CN111047094A - Meter reading data anomaly analysis method based on deep learning algorithm - Google Patents
Meter reading data anomaly analysis method based on deep learning algorithm Download PDFInfo
- Publication number
- CN111047094A CN111047094A CN201911272670.8A CN201911272670A CN111047094A CN 111047094 A CN111047094 A CN 111047094A CN 201911272670 A CN201911272670 A CN 201911272670A CN 111047094 A CN111047094 A CN 111047094A
- Authority
- CN
- China
- Prior art keywords
- meter reading
- reading data
- abnormal
- user
- model
- 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.)
- Pending
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 53
- 238000004458 analytical method Methods 0.000 title claims abstract description 21
- 230000002159 abnormal effect Effects 0.000 claims abstract description 149
- 238000012549 training Methods 0.000 claims abstract description 76
- 238000000034 method Methods 0.000 claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 18
- 230000005611 electricity Effects 0.000 claims description 84
- 230000006870 function Effects 0.000 claims description 60
- 230000005856 abnormality Effects 0.000 claims description 29
- 230000004913 activation Effects 0.000 claims description 26
- 230000008569 process Effects 0.000 claims description 24
- 210000002569 neuron Anatomy 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 7
- 230000001174 ascending effect Effects 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 6
- 238000010801 machine learning Methods 0.000 claims description 4
- 230000002441 reversible effect Effects 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 4
- 238000001514 detection method Methods 0.000 abstract description 3
- 238000003908 quality control method Methods 0.000 abstract description 3
- 239000000463 material Substances 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 14
- 239000013598 vector Substances 0.000 description 14
- 210000004027 cell Anatomy 0.000 description 13
- 230000006399 behavior Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 238000012937 correction Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 230000010355 oscillation Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004171 remote diagnosis Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
-
- 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
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- 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)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Tourism & Hospitality (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a meter reading data abnormity analysis method based on a deep learning algorithm, and relates to the field of power consumer meter reading abnormity judgment methods. Conventionally, a field detection method is mostly adopted for processing abnormal opinion complaints of meter reading data, so that material resources and human resources are consumed, the efficiency is low, and the problems of single method, low accuracy and the like exist in the judgment of the error electric quantity of a determined abnormal meter reading data user. The method adopts a BP neural network subjected to deep learning training to establish a user meter reading data abnormal attribute judgment algorithm model and an optimized configuration strategy, realizes the rapid and accurate judgment of the work order user meter reading data abnormal attribute, and realizes the accurate estimation of error time and error electric quantity by establishing a meter reading data abnormal user daily electric quantity prediction model and an optimized configuration strategy. Therefore, the working efficiency of 95598 index analysis and quality control is improved, and an auxiliary decision making effect is played.
Description
Technical Field
The invention relates to the field of power consumer meter reading abnormity judgment methods, in particular to a meter reading data abnormity analysis method based on a deep learning algorithm.
Background
The method can quickly and accurately process the abnormal complaints of the meter reading data of the user, and the method not only relates to the economic benefit of the power company, but also relates to the service quality of the power company. The existing method for processing abnormal opinion complaints of meter reading data mostly adopts a field detection method, namely, a power grid technician checks a power utilization field, the processing mode extremely consumes material resources and human resources, the efficiency is low, the effect is poor, and the feedback of the processing result also has great human factors, so that the management of the power industry is not facilitated. Meanwhile, the problems of single method, low accuracy and the like exist in the judgment of the error electric quantity of the user with abnormal meter reading data.
Disclosure of Invention
The technical problem to be solved and the technical task provided by the invention are to perfect and improve the prior technical scheme, and provide a meter reading data abnormity analysis method based on a deep learning algorithm, so as to improve the accuracy rate of meter reading abnormity judgment and improve the judgment efficiency. Therefore, the invention adopts the following technical scheme.
A meter reading data anomaly analysis method based on a deep learning algorithm comprises the following steps:
1) performing feature extraction on a data set of a work order user with abnormal meter reading data;
2) reading a user meter reading data abnormal attribute and reality judgment algorithm model subjected to deep learning from a database;
3) judging whether the data is abnormal or not through the identification of an abnormal actual judgment algorithm model of the user meter reading data, if not, returning a judgment result, and if so, executing the next step;
4) according to the characteristics of daily electricity consumption data of users with abnormal meter reading data and the service requirement of error electricity consumption of users with abnormal meter reading data needing to be estimated, establishing a daily electricity consumption prediction algorithm model of users with abnormal meter reading data by adopting an LSTM deep learning algorithm, and designing an LTSM neural network model structure and model parameters;
5) acquiring a training data set of a daily electricity consumption prediction algorithm model of a user with abnormal meter reading data;
6) deep learning training is carried out on the daily electricity prediction algorithm model of the user with abnormal meter reading data, training sample data classification and normalization processing are completed, and deep learning training is completed through a random gradient descent optimization algorithm;
7) and calculating abnormal days and error electric quantity through a daily electric quantity prediction algorithm model of the user with abnormal meter reading data so as to judge and correct abnormal electricity consumption.
The abnormal attribute judgment of the meter reading data of the worksheet user is realized through the deeply learned abnormal attribute judgment algorithm model of the meter reading data of the user, the error time and the error electric quantity are accurately estimated through establishing the daily electric quantity prediction model of the abnormal meter reading user, the abnormal meter reading judgment accuracy and the judgment efficiency are effectively improved, and therefore the working efficiency of index analysis and quality control is improved.
As a preferable technical means: in the step 2), the establishment of the abnormal attribute and reality judgment algorithm model of the user meter reading data comprises the following steps:
201) defining learning model parameters including learning rate, regular parameters, iteration times and the number of neural network layers;
202) establishing an abnormal attribute and reality judgment algorithm model of the user meter reading data by feature extraction and adopting a BP neural network;
203) acquiring a latest training data set of an abnormal attribute and reality judgment algorithm model of user meter reading data;
204) performing machine learning training on the abnormal attribute and reality judgment algorithm model of the user meter reading data, and performing unified training on all parameters of the model by adopting a random gradient descent algorithm to adjust hidden layer parameters and finally obtain output layer parameters;
205) optimally configuring key parameters related to an abnormal attribute and reality judgment algorithm model of user meter reading data, and balancing model parameter setting by comparing a loss function descending curve trend and accuracy in the model training and testing processes;
206) and finishing the training and learning of the model, and writing the model back to the database.
The method has the advantages that the client meter reading data abnormal attribute and reality judging algorithm model is built, an effective client meter reading data abnormal attribute and reality judging algorithm model can be provided through deep learning training, and the user meter reading data abnormal attribute and reality can be quickly and accurately judged.
As a preferable technical means: the process of deep learning training of the user meter reading data abnormal user daily electricity consumption prediction algorithm model in the step 6) and the user meter reading data abnormal attribute judgment algorithm model in the step 204) comprises the following steps:
601) training data vectorization processing;
602) performing classified iterative learning;
603) judging whether the iteration times is less than the iteration times, if not, finishing model training and learning, and if so, executing the next step;
604) updating the weights and the offsets and returning to step 603).
And deep learning training is effectively realized.
As a preferable technical means: the prediction algorithm model for the daily electricity consumption of the user with abnormal meter reading data is provided with a 4-layer network structure, 5 neurons of an input layer are provided, the number of neurons of a hidden layer is 200, 100 and 100 respectively, an output target is 1, a tanh nonlinear function is selected as an activation function of the hidden layer, and an identity function is selected as an activation function of an output layer. And the LTSM neural network model structure and model parameters are effectively realized.
As a preferable technical means: in the step 6), the training sample data classification is to split daily electricity consumption data of the user, which is calibrated to be abnormal through the abnormal attribute judgment model of the user meter reading data, into two data sets of daily electricity consumption in holidays and daily electricity consumption in working days to form two time sequences, and each time sequence is independently modeled. The two data sets are established mainly in consideration of the fact that the large difference exists between the electricity behaviors of the holidays and the electricity behaviors of the working days, and the targeted training can be effectively achieved.
As a preferable technical means: in step 6), during normalization, the sequence data is normalized to [0, 1]]Within the scope, the transfer function is:
in the formula xiRepresenting the ith input value of the time series, x representing all inputs of the series, yiThe ith output value representing a time series. The normalization processing is mainly because the characteristics of the activation function sigmoid in the learning training process are considered, and the purpose is to reduce the time consumption and the resource occupation in the data training process.
As a preferable technical means: in step 7), the optimal configuration strategy needs to be repeatedly verified through multiple experimental tests to determine the value of each hyper-parameter. The training of the daily electricity consumption prediction model of the abnormal meter reading data user is completed within a certain time according to requirements on services, the final result is fed back, namely the abnormal days and the error electricity consumption of the abnormal meter reading data user are calculated, the optimal parameter configuration can be effectively realized by determining the value of each hyper-parameter through repeated verification, and better performance and accuracy are obtained.
As a preferable technical means: in step 202), the feature extraction content comprises the average daily power consumption per month, the ascending and descending trend of the daily power consumption, the occurrence frequency of various metering abnormalities and the average daily temperature of the grade city where the house number is located.
As a preferable technical means: in step 202), when modeling is performed by using a BP neural network, in the aspect of setting a BP deep learning algorithm structure, a 5-layer network structure is set, 45 input layer neurons are provided, the number of hidden layer neurons is 500, 200 and 50 respectively, an output target is 1, a tanh nonlinear function is selected as an activation function of a hidden layer, and a sigmoid function is selected as an activation function of an output layer. And the model structure and the model parameters of the BP neural network are effectively realized.
As a preferable technical means: the number of times of occurrence of the metering abnormality comprises 11 metering abnormalities, namely uneven electric energy representation value, flying away of the electric energy meter, backward walking of the electric energy meter, stopping of the electric energy meter, voltage phase failure, clock abnormality of the electric energy meter, reverse electric quantity abnormality, reversed trend, voltage loss, automatic checking abnormality and electric quantity fluctuation abnormality. Considering the influence of metering abnormality and the occurrence frequency of various metering abnormalities of users with abnormal meter reading data in nearly 6 months, 11 kinds of metering abnormalities are screened out as input characteristics.
As a preferable technical means: and updating the abnormal attribute and reality judgment model of the meter reading data periodically by adopting a mode of combining off-line learning and on-line learning. So as to maintain the timeliness and adaptability of the model.
Has the advantages that:
1. based on a deep learning technology, by establishing a scientific abnormal recognition and daily electricity quantity prediction model and researching the mathematical relationship of each index, the intelligent recognition of users with abnormal meter reading data and the scientific estimation of abnormal days and error electricity quantity of the meter reading data are realized, and the working efficiency of customer service index analysis and quality control is improved.
2. The method supports regular and real-time model training and correction, and the user meter reading data abnormal attribute and real-time judgment algorithm model which is trained well in an initial off-line mode can be automatically updated in a timed mode by newly-added supplementary data, so that the model can be better adapted to various conditions, and the model can be updated in real time according to needs. The daily electricity quantity prediction model of the user with abnormal meter reading data is trained and predicted in real time, namely after the abnormal user meter reading data is identified as abnormal by the aid of the judgment algorithm model, daily electricity quantity, daily maximum temperature and daily minimum temperature data of the user are obtained from the database in real time, model training and prediction are conducted, and the capacity of estimating abnormal days and error electricity quantity of the meter reading data is improved by means of the set correction strategy.
3. Machine real-time attribute study and judgment are carried out aiming at the abnormal work order complaints of sensitive and seedling-end meter reading data, the remote diagnosis and field verification work of electricity inspection is assisted, and metering meter reading complaints are reduced; meanwhile, aiming at the meter reading abnormal phenomenon which is not sensed by the customer, a large number of customer carpet type machine active studying and judging mechanisms are established, suspected abnormal meter reading users are actively identified and pushed to the electricity consumption acquisition system in real time, and the customer is supported to carry out active investigation on wrong meter reading and missed meter reading.
Drawings
FIG. 1 is a schematic flow diagram of the present invention.
FIG. 2 is a schematic diagram of a flow of implementing an abnormal attribute and reality determination algorithm model of user meter reading data.
FIG. 3 is a schematic diagram of deep learning training process of the user daily electricity quantity prediction algorithm model and the user meter reading data abnormity attribute judgment algorithm model of the abnormal meter reading data.
FIG. 4 is a schematic structural diagram of a prediction model of daily electricity consumption of users with abnormal meter reading data based on an LSTM deep learning algorithm.
FIG. 5 is a diagram showing the comparison result between different learning rates and loss functions in the daily electricity consumption model training process.
FIG. 6 is a diagram showing the comparison result between different learning rates and the accuracy rate in the daily electricity consumption model training process.
Fig. 7 is a schematic diagram of comparison results of different activation functions and accuracy rates in the daily electricity consumption model training process.
FIG. 8 is a schematic diagram of a model structure of an abnormal attribute and reality determination algorithm for user meter reading data according to the present invention.
FIG. 9 is a schematic diagram of comparison results of different learning rates and loss functions in the user meter reading data abnormal attribute and reality determination algorithm model training process.
FIG. 10 is a schematic diagram of comparison results of different learning rates and accuracy rates in a user meter reading data abnormal attribute and reality judging algorithm model training process.
FIG. 11 is a schematic diagram of a comparison result between different regularization parameters L2 and a loss function in the abnormal meter reading data model training process.
FIG. 12 is a schematic diagram of comparison results of different regular parameters L2 and accuracy in the user meter reading data abnormal attribute and reality judgment algorithm model training process.
FIG. 13 is a schematic diagram of comparison results of different activation functions and accuracy rates in a user meter reading data abnormal attribute and reality judgment algorithm model training process.
FIG. 14 is a schematic diagram of the structure of LSTM neural network cells.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
As shown in fig. 1, a meter reading data anomaly analysis method based on a deep learning algorithm includes the following steps:
1) performing feature extraction on a data set of a work order user with abnormal meter reading data;
2) reading a user meter reading data abnormal attribute and reality judgment algorithm model subjected to deep learning in a database;
3) judging whether the data is abnormal or not through the identification of an abnormal actual judgment algorithm model of the user meter reading data, if not, returning a judgment result, and if so, executing the next step;
4) according to the characteristics of daily electricity consumption data of users with abnormal meter reading data and the service requirement of error electricity consumption of users with abnormal meter reading data needing to be estimated, establishing a daily electricity consumption prediction algorithm model of users with abnormal meter reading data by adopting an LSTM deep learning algorithm, and designing an LTSM neural network model structure and model parameters;
5) acquiring a training data set of a daily electricity consumption prediction algorithm model of a user with abnormal meter reading data;
6) deep learning training is carried out on the daily electricity prediction algorithm model of the user with abnormal meter reading data, training sample data classification and normalization processing are completed, and deep learning training is completed through a random gradient descent optimization algorithm;
7) and calculating abnormal days and error electric quantity through a daily electric quantity prediction algorithm model of the user with abnormal meter reading data so as to judge and correct abnormal electricity consumption.
In order to realize the rapid and accurate determination of the abnormal attribute of the user meter reading data, as shown in fig. 2, in step 2), the establishment of the abnormal attribute determination algorithm model of the user meter reading data comprises the following steps:
201) defining learning model parameters including learning rate, regular parameters, iteration times and the number of neural network layers;
202) establishing an abnormal attribute and reality judgment algorithm model of the user meter reading data by feature extraction and adopting a BP neural network;
203) acquiring a latest training data set of an abnormal attribute and reality judgment algorithm model of user meter reading data;
204) performing machine learning training on the abnormal attribute and reality judgment algorithm model of the user meter reading data, and performing unified training on all parameters of the model by adopting a random gradient descent algorithm to adjust hidden layer parameters and finally obtain output layer parameters;
205) optimally configuring key parameters related to an abnormal attribute and reality judgment algorithm model of user meter reading data, and balancing model parameter setting by comparing a loss function descending curve trend and accuracy in the model training and testing processes;
206) and finishing the training and learning of the model, and writing the model back to the database.
The method has the advantages that the client meter reading data abnormal attribute and reality judging algorithm model is built, an effective client meter reading data abnormal attribute and reality judging algorithm model can be provided through deep learning training, and the user meter reading data abnormal attribute and reality can be quickly and accurately judged.
In order to realize deep learning training, as shown in fig. 3, the processes of deep learning training of the user daily electricity consumption prediction algorithm model with abnormal meter reading data in step 6) and the user meter reading data abnormal attribute judgment algorithm model in step 204) both include the following steps:
601) training data vectorization processing;
602) performing classified iterative learning;
603) judging whether the iteration times is less than the iteration times, if not, finishing model training and learning, and if so, executing the next step;
604) updating the weights and the offsets and returning to step 603).
And deep learning training is effectively realized.
In order to realize the LTSM neural network model structure, as shown in fig. 4, the daily electricity consumption prediction algorithm model for users with abnormal meter reading data is provided with a 4-layer network structure, 5 input layer neurons, 200, 100 and 100 hidden layer neurons, respectively, an output target is 1, a tanh nonlinear function is selected as an activation function of a hidden layer, and an identity function is selected as an activation function of an output layer. { x, y } denotes input variablesQuantity and output variable, where x ═ x1，x2，x3，x4，x5Denotes 5 characteristic values, x, of the input layer1Denotes the daily power, x, for t days2Denotes the maximum temperature, x, for t days3Denotes the minimum temperature, x, for t days4Denotes the maximum temperature, x, for t +1 days5The maximum temperature for t +1 day is indicated. And the LTSM neural network model structure and model parameters are effectively realized.
In order to realize training aiming at power consumption behavior differences at different times, in the step 6), training sample data classification is to divide daily power consumption data of a user which is marked as abnormal through a user meter reading data abnormal attribute judgment model into two data sets of daily power consumption in holidays and daily power consumption in working days to form two time sequences, and each time sequence is independently modeled. The two data sets are established mainly in consideration of the fact that the large difference exists between the electricity behaviors of the holidays and the electricity behaviors of the working days, and the targeted training can be effectively achieved.
In order to reduce the time consumption and the resource occupation in the data training process, in the step 6), during the normalization processing, the sequence data are normalized to [0, 1]]Within the scope, the transfer function is:
in the formula xiRepresenting the ith input value of the time series, x representing all inputs of the series, yiThe ith output value representing a time series. The normalization processing is mainly because the characteristics of the activation function sigmoid in the learning training process are considered, and the purpose is to reduce the time consumption and the resource occupation in the data training process.
In order to implement the optimal configuration strategy, in step 7), the optimal configuration strategy needs to be repeatedly verified through multiple experimental tests to determine the value of each hyper-parameter. The training of the daily electricity consumption prediction model of the abnormal meter reading data user is completed within a certain time according to requirements on services, the final result is fed back, namely the abnormal days and the error electricity consumption of the abnormal meter reading data user are calculated, the optimal parameter configuration can be effectively realized by determining the value of each hyper-parameter through repeated verification, and better performance and accuracy are obtained. According to repeated verification of experiments, in the example, the Batch size is set to be 50, the accuracy is relatively stable, and the regularization parameter is set to be 0.001. The momentum coefficient was set to 0.95 by default, and the specific model parameter configuration is shown in table 1 below:
table 1 abnormal user daily electricity consumption prediction algorithm model training parameters of meter reading data
As shown in fig. 5, the loss values at different learning rates generally tend to decrease first and eventually tend to stabilize. And if the Epoch value reaches about 50 and the loss value tends to be stable at each learning rate, taking the Epoch value as 50.
As shown in fig. 6, the accuracy rate at different learning rates first shows a growing trend, and after several epochs, the small oscillation tends to be stable. When the learning rate is 0.01, the accuracy is high, and the learning rate is set to 0.01.
As shown in FIG. 7, different hidden layer activation functions have a significant effect on the accuracy, and the lowest curve represents when the hidden layer activation function is sigThe accuracy corresponding to different epochs in the moid can be found to be always at a low level. The uppermost curve represents the accuracy corresponding to different epochs when the hidden layer activation function is set to tanh, and the accuracy is higher when the activation function is tanh, and therefore, the hidden layer activation function is set to tanh.
In step 202) for feature extraction, during feature extraction, the extracted content includes the average monthly electricity consumption, the ascending and descending trend of the monthly electricity consumption, the occurrence frequency of various metering anomalies, and the average monthly air temperature of the grade city where the household number is located, and it is assumed that the used data set is X ═ { X ═ X {nAnd N is 1,2, …, N, the data set comprises N daily power users, each user is divided into M months of power consumption, and K quantity abnormality is counted.
Because the randomness of the electricity utilization data of a single user is strong, whether the electricity utilization data are abnormal or not is judged by depending on the extracted single characteristic indexes of the electricity utilization user, and the accuracy is not ideal. Proper and efficient selection of the appropriate one of the many features is directly related to the ultimate function and achievement of the goal. And selecting the average monthly electricity consumption, the ascending and descending trend of the monthly electricity consumption and the average monthly temperature as input characteristics based on engineering experience and balance. Considering the influence of metering abnormality and the occurrence frequency of various metering abnormalities of users with abnormal meter reading data in nearly 6 months, 11 kinds of metering abnormalities are screened out as input characteristics.
Monthly daily power consumption: assume a sequence of monthly power usage by each user of
xn＝{xnmM1, 2 … M, and the sequence of days per month for each user is yn＝{ynmAnd M is 1,2, …, M }, the average electricity consumption of the user is every month and day
enm＝xnm/ynmn＝1,2…,N m＝1,2,M。
The trend of the rise and fall of monthly electricity consumption: the trend of the monthly power consumption is to make a prediction of the next monthly power consumption according to the continuous months of power consumption of the user and compare the predicted power consumption with the actual power consumption of the next month. Here, a moving average method is used. Monthly power consumption ascending and descending trend sequence s ═ snN is 1,2, … N, and the ascending and descending trend sequence of each user is sn＝{snm,m＝3,…M}，
The occurrence frequency of various metering abnormalities is as follows: the times of the metering abnormality occurrence comprise 11 metering abnormalities which are respectively uneven electric energy representation value, flying electric energy meter, falling electric energy meter, stopping electric energy meter, voltage phase failure, clock abnormality of electric energy meter, reverse electric quantity abnormality, tide reverse, voltage loss, automatic checking abnormality and electric quantity fluctuation abnormality, and the time sequence of the metering abnormality occurrence of the electric energy meter used by a user is jn＝{jnk,k＝1,2…K},n＝1,2…N。
Average temperature per month in grade city where house number is located: monthly and daily average air temperature of city of the place level of each userSequence is tn＝{tnm,m＝1,2…M}。
To implement the model structure of the BP neural network, as shown in fig. 8, in step 202), when modeling is performed using the BP neural network, in terms of setting the structure of the BP deep learning algorithm, a 5-layer network structure is set, 45 neurons in the input layer are provided, the numbers of neurons in the hidden layer are 500, 200, and 50, respectively, the output target is 1, a tanh nonlinear function is selected as an activation function of the hidden layer, a sigmoid function is used as an activation function of the output layer, and { x, y } represents an input variable and an output variable, where x ═ x [ x, y ]1，x2，x3…x45And the values of 45 characteristics of the input layer are represented, and y represents whether the user is abnormal in meter reading data. f (net)j) Is a non-linear activation function, and netj＝w11x1+w21x2+…w(45)1x45. And the model structure and the model parameters of the BP neural network are effectively realized.
The learning and training purpose of the abnormal attribute-to-reality model of the user meter reading data is to find the weight and the bias of the function C (w, b) capable of minimizing the loss of the mean square error. The update rule equation of the weight w and the bias b is as follows:
wherein XkTaking the kth input minimum sampling block from all opinion work order learning training samples, m being the total number of the minimum sampling blocks divided from all opinion work order learning training samples, η being the learning step size, l representing the network layer,
for the loss function C (w, b) a calculus partial derivative based on the weight w,
the partial derivatives are based on the deviation b calculus for the loss function C (w, b).
In step 205), in order to better adapt to practical application, better performance parameters are obtained through multiple experimental tests, and the Batch size is set to 200 according to repeated verification of experiments, so that the accuracy is relatively stable. The momentum coefficient is set to 0.95 by default. Specific model parameter configurations are shown in table 2 below, where 1 Epoch indicates that all training sample data is trained once through the network.
TABLE 2 user data for recording meter and judging abnormal attribute and actual attribute deep learning model training parameters
As shown in fig. 9, the loss values at different learning rates generally tend to decrease first and eventually tend to stabilize. And if the Epoch value reaches about 104 and the loss value tends to be stable at each learning rate, taking the Epoch value as 104.
As shown in fig. 10, the accuracy rate at different learning rates first shows a growing trend, and after several epochs, the small oscillation tends to be stable, and it can be found through observation that, when the learning rate is 0.001, the accuracy rate is higher, and then the learning rate is set to 0.001.
As shown in fig. 11, the influence of the different regularization parameters L2 on the loss function is relatively significant.
As shown in fig. 12, if the influence of the different regularization parameters L2 on the accuracy is not significant, the regularization parameter L2 is selected to be 0.001.
As shown in fig. 13, different hidden layer activation functions have a significant effect on the accuracy, and the uppermost curve represents the accuracy corresponding to different epochs when the hidden layer activation function is set to tanh, and it is easy to find that the accuracy is higher when the activation function is tanh, and therefore, the hidden layer activation function is set to tanh.
In order to maintain the timeliness and the adaptability of the model, the abnormal attribute and reality judgment model of the meter reading data is updated regularly in a mode of combining offline learning and online learning. So as to maintain the timeliness and adaptability of the model.
The example relies on an artificial intelligence 2.0 technology represented by deep learning, and adopts a DL4j open source deep learning technical framework.
The daily electricity consumption prediction model of the users with abnormal meter reading data, trained through machine learning, can realize the dynamic intelligent prediction of the daily electricity consumption of the users with abnormal meter reading data through self-adaptive learning along with time, and is remarkable in that the daily electricity consumption data output by a target can only be used as a theoretical prediction value, and the upper limit and the lower limit of the prediction value are used as reasonable confidence intervals. Suppose yt theoryRepresents the t-day daily power consumption, y, predicted by the modelt observationRepresenting the actual daily electricity consumption of t days, wherein delta is a movement early warning threshold (the default is 25%), when yt observation∈[yt theory*(1-δ)，yt theory*(1+δ)]And judging that the daily electricity consumption of the user is normal. When y ist observation＜yt theory(1- δ) or yt observation＞yt theory(1+ δ), determining that the daily power consumption of the user t is abnormal, and using yt theoryFor yt observationMake a correction, i.e. order
Considering that the abnormal daily electricity consumption data of the users with abnormal meter reading data exist for a long time in practice, the abnormal daily electricity consumption data needs to be accurately identified and corrected to ensure the original prediction trend of the model, and y is obtained by predicting a certain daytheoryAnd the real daily electricity yobservationAnd (4) carrying out abnormal motion prejudgment, and carrying out assignment judgment through the abnormal motion coefficient delta. If the user really uses the daily electricity yobservationDaily electricity consumption interval [ y ] no longer settheory*(1-δ),ytheory*(1+δ)]If yes, the user is indicated that the daily electricity data of the user is abnormal, and y is usedtheoryMake a correction, i.e. yobservation＝ytheory，yobservationAs a transfusionThe data is used for predicting the next round of daily electricity consumption.
The following concrete implementation examples of abnormal work order user identification and daily electricity quantity prediction technology of meter reading data based on BP and LSTM deep learning algorithm:
548 abnormal meter reading data worksheet users in 2019 are selected, the abnormal meter reading data attribute and real judgment model of the users is used for identification judgment, and the abnormal meter reading data attribute and real judgment model is compared with a subsequent actual processing result to obtain a confusion matrix, wherein 0 represents abnormal and 1 represents normal as shown in a table 3. ACC (accuracy, i.e. the proportion of the total number of samples of all correctly predicted results in the classification model), PPV (accuracy, i.e. the proportion of correct model predictions in all results where the model predictions are abnormal), TPR (sensitivity, i.e. the proportion of correct model predictions in all results where the true values are abnormal), TNR (specificity, i.e. the proportion of correct model predictions in all results where the true values are normal) can be calculated from the confusion matrix. Simple calculation results in ACC of 82.48%, PPV of 81.41%, TPR of 82.64%, and TNR of 82.33. One work order user is judged to be an abnormal user through the model, and the subsequent actual processing result is abnormal meter reading data caused by wrong wiring reasons. The complaint time of the work order user is 2019-06-9, daily electricity and temperature data from 2017-04-01 to 2019-03-31 are selected to generate a training set, and 2019-04-01 to 2019-06-08 are selected as suspected abnormal time intervals to determine abnormal days and error electricity of the training set. Table 4 shows the ratio of the actual daily power consumption and the predicted daily power consumption for the working day and the non-working day in the suspected abnormal time interval. Y is the same as the number of abnormal days and the error electric quantity part of the meter reading dataobservationRepresents the actual daily electricity consumption, ytheoryIndicating the predicted daily power consumption
When η is more than or equal to 20%, the data of the daily actual electricity consumption is marked as abnormal, and the error electricity quantity is marked as e-yobservation-ytheory. The number of abnormal days of the meter reading data is 29 days, and the error electricity quantity is 198.74 degrees.
TABLE 3 confusion matrix
| Exception (Predicted) | Normal (Predicted) | |
| Exception (Actual) | 219 | 46 |
| Normal (Actual) | 50 | 233 |
Table 4 actual daily electricity consumption and actual daily electricity consumption
The technology for recognizing abnormal work order users of meter reading data and predicting daily electric quantity based on BP and LSTM deep learning algorithm has the following advantages:
the ability of 95598 problem location and appeal check is improved. By means of the meter reading data abnormity attribute and reality judgment model based on the BP deep learning algorithm, users with abnormal meter reading data can be quickly screened out, field detection can be preferentially carried out on the screened out users, and working efficiency is improved. And the behavior of the processing result of the fictitious report client appeal is reduced by comparing the model prediction result with the actual processing result of the client appeal.
Secondly, the service quality of 95598 is improved. By means of the daily electricity consumption prediction model of the users with abnormal meter reading data based on the LSTM deep learning algorithm, abnormal daily electricity consumption data are accurately identified and corrected, and the abnormal days and error electricity consumption of the meter reading data can be scientifically and accurately estimated, so that customer satisfaction is improved, and the high-quality service level is improved.
And thirdly, the deep learning technology breakthrough of the mainstream open source DL4j is realized, all functions are packaged in a component mode, the expansibility is better, the adaptability is strong, the model training process is uniformly monitored, all modeling parameter functions can be completed through foreground page configuration, the stress of developers is reduced, and the timeliness of demand response is improved.
The following contents are introduced to the LSTM network deep learning technology:
the long-short term memory neural Network LSTM is an improved time-series Recurrent Neural Network (RNN), a variant of which is proposed by Sepp Hochreiter and Juergen Schmidhuber, with so-called long-short term memory cells, which can solve the problem of gradient disappearance. The LSTM can learn the long-term and short-term dependency information of the time sequence, and is suitable for processing and predicting interval and delay events in the time sequence due to the fact that a time memory unit is contained in the neural network.
The LSTM stores information in gating cells outside the normal flow of information in a circulating network. These units can store, write, or read information as data in a computer memory. The cell determines which information is stored, and when the information is allowed to be read, written, or cleared, by the switching of the gates. The memory function of the LSTM network is realized by the valve nodes of each layer. The valves are of type 3: forgetting gate, input gate and output gate. The input gate accepts new input from the outside and processes new data; a forgetting gate determines when to forget to output the result, thereby selecting an optimal time delay for the input sequence; the output gate will calculate all the results and generate output for the LSTM network element; these valves can be opened or closed to add a determination of whether the memory state of the model network (the state of the previous network) at the layer output reaches a threshold value to the current layer calculation. As shown in fig. 1, the valve node calculates the memory state of the network as input using a sigmoid function; if the output result reaches the threshold value, multiplying the valve output by the calculation result of the current layer to be used as the input of the next layer; and if the threshold value is not reached, forgetting the output result. The weights for each layer, including the valve nodes, are updated during each back-propagation training of the model.
These gates are switched on and off in response to the received signals, and, like the nodes of the neural network, they filter the information using their own set of weights, determining whether to allow the information to pass based on their strengths and importation. These weights, just like the weights of the modulation input and the hidden state, are adjusted by the learning process of the recurrent network. That is, the memory unit learns when data is allowed to enter, leave, or be deleted by guessing, error back-propagation, iterative process with gradient down adjustment weights.
As shown in fig. 14, the input time series is represented as X ═ X (X)1，x2，...，xn) The hidden state of the memory cell is H (H)1，h2，...，hn) The output time sequence is Y ═ Y (Y)1，y2，...，yn) The LSTM neural network is calculated as follows:
ht＝H(Whxxt+Whhxt-1+bh)
in the formula: h istOutput representing the state of the hidden unit at time t, WhxRepresenting a time series input xtTo hidden unit state htWeight vector between, WhhIndicates the hidden unit state htAnd ht-1Weight vector between, ht-1Hidden cell state output representing the time immediately before t, bhOutput h representing the state of a hidden unittThe deviation of (2).
pt＝Whyyt-1+by
In the formula: whyIndicates the hidden unit state htTo the output time series yt-1Weight vector of yt-1Indicating the output timing data at the time immediately before t, ptRepresenting the target prediction output at time t, byRepresenting a target prediction output ptThe deviation of (2).
Further decomposition, where the hidden unit state is calculated in the following formula:
it＝σ(Wixxt+Whhht-1+Wicct-1+bi)
in the formula: i.e. itInput valve, WixRepresenting a time series input xtTo the input valve itWeight vector between, WhhIndicates the hidden unit state htAnd ht-1Weight vector between, WicIndicating input valve itTo cell state output ct-1Weight vector of biIndicating input valve itThe deviation of (2).
ft＝σ(Wfxxt+Whhht-1+Wfcct-1+bf)
In the formula: f. oftForgetting the valve, WfxRepresenting a time series input xtTo forget valve ftWeight vector between, WfcValve f for indicating forgettingtTo cell state output ct-1Weight vector of bfValve f for indicating forgettingtσ denotes a sigmoid function.
ct＝＝ft*ct-1+it*g(Wcxxt+Whhht-1+Wccct-1bc
In the formula: c. CtRepresenting cell status output, ct-1Represents the cell state output, W, at the previous timecxRepresenting a time series input xtCell state output c to the previous timet-1Weight vector between, WccRepresents cell state output ctCell state output c to the previous timet-1Weight vector of bcRepresents cell state output ctG is a sigmoid function in the range of [ -2, 2]。
ot＝σ(Woxxt+Whhht-1+Wocct-1+bo)
In the formula: otIndicating the output valve, WoxRepresenting a time series input xtTo the output valve otWeight vector between, WocIndicating the output valve otWeight vector to cell state output c betweentWeight vector of boIndicating the output valve otThe deviation of (2).
ht＝ot*h(ct)
In the formula: denotes the scalar product of two vectors or matrices, and h is the sigmoid function range [ -1, 1 ].
For the objective function, we use the square loss function represented by the following formula:
the meter reading data anomaly analysis method based on the deep learning algorithm shown in fig. 1 to 13 is a specific embodiment of the present invention, already embodies the outstanding substantive features and significant progress of the present invention, and can make equivalent modifications in the aspects of shape, structure and the like according to the practical use requirements and under the teaching of the present invention, which are within the protection scope of the present scheme.
Claims (10)
1. A meter reading data anomaly analysis method based on a deep learning algorithm is characterized by comprising the following steps:
1) performing feature extraction on a data set of a work order user with abnormal meter reading data;
2) reading a user meter reading data abnormal attribute and reality judgment algorithm model subjected to deep learning from a database;
3) judging whether the data is abnormal or not through the identification of an abnormal actual judgment algorithm model of the user meter reading data, if not, returning a judgment result, and if so, executing the next step;
4) according to the characteristics of daily electricity consumption data of users with abnormal meter reading data and the service requirement of error electricity consumption of users with abnormal meter reading data needing to be estimated, establishing a daily electricity consumption prediction algorithm model of users with abnormal meter reading data by adopting an LSTM deep learning algorithm, and designing an LTSM neural network model structure and model parameters;
5) acquiring a training data set of a daily electricity consumption prediction algorithm model of a user with abnormal meter reading data;
6) deep learning training is carried out on the daily electricity prediction algorithm model of the user with abnormal meter reading data, training sample data classification and normalization processing are completed, and deep learning training is completed through a random gradient descent optimization algorithm;
7) and calculating abnormal days and error electric quantity through a daily electric quantity prediction algorithm model of the user with abnormal meter reading data so as to judge and correct abnormal electricity consumption.
2. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 1, characterized in that: in the step 2), the establishment of the abnormal attribute and reality judgment algorithm model of the user meter reading data comprises the following steps:
201) defining learning model parameters including learning rate, regular parameters, iteration times and the number of neural network layers;
202) establishing an abnormal attribute and reality judgment algorithm model of the user meter reading data by feature extraction and adopting a BP neural network;
203) acquiring a latest training data set of an abnormal attribute and reality judgment algorithm model of user meter reading data;
204) performing machine learning training on the abnormal attribute and reality judgment algorithm model of the user meter reading data, and performing unified training on all parameters of the model by adopting a random gradient descent algorithm to adjust hidden layer parameters and finally obtain output layer parameters;
205) optimally configuring key parameters related to an abnormal attribute and reality judgment algorithm model of user meter reading data, and balancing model parameter setting by comparing a loss function descending curve trend and accuracy in the model training and testing processes;
206) and finishing the training and learning of the model, and writing the model back to the database.
3. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 2, characterized in that: the process of deep learning training of the user meter reading data abnormal user daily electricity consumption prediction algorithm model in the step 6) and the user meter reading data abnormal attribute judgment algorithm model in the step 204) comprises the following steps:
601) training data vectorization processing;
602) performing classified iterative learning;
603) judging whether the iteration times is less than the iteration times, if not, finishing model training and learning, and if so, executing the next step;
604) updating the weights and the offsets and returning to step 603).
4. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 3, characterized in that: the prediction algorithm model for the daily electricity consumption of the user with abnormal meter reading data is provided with a 4-layer network structure, 5 neurons of an input layer are provided, the number of neurons of a hidden layer is 200, 100 and 100 respectively, an output target is 1, a tanh nonlinear function is selected as an activation function of the hidden layer, and an identity function is selected as an activation function of an output layer.
5. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 1, characterized in that: in the step 6), the training sample data classification is to split daily electricity consumption data of the user, which is calibrated to be abnormal through the abnormal attribute judgment model of the user meter reading data, into two data sets of daily electricity consumption in holidays and daily electricity consumption in working days to form two time sequences, and each time sequence is independently modeled.
6. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 1, characterized in that: in step 6), during normalization, the sequence data is normalized to [0, 1]]Within the scope, the transfer function is:
in the formula xiRepresenting the ith input value of the time series, x representing all inputs of the series, yiThe ith output value representing a time series.
7. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 1, characterized in that: in step 7), the optimal configuration strategy needs to be repeatedly verified through multiple experimental tests to determine the value of each hyper-parameter.
8. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 2, characterized in that: in step 202), the feature extraction content comprises the average daily power consumption per month, the ascending and descending trend of the daily power consumption, the occurrence frequency of various metering abnormalities and the average daily temperature of the grade city where the house number is located.
9. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 2, characterized in that: in step 202), when modeling is performed by using a BP neural network, in the aspect of setting a BP deep learning algorithm structure, a 5-layer network structure is set, 45 input layer neurons are provided, the number of hidden layer neurons is 500, 200 and 50 respectively, an output target is 1, a tanh nonlinear function is selected as an activation function of a hidden layer, and a sigmoid function is selected as an activation function of an output layer.
10. The meter reading data anomaly analysis method based on the deep learning algorithm according to claim 8, characterized in that: the number of times of occurrence of the metering abnormality comprises 11 metering abnormalities, namely uneven electric energy representation value, flying away of the electric energy meter, backward walking of the electric energy meter, stopping of the electric energy meter, voltage phase failure, clock abnormality of the electric energy meter, reverse electric quantity abnormality, reversed trend, voltage loss, automatic checking abnormality and electric quantity fluctuation abnormality.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911272670.8A CN111047094A (en) | 2019-12-12 | 2019-12-12 | Meter reading data anomaly analysis method based on deep learning algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911272670.8A CN111047094A (en) | 2019-12-12 | 2019-12-12 | Meter reading data anomaly analysis method based on deep learning algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111047094A true CN111047094A (en) | 2020-04-21 |
Family
ID=70236021
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911272670.8A Pending CN111047094A (en) | 2019-12-12 | 2019-12-12 | Meter reading data anomaly analysis method based on deep learning algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111047094A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111539492A (en) * | 2020-07-08 | 2020-08-14 | 武汉格蓝若智能技术有限公司 | Abnormal electricity utilization judgment system and method based on reinforcement learning |
| CN111612651A (en) * | 2020-05-27 | 2020-09-01 | 福州大学 | Abnormal electric quantity data detection method based on long-term and short-term memory network |
| CN111652417A (en) * | 2020-05-25 | 2020-09-11 | 成都千嘉科技有限公司 | Automatic correction method and system for remote meter reading data |
| CN111652570A (en) * | 2020-05-12 | 2020-09-11 | 国网福建省电力有限公司 | Automatic auditing system and method |
| CN111651935A (en) * | 2020-05-25 | 2020-09-11 | 成都千嘉科技有限公司 | Multi-dimensional expansion prediction method and device for non-stationary time series data |
| CN111781554A (en) * | 2020-06-28 | 2020-10-16 | 南方电网科学研究院有限责任公司 | Metering device fault determination method and system based on four-quadrant electric energy data |
| CN111815487A (en) * | 2020-06-28 | 2020-10-23 | 珠海中科先进技术研究院有限公司 | Health education assessment method, device and medium based on deep learning |
| CN111833583A (en) * | 2020-07-14 | 2020-10-27 | 南方电网科学研究院有限责任公司 | Training method, device, equipment and medium for power data anomaly detection model |
| CN112230083A (en) * | 2020-10-10 | 2021-01-15 | 国网四川省电力公司电力科学研究院 | Gateway metering device abnormal event identification method and system |
| CN112540629A (en) * | 2020-11-18 | 2021-03-23 | 河南卓正电子科技有限公司 | Intelligent valve flow calibration method and device |
| CN112560338A (en) * | 2020-12-10 | 2021-03-26 | 东北大学 | Complex industrial system intelligent forecasting method, device, equipment and storage medium based on adaptive deep learning |
| CN112926686A (en) * | 2021-03-30 | 2021-06-08 | 武汉工程大学 | BRB (Brillouin bus) and LSTM (least Square) model-based power big data power utilization anomaly detection method and device |
| CN113033089A (en) * | 2021-03-23 | 2021-06-25 | 中国联合网络通信集团有限公司 | Method and device for identifying abnormal power consumption user |
| CN113642601A (en) * | 2021-07-02 | 2021-11-12 | 国网江苏省电力有限公司电力科学研究院 | Medium voltage distribution network transfer operation identification method, device and equipment |
| CN113762470A (en) * | 2021-08-23 | 2021-12-07 | 迟源 | Prediction model construction method, prediction method, device, equipment and medium |
| TWI831033B (en) * | 2021-07-28 | 2024-02-01 | 財團法人紡織產業綜合研究所 | Electricity usage scheduling system and electricity usage scheduling method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170169357A1 (en) * | 2015-12-15 | 2017-06-15 | Deep Instinct Ltd. | Methods and systems for data traffic analysis |
| CN109685240A (en) * | 2018-10-24 | 2019-04-26 | 国网浙江省电力有限公司 | A kind of prediction of 95598 traffic work orders and unusual fluctuation method for early warning based on LSTM deep learning |
-
2019
- 2019-12-12 CN CN201911272670.8A patent/CN111047094A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170169357A1 (en) * | 2015-12-15 | 2017-06-15 | Deep Instinct Ltd. | Methods and systems for data traffic analysis |
| CN109685240A (en) * | 2018-10-24 | 2019-04-26 | 国网浙江省电力有限公司 | A kind of prediction of 95598 traffic work orders and unusual fluctuation method for early warning based on LSTM deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 罗慧等: "基于长短期记忆网络的智能用电数据甄别方法", 《广东电力》 * |
| 陈启鑫等: "异常用电的检测方法:评述与展望", 《电力系统自动化》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652570A (en) * | 2020-05-12 | 2020-09-11 | 国网福建省电力有限公司 | Automatic auditing system and method |
| CN111652417A (en) * | 2020-05-25 | 2020-09-11 | 成都千嘉科技有限公司 | Automatic correction method and system for remote meter reading data |
| CN111651935A (en) * | 2020-05-25 | 2020-09-11 | 成都千嘉科技有限公司 | Multi-dimensional expansion prediction method and device for non-stationary time series data |
| CN111612651B (en) * | 2020-05-27 | 2022-07-26 | 福州大学 | Abnormal electric quantity data detection method based on long-term and short-term memory network |
| CN111612651A (en) * | 2020-05-27 | 2020-09-01 | 福州大学 | Abnormal electric quantity data detection method based on long-term and short-term memory network |
| CN111815487A (en) * | 2020-06-28 | 2020-10-23 | 珠海中科先进技术研究院有限公司 | Health education assessment method, device and medium based on deep learning |
| CN111815487B (en) * | 2020-06-28 | 2024-02-27 | 珠海中科先进技术研究院有限公司 | Deep learning-based health education assessment method, device and medium |
| CN111781554A (en) * | 2020-06-28 | 2020-10-16 | 南方电网科学研究院有限责任公司 | Metering device fault determination method and system based on four-quadrant electric energy data |
| CN111539492A (en) * | 2020-07-08 | 2020-08-14 | 武汉格蓝若智能技术有限公司 | Abnormal electricity utilization judgment system and method based on reinforcement learning |
| CN111539492B (en) * | 2020-07-08 | 2020-11-20 | 武汉格蓝若智能技术有限公司 | Abnormal electricity utilization judgment system and method based on reinforcement learning |
| CN111833583B (en) * | 2020-07-14 | 2021-09-03 | 南方电网科学研究院有限责任公司 | Training method, device, equipment and medium for power data anomaly detection model |
| CN111833583A (en) * | 2020-07-14 | 2020-10-27 | 南方电网科学研究院有限责任公司 | Training method, device, equipment and medium for power data anomaly detection model |
| CN112230083A (en) * | 2020-10-10 | 2021-01-15 | 国网四川省电力公司电力科学研究院 | Gateway metering device abnormal event identification method and system |
| CN112230083B (en) * | 2020-10-10 | 2022-08-30 | 国网四川省电力公司电力科学研究院 | Method and system for identifying abnormal events of gateway metering device |
| CN112540629A (en) * | 2020-11-18 | 2021-03-23 | 河南卓正电子科技有限公司 | Intelligent valve flow calibration method and device |
| CN112560338B (en) * | 2020-12-10 | 2022-03-25 | 东北大学 | Complex industrial system intelligent forecasting method, device, equipment and storage medium based on adaptive deep learning |
| CN112560338A (en) * | 2020-12-10 | 2021-03-26 | 东北大学 | Complex industrial system intelligent forecasting method, device, equipment and storage medium based on adaptive deep learning |
| CN113033089A (en) * | 2021-03-23 | 2021-06-25 | 中国联合网络通信集团有限公司 | Method and device for identifying abnormal power consumption user |
| CN112926686B (en) * | 2021-03-30 | 2023-11-17 | 武汉工程大学 | BRB and LSTM model-based power consumption anomaly detection method and device for big power data |
| CN112926686A (en) * | 2021-03-30 | 2021-06-08 | 武汉工程大学 | BRB (Brillouin bus) and LSTM (least Square) model-based power big data power utilization anomaly detection method and device |
| CN113642601A (en) * | 2021-07-02 | 2021-11-12 | 国网江苏省电力有限公司电力科学研究院 | Medium voltage distribution network transfer operation identification method, device and equipment |
| CN113642601B (en) * | 2021-07-02 | 2023-11-17 | 国网江苏省电力有限公司电力科学研究院 | Medium voltage distribution network transfer operation identification method, device and equipment |
| TWI831033B (en) * | 2021-07-28 | 2024-02-01 | 財團法人紡織產業綜合研究所 | Electricity usage scheduling system and electricity usage scheduling method |
| CN113762470A (en) * | 2021-08-23 | 2021-12-07 | 迟源 | Prediction model construction method, prediction method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111047094A (en) | Meter reading data anomaly analysis method based on deep learning algorithm | |
| Tikhamarine et al. | Improving artificial intelligence models accuracy for monthly streamflow forecasting using grey Wolf optimization (GWO) algorithm | |
| CN109685240B (en) | 95598 ticket prediction and transaction early warning method based on LSTM deep learning | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| CN106022954B (en) | Multiple BP neural network load prediction method based on grey correlation degree | |
| CN106952181A (en) | Electric Load Prediction System based on long Memory Neural Networks in short-term | |
| CN106960252A (en) | Methods of electric load forecasting based on long Memory Neural Networks in short-term | |
| CN111639783A (en) | Line loss prediction method and system based on LSTM neural network | |
| CN112418491A (en) | Dynamic configuration method for residual sediment storage capacity of reservoir | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN113722997A (en) | New well dynamic yield prediction method based on static oil and gas field data | |
| CN112734002A (en) | Service life prediction method based on data layer and model layer joint transfer learning | |
| CN112116002A (en) | Determination method, verification method and device of detection model | |
| CN112990585A (en) | Hen laying rate prediction method based on LSTM-Kalman model | |
| Turetskyy et al. | Artificial neural network enabled p2d model deployment for end-of-line battery cell characterization | |
| CN113591957B (en) | Wind power output short-term rolling prediction and correction method based on LSTM and Markov chain | |
| CN114819395A (en) | Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model | |
| CN115329823A (en) | Dynamic depth migration prediction method and device for furnace tube temperature of delayed coking heating furnace | |
| CN115618987A (en) | Production well production data prediction method, device, equipment and storage medium | |
| CN117113086A (en) | Energy storage unit load prediction method, system, electronic equipment and medium | |
| CN112232570A (en) | Forward active total electric quantity prediction method and device and readable storage medium | |
| CN111061708A (en) | Electric energy prediction and restoration method based on LSTM neural network | |
| D’Ambrosio et al. | Meta-model assisted evolutionary optimization of cellular automata: An application to the SCIARA model | |
| CN114169416B (en) | Short-term load prediction method based on migration learning under small sample set |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210128 Address after: 310007 No. 8 Huanglong Road, Hangzhou, Zhejiang, Xihu District Applicant after: STATE GRID ZHEJIANG ELECTRIC POWER Co.,Ltd. Applicant after: Marketing service center of State Grid Zhejiang Electric Power Co.,Ltd. Applicant after: ZHEJIANG HUAYUN INFORMATION TECHNOLOGY Co.,Ltd. Address before: 310007 No. 8 Huanglong Road, Hangzhou, Zhejiang, Xihu District Applicant before: STATE GRID ZHEJIANG ELECTRIC POWER Co.,Ltd. Applicant before: STATE GRID ZHEJIANG ELECTRIC POWER COMPANY LIMITED ELECTRIC POWER Research Institute Applicant before: ZHEJIANG HUAYUN INFORMATION TECHNOLOGY Co.,Ltd. |
|
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200421 |