023-01-04 2023-05-12 上海电力大学 LSTM-AE-DT model-based medium-low pressure gas pressure regulator abnormality detection method
WO2023114121A1 (en) * 2021-12-13 2023-06-22 Mars, Incorporated A computer-implemented method of predicting quality of a food product sample
CN116993537A (en) * 2023-08-03 2023-11-03 浙江工商大学 Power load abnormality detection method and system based on serial GRU (generic routing unit) self-encoder

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023114121A1 (en) * 2021-12-13 2023-06-22 Mars, Incorporated A computer-implemented method of predicting quality of a food product sample
CN114707698A (en) * 2022-03-03 2022-07-05 深圳智芯微电子科技有限公司 Long-term power load prediction method based on comprehensive shape error and time error
CN116108885A (en) * 2023-01-04 2023-05-12 上海电力大学 LSTM-AE-DT model-based medium-low pressure gas pressure regulator abnormality detection method
CN116993537A (en) * 2023-08-03 2023-11-03 浙江工商大学 Power load abnormality detection method and system based on serial GRU (generic routing unit) self-encoder

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Anomaly Detection of an Air Compressor from Time-series Measurement Data;Kim, MJ等;《22nd International Conference on Control, Automation and Systems》;20221231;全文 *
基于集成深度学习算法的燃气负荷预测方法;王晓霞;徐晓钟;张彤;高超伟;;计算机系统应用;20191215(12);全文 *

Also Published As

Publication number Publication date
CN117591942A (en) 2024-02-23

Similar Documents

Publication Publication Date Title
CN110610280B (en) Short-term prediction method, model, device and system for power load
CN112784965B (en) Large-scale multi-element time series data anomaly detection method oriented to cloud environment
CN112148955B (en) Method and system for detecting abnormality of time sequence data of Internet of things
CN111914873A (en) Two-stage cloud server unsupervised anomaly prediction method
CN112990556A (en) User power consumption prediction method based on Prophet-LSTM model
CN113312447B (en) Semi-supervised log anomaly detection method based on probability label estimation
CN111241755A (en) Power load prediction method
CN111783684A (en) Bi-GRU-SA-based household electrical load decomposition method and system
CN112257263A (en) Equipment residual life prediction system based on self-attention mechanism
Legrand et al. Study of autoencoder neural networks for anomaly detection in connected buildings
CN112948155A (en) Model training method, state prediction method, device, equipment and storage medium
CN113284001A (en) Power consumption prediction method and device, computer equipment and storage medium
CN113762716A (en) Method and system for evaluating running state of transformer area based on deep learning and attention
CN111881299A (en) Outlier event detection and identification method based on duplicate neural network
CN117131022B (en) Heterogeneous data migration method of electric power information system
Wilfling et al. Learning non-linear white-box predictors: A use case in energy systems
CN113779879A (en) Medium-and-long-term electricity utilization abnormity detection method based on LSTM-seq2seq-attention model
CN117591942B (en) Power load data anomaly detection method, system, medium and equipment
CN116662866A (en) End-to-end incomplete time sequence classification method based on data interpolation and characterization learning
CN117154680A (en) Wind power prediction method based on non-stationary transducer model
CN115952928A (en) Short-term power load prediction method, device, equipment and storage medium
CN115330050A (en) Building load prediction method based on hybrid model
CN115766504A (en) Method for detecting cycle time sequence abnormity
CN115217152A (en) Method and device for predicting opening and closing deformation of immersed tunnel pipe joint
KR20230088967A (en) Anomaly detiection and repair based electrical load forecasting device and method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant