CN111967512A - Abnormal electricity utilization detection method, system and storage medium - Google Patents

Abstract

The invention discloses a method, a system and a storage medium for detecting abnormal power consumption, and belongs to the technical field of abnormal power consumption behavior identification of a smart grid. The abnormal electricity consumption detection method includes the following steps: acquiring smart meter data to be detected; converting the smart electricity meter data into multi-time-series data; inputting the multi-time-series data into the trained abnormal electricity consumption detection model to obtain the predicted electricity consumption type Label; take the electricity consumption type with the highest probability among the predicted electricity consumption type labels as the abnormal electricity consumption detection result of the smart meter data to be detected. The invention can fully and comprehensively utilize the big data of electricity consumption, deeply mine the advanced features of abnormal electricity consumption behavior, improve the generalization performance of abnormal electricity consumption detection, reduce the proportion of misjudgment, and detect abnormal electricity consumption efficiently and accurately.

Description

A kind of abnormal power consumption detection method, system and storage medium

technical field

The invention relates to an abnormal power consumption detection method, system and storage medium, and belongs to the technical field of abnormal power consumption behavior identification of a smart grid.

Background technique

For a long time, abnormal electricity consumption such as electricity theft, meter failure and installation error has brought huge economic losses to grid operators every year. Moreover, due to the huge errors in the collected power consumption data, it will also affect the scheduling and management of the power grid, as well as the operation safety. Therefore, abnormal power consumption detection is one of the important supports for safe power consumption in the process of smart grid operation and maintenance, which is of great significance. By identifying whether the electricity consumption is normal or not, it is possible to make up for the under-metered electricity bill caused by abnormal electricity consumption. In addition, it can also correct the distorted power consumption data, improve the quality of power grid data, and provide guarantee for a wider range of power big data analysis. At present, industrial and commercial users in major countries in the world, including my country, account for a very large proportion of electricity consumption in the whole society. In my country, the proportion has exceeded 80%. Therefore, it is particularly important to detect abnormal electricity consumption among industrial and commercial users.

More and more researchers are keen on methods based on big data analysis, using statistics, data mining, machine learning and other technical means to find abnormal electricity consumption behavior. Compared with hardware-based solutions, such methods have the advantages of high efficiency and fast detection speed; and as the power sector accumulates more and more smart meter data, such methods are gradually becoming mainstream. The abnormal power consumption detection method based on data analysis mainly relies on the implementation mode of artificial feature modeling and discriminative algorithm design; especially the performance of the feature model directly determines the pros and cons of the detection method. Often due to the insufficient generalization ability of the feature model, a large number of false positives are caused. At present, the shortcomings of artificial feature modeling are embodied in the following aspects:

1) The data phenomena of abnormal electricity consumption behavior are extremely diverse, and the data scale of smart meters is extremely large. When manually designing data features, it is impossible to traverse all possible data phenomena. The artificially designed data features are often based on a small amount of data, which leads to the gradual degradation of the accuracy of the detection model over time;

2) The data characteristics of abnormal electricity consumption are multifaceted, including both short-term and long-term characteristics; both local and global characteristics; it is difficult to manually design a feature model with excellent performance in a short period of time;

3) The smart meter data contains a large amount of various types of noise, which will interfere with human experts for data analysis and feature modeling;

4) The artificially designed data features often have strong limitations, which are directly related to the technical level and experience of human experts, resulting in strong subjectivity and poor scalability of the feature model, which cannot meet the application requirements.

It can be seen that the traditional abnormal power consumption detection methods have been unable to meet the operation, maintenance and development needs of smart grids. How to make full use of electricity consumption big data, deeply mine the advanced characteristics of abnormal electricity consumption behavior, and efficiently and accurately detect abnormal electricity consumption has become the main trend of smart grid development.

SUMMARY OF THE INVENTION

The purpose of the present invention is to overcome the deficiencies in the prior art, and to provide a method, system and storage medium for detecting abnormal electricity consumption, which can fully and comprehensively utilize the big data of electricity consumption, deeply mine the advanced features of abnormal electricity consumption behavior, and improve the abnormality of electricity consumption. The generalization performance of electricity detection, reducing the proportion of misjudgments, and detecting abnormal electricity efficiently and accurately.

To achieve the above object, the present invention adopts the following technical solutions to realize:

first:

A method for detecting abnormal electricity consumption, the method for detecting abnormal electricity consumption comprises the following steps:

Obtain the data of the smart meter to be detected;

Convert smart meter data to multi-time series data;

Input the multi-time series data into the trained abnormal electricity consumption detection model to obtain the predicted electricity consumption type label;

The electricity consumption type with the highest probability in the predicted electricity consumption type label is used as the abnormal electricity consumption detection result of the smart meter data to be detected.

Further, the method for converting smart meter data into multi-time series data includes the following steps:

Convert smart meter data into the form of multi-time series data, construct and append data quality series and date type series;

Fill in the missing data for the acquired sequence data and divide it according to the natural week, and construct it as multi-time series data;

Independent data normalization for multiple time series data.

Further, the method for constructing the data quality sequence includes the following steps:

The number of missing smart meter data is counted according to the timestamp, and the corresponding timestamp is assigned to the counted missing number to construct a data quality sequence.

Further, the method for constructing a date type sequence includes the following steps:

According to the working day or holiday attribute of each natural day, one date type data is given for each hour of the natural day, and a date type sequence is constructed.

Further, the data normalization processing adopts the following classification normalization methods;

The voltage data is normalized as follows:

The current data is normalized as follows:

The total active power is normalized as follows:

Data quality is normalized as follows:

Further, the training method of the abnormal electricity consumption detection model includes the following steps:

Obtain smart meter data of normal and abnormal electricity consumption cases, and convert smart meter data into multi-time series data samples;

The multi-time series data samples are calibrated by electricity type labels, and divided into training sample sets and verification sample sets;

Randomly sample a batch of data samples from the training sample set, input them into the deep hybrid neural network, calculate the loss and optimize the parameters of the deep hybrid neural network;

Use the parameter-optimized deep hybrid neural network to traverse the entire training sample set to complete a round of training;

After each round of training, use the verification sample set to evaluate the training effect of the deep hybrid neural network, and save a snapshot of the abnormal electricity consumption detection model;

Repeat the training to reach a preset number of epochs, make the loss of the deep hybrid neural network level off, and end the training;

According to the evaluation results after each round of training, the trained abnormal electricity consumption detection model is determined.

Further, the method for obtaining the electricity consumption type label includes the following steps:

Extract local features of smart meter data;

Extract global features of smart meter data;

The local features and global features are concatenated and merged, and converted into labels of electricity consumption types.

Further, the data structure sequence of the multi-time-series data sequentially includes A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase current, C-phase current, total active power, total power factor, data quality and Date type.

Second aspect:

An abnormal power consumption detection system, the system includes the following modules:

The data acquisition module is used to acquire the data of the smart meter to be detected;

Data conversion module for converting smart meter data into multi-time series data;

The abnormal electricity consumption detection model is used to obtain the predicted electricity consumption type label according to the input multi-time series data;

The detection output module is used for taking the electricity consumption type with the highest probability in the predicted electricity consumption type label as the abnormal electricity consumption detection result of the smart meter data to be detected.

Further, the abnormal power consumption detection model adopts a deep hybrid neural network, and the deep hybrid neural network includes a local feature network, a global feature network and a classification network;

The local feature network is used to extract local features of smart meter data;

The global feature network is used to extract global features of smart meter data;

The classification network is used for splicing and merging local features and global features, and converting them into labels of electricity consumption types.

Further, the local feature network includes a plurality of cascaded one-dimensional convolution modules and a Flatten layer;

The one-dimensional convolution module is used to learn and extract the local features of the smart meter data; the Flatten layer is used to expand and output the local features extracted by the one-dimensional convolution module into a one-dimensional vector;

The one-dimensional convolution module includes a one-dimensional convolution layer and a one-dimensional average pooling layer;

The one-dimensional convolution layer is used to keep the output time axis dimension consistent with the input time axis dimension;

The one-dimensional average pooling layer is used to reduce the dimension of the input time axis on the time axis by using the averaging operator.

Further, the global feature network includes a Flatten layer and multiple cascaded fully connected layers;

The Flatten layer is used to expand the multi-time series smart meter data into a one-dimensional vector;

The fully connected layer is used to learn and extract the global features of the smart meter data from the one-dimensional vector output by the Flatten layer of the global feature network.

Further, the classification network includes a splicing layer and a plurality of cascaded fully connected layers;

The splicing layer is used for splicing the local features and global features output by the local feature network and the global feature network into a one-dimensional feature vector, and inputting multiple stacked fully connected layers;

The fully connected layer is used to calculate and output the type labels of normal power consumption and abnormal power consumption.

The third aspect:

An abnormal power consumption detection system, comprising a processor and a storage medium;

the storage medium is used for storing instructions;

The processor is configured to operate according to the instructions to perform the steps of any one of the abnormal power usage detection methods according to the first aspect.

Fourth aspect:

A computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of any one of the abnormal power usage detection methods of the first aspect.

Compared with the prior art, the beneficial effects achieved by the present invention:

The invention provides an abnormal electricity consumption detection method based on a deep hybrid neural network. By learning the local features and global characteristics of the smart meter data, the advanced features of the electricity consumption behavior are more comprehensively extracted, so as to accurately identify the abnormal electricity consumption behavior. It can better ensure the long-term stability of abnormal power consumption detection performance and the generalization ability in different application scenarios;

The use of multi-time series data, with natural weeks as the unit time span, can better describe the periodicity of users' electricity consumption behavior; by supplementing date types and data quality, the information content of data samples is further enriched, thereby reducing data missing to anomalies The interference of electricity consumption identification can improve the influence of dynamic adjustment of working days and holidays on abnormal electricity consumption detection, ensure that the abnormal electricity consumption detection performance is more stable, reduce the misjudgment rate, and improve the operation efficiency of power companies.

Description of drawings

1 is a flowchart of a method for detecting abnormal power consumption provided by an embodiment of the present invention;

2 is a flow chart of the construction of multi-time series data provided by an embodiment of the present invention;

3 is a schematic structural diagram of multi-time series data provided by an embodiment of the present invention;

Fig. 4 is the training flow chart of the abnormal electricity consumption detection model provided by an embodiment of the present invention;

5 is a flowchart of obtaining a power consumption type label provided by an embodiment of the present invention;

6 is a schematic diagram of a framework of an abnormal power consumption detection system provided by an embodiment of the present invention;

7 is a schematic diagram of a framework of an abnormal power consumption detection model provided by an embodiment of the present invention;

FIG. 8 is a schematic structural diagram of a deep hybrid neural network provided by an embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solutions of the present invention more clearly, and cannot be used to limit the protection scope of the present invention.

Example 1:

As shown in FIG. 1 , based on the requirements of efficiently and accurately detecting abnormal power consumption and smart grid operation, maintenance and development, an embodiment of the present invention provides a method for detecting abnormal power consumption, which includes the following steps:

Step 101, acquiring smart meter data to be detected;

Step 102, converting smart meter data into multi-time series data;

Step 103, input the multi-time series data into the trained abnormal electricity consumption detection model to obtain the predicted electricity consumption type label;

Step 104 , identifying the electricity consumption type with the highest probability in the predicted electricity consumption type label as the abnormal electricity consumption detection result of the smart meter data to be detected.

Modern smart grids collect massive amounts of smart meter data, which are characterized by large scale, high latitude, and various types. In view of the data characteristics of smart meters, in order to improve the data quality of smart meters in the power grid, the embodiment of the present invention provides a method for constructing multi-time series data as shown in FIG. Follow the steps below:

Step 201: Acquire smart meter data, and convert the smart meter data into eight items including A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase current, C-phase current, total active power, and total power factor. Time series multi-time series data;

Step 202: Count the number of missing items in the A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase current, C-phase current, total active power, and total power factor corresponding to each time stamp; The value of quality, the lowest is 0, which means the data is complete; the highest is 8, which means that all data are missing;

Step 203, assign a corresponding time stamp to the missing number counted in step 202, and construct a data quality sequence; the data quality sequence represents the A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase voltage corresponding to each time stamp The number of missing data items such as phase current, C-phase current, total active power and total power factor;

Step 204: According to the attributes of working days or holidays of each natural day, one date type data per hour of the day is given, and a date type sequence is constructed; The date type data content of day is the same;

Step 205: Add the data quality sequence and date type sequence constructed in steps 203 and 204 to the smart meter data sequence according to the time stamp, so that the generated new smart meter data sequence includes: phase A voltage, phase B voltage, phase C 10 time series such as voltage, A-phase current, B-phase current, C-phase current, total active power, total power factor, data quality and date type; by adding data quality and date type, the information content of data samples is further enriched, It can improve the influence of dynamic adjustment of working days and holidays on the detection of abnormal electricity consumption;

In step 206, based on the data adjacent to the missing value, the missing data items are filled by the linear interpolation method, which can reduce the interference of the missing data on the identification of abnormal electricity consumption;

Step 207, dividing the new smart meter data sequence according to the natural week, to form a smart meter data sample with the natural week as the time span; with the natural week as the unit time span, the periodicity of the user's electricity consumption behavior can be better described;

Step 208 , normalize the smart meter data samples according to the data type; wherein, the normalization method of the three-phase voltage data is as follows:

Among them, the normalization method of the three-phase current data is as follows:

Among them, the normalization of the total active power is as follows:

Among them, the standard range of the total power factor is already between [0, 1], so there is no need for normalization;

Among them, the normalization method of data quality is as follows:

Among them, the value of date type data only includes 0 and 1, so no normalization is required.

Specifically, as shown in FIG. 3, an embodiment of the present invention provides a multi-channel time series data structure, the data structure includes: A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase current, 10 time series such as C-phase current, total active power, total power factor, data quality and date type; among them, the data collection frequency is 1 data/hour, and the timestamps of each channel are strictly aligned; by supplementing the date type and data quality, further enriching the information of the data samples.

As shown in Figure 3, the multi-channel time series are arranged in the following order: A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase current, C-phase current, total active power, total power factor, data quality and Date type. Among them, the time span of the multi-channel time series data sample structure is a natural week, starting at 0:00 on Monday and ending at 23:00 on Sunday.

Specifically, as shown in FIG. 4 , in order to improve the stability and reliability of the abnormal electricity consumption detection model, an embodiment of the present invention provides a training method for the abnormal electricity consumption detection model, which specifically includes the following steps:

Step 401: Acquire smart meter data of normal electricity consumption and abnormal electricity consumption cases, and convert the acquired smart electricity meter data into multi-time series data samples according to the multi-time-series data construction method shown in FIG. 2;

Step 402: Assign a label of the electricity consumption type to the multi-time-series data sample according to the user's electricity consumption type to which the multi-time-series data sample belongs, wherein the normal electricity consumption sample is 0, and the abnormal electricity consumption sample is 1; The labeled multi-time-series data samples are divided into a training sample set and a verification sample set according to a certain ratio; as an embodiment of the present invention, the division ratio of the training sample set and the verification sample set may be 7:3;

Step 403: Randomly sample a batch of multi-time-series data samples from the training sample set, obtain sample data and their type labels, and encode the labels in a one-hot manner. The sample batch size can be 200;

Step 404, using the multi-category cross entropy as the loss function, using the Adam optimizer to calculate the gradient of each parameter of the hybrid neural network, and using the back propagation algorithm to update the parameters of the hybrid neural network according to the set learning rate. In one embodiment, the learning rate can be set to 0.001;

Step 405, if all data samples in the training sample set have been traversed, go to Step 406, otherwise go to Step 403;

Step 406, use the deep hybrid neural network model to predict the labels of all samples in the sample set, and use the real labels to calculate the AUC score of the abnormal electricity consumption category, and evaluate the training effect of the deep hybrid neural network; wherein, the AUC score (Area Under ROC The ROC (Receiver Operating Characteristic) of ) is the receiver operating characteristic curve, which is a popular method to verify the performance of classifiers on imbalanced datasets;

Step 407, if the abnormal power consumption category AUC score is the highest among all completed training rounds, go to step 408, otherwise go to step 409;

Step 408, save the deep hybrid neural network model as a snapshot file, if the snapshot file already exists, overwrite the existing snapshot file with the latest snapshot file, and proceed to step 409;

In step 409, if the number of training rounds has reached the preset maximum number of training rounds, step 410 is performed, otherwise, step 403 is performed, wherein, as an embodiment of the present invention, the maximum number of training rounds can be selected as 50;

Step 410: End the training, and use the saved optimal snapshot file as the trained abnormal electricity consumption detection model.

The trained abnormal electricity consumption detection model can extract the advanced features of electricity consumption behavior more comprehensively by learning the local and global features of smart meter data, and has higher abnormal electricity consumption evaluation efficiency and higher abnormal electricity consumption detection accuracy. rate, thereby saving a lot of human and material resources and other resources.

Specifically, as shown in FIG. 5 , an embodiment of the present invention provides a method for obtaining a power consumption type label, which specifically includes the following steps:

Step 501, extracting local features of smart meter data;

Step 502, extracting global features of smart meter data;

Step 503, splicing and merging the local features and the global features, and converting them into labels of electricity consumption types.

It should be noted that, in the process of obtaining electricity type labels, there is no strict sequence restriction on the extraction of local features and global features of smart meter data, as long as the relevant features can be successfully extracted.

From the above, it can be seen that the abnormal power consumption detection method provided in the embodiment of the present invention can more comprehensively extract the advanced features (local features and global features) of the power consumption behavior, can reduce the interference of data missing on the identification of abnormal power consumption, and improve the working days and holidays. The impact of dynamic adjustment on abnormal power consumption detection ensures that the abnormal power consumption detection performance is more stable, reduces the misjudgment rate, and improves the operation efficiency of power companies and the generalization ability in different application scenarios.

Embodiment 2:

As shown in FIG. 6 , an embodiment of the present invention provides an abnormal power consumption detection system, and the abnormal power consumption detection system includes the following modules:

The data acquisition module is used to acquire the data of the smart meter to be detected;

Data conversion module for converting smart meter data into multi-time series data;

The abnormal electricity consumption detection model is used to obtain the predicted electricity consumption type label according to the input multi-time series data;

The detection output module is used for taking the electricity consumption type with the highest probability in the predicted electricity consumption type label as the abnormal electricity consumption detection result of the smart meter data to be detected.

Specifically, as shown in Figures 7 and 8, the abnormal power consumption detection model of the abnormal power consumption detection system adopts a deep hybrid neural network, wherein the deep hybrid neural network includes a local feature network, a global feature network and a classification network;

Among them, the local feature network is used to extract the local features of the smart meter data; the global feature network is used to extract the local features of the smart meter data; the classification network is used to splicing and merging the local features and global features, and converting them into labels of electricity consumption types .

Specifically, the local feature network is based on multiple cascaded one-dimensional convolution modules, processes smart meter data along the time axis, and uses the local perception characteristics of one-dimensional convolution units to learn and extract local features of smart meter data; The Flatten layer expands the local features output by the one-dimensional convolution module into a one-dimensional vector. Among them, the one-dimensional convolution module includes a one-dimensional convolution layer and a one-dimensional average pooling layer. The time axis dimension of the output of the one-dimensional convolution layer is consistent with the input. dimensionality reduction on the axis.

As shown in Figure 8, in this embodiment, the local feature network includes three cascaded 1-D convolution modules and one Flatten layer. Among them, the size of the convolution kernel of the 1-dimensional convolution layer-1 of the 1-dimensional convolution module 1 is 1*5, the number of filters is 16, and the size of the pooling kernel of the 1-dimensional average pooling layer-1 is 1*4; The size of the convolution kernel of the 1D convolution layer-2 of the 1D convolution module 2 is 1*3, the number of filters is 32, and the size of the pooling kernel of the 1D average pooling layer-2 is 1*3; The convolution kernel size of the 1D convolution layer-3 of the convolution module 3 is 1*3, the number of filters is 64, and the pooling kernel size of the 1D average pooling layer-3 is 1*2; and three 1s The activation functions of the dimensional convolutional layers are all ReLUs.

Specifically, the global feature network includes a Flatten layer and multiple cascaded fully connected layers. First, the multi-time series smart meter data is expanded into a one-dimensional vector by the Flatten layer, and then input to the stacked fully connected layer, using the feature that each neuron of the fully connected layer is connected to the input, to learn and extract the smart meter data. global features.

As shown in FIG. 8 , in this embodiment, the global feature network includes one Flatten layer and two cascaded fully connected layers. Among them, the number of neurons in the fully connected layer-1 is 1000 and the number of neurons in the fully connected layer-2 is 500; and the activation functions of the two fully connected layers are ReLU.

Specifically, the classification network consists of a concatenation layer and multiple cascaded fully connected layers. Among them, the splicing layer splices the local features and global features output by the local feature network and the global feature network into a larger one-dimensional feature vector, and then inputs multiple stacked fully connected layers, and finally calculates and outputs normal power consumption and abnormal power consumption. Type labels of electricity, and the probability of obtaining each type of label.

As shown in FIG. 8 , in this embodiment, the classification network includes a splicing layer and two cascaded fully connected layers. Among them, the number of neurons in the fully connected layer-3 is 1000 and the number of neurons in the fully connected layer-4 is 100. The activation function of the fully connected layer-3 is ReLU, and the activation function of the fully connected layer-4 is Softmax.

It can be seen from the above that except the activation function of the last fully connected layer of the classification network is the Softmax function, the activation functions of the one-dimensional convolutional layer and the fully connected layer of all other networks are ReLU functions.

The abnormal power consumption detection system and its abnormal power consumption detection model provided by the embodiments of the present invention can learn the local and global features of the smart meter data, and more comprehensively extract the high-level features of the power consumption behavior, so as to accurately identify the abnormal power consumption behavior. It can better ensure the long-term stability of abnormal power consumption detection performance and the generalization ability in different application scenarios.

Embodiment three:

The abnormal electricity detection method provided by the present invention is described below through specific experimental examples:

Using the smart meter data of 461 three-phase four-wire 10kV users in a provincial power company as a case, training and verification are carried out.

The smart meter data is acquired by the abnormal power consumption detection method as shown in FIG. 1 provided by the present invention, and converted into the form of multi-time series data; the data collection frequency of each user is 96 collection points/day, and each data collection point Including A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase current, C-phase current, total active power, total power factor and other eight data.

The data sample structure shown in FIG. 3 is obtained by the data sample construction method as shown in FIG. 2 provided by the present invention, and the data is down-sampled, the frequency is changed to 24 collection points/day, and the data items are unchanged; After segmentation, a total of 51,289 samples were obtained, including 35,506 samples of normal power consumption and 15,783 samples of abnormal power consumption.

Using the structure of the deep hybrid neural network shown in Figure 8, the detection model is trained according to the training method shown in Figure 4. The training sample batch size is 200 samples/batch, the learning rate is fixed at 0.001, and the optimizer is Adam, The training round is 50 rounds, so as to obtain the trained abnormal electricity consumption detection model.

Five-fold cross-validation is performed on the abnormal electricity consumption detection model provided by the present invention as shown in Figure 7, and the training samples and test samples of each fold are classified according to 80% and 20% of the normal electricity consumption category and the abnormal electricity consumption category. . The five-fold cross-validation performance of the abnormal power consumption detection model provided by the present invention is shown in the following table:

It can be seen from the results in the above table that the abnormal power consumption detection results using the abnormal power consumption detection method of the first embodiment and the abnormal power consumption detection system of the second embodiment have better abnormal power consumption identification accuracy, and the In the cross-validation experiments, the stability of the recognition performance is shown, and generally good results are obtained.

Embodiment 4:

An embodiment of the present invention provides an abnormal power consumption detection system, including a processor and a storage medium;

the storage medium is used for storing instructions;

The processor is configured to operate according to the instruction to execute the steps of any one of the abnormal power consumption detection methods according to Embodiment 1.

Embodiment 5:

An embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of any one of the abnormal power consumption detection methods of the first embodiment.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the technical principle of the present invention, several improvements and modifications can also be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (15)

1. An abnormal electricity consumption detection method, characterized by comprising the steps of:

acquiring data of the intelligent ammeter to be detected;

converting the data of the intelligent electric meter into multi-time sequence data;

inputting the multi-time sequence data into a trained abnormal electricity utilization detection model to obtain a predicted electricity utilization type label;

and taking the power utilization type with the highest probability in the predicted power utilization type labels as an abnormal power utilization detection result of the data of the intelligent electric meter to be detected.

2. The abnormal electricity utilization detection method according to claim 1, wherein the method for converting the smart meter data into the multi-time series data comprises the steps of:

converting the data of the intelligent electric meter into a multi-time sequence data form, and constructing and attaching a data quality sequence and a date type sequence;

filling missing data into the obtained sequence data, dividing the data according to a natural circumference, and constructing the data into multi-time sequence data;

and carrying out independent data normalization processing on the multi-time sequence data.

3. The abnormal electricity utilization detection method according to claim 2, wherein the method of constructing the data quality sequence comprises the steps of:

and counting the missing number of the data of the intelligent electric meter according to the time stamp, endowing the counted missing number with the corresponding time stamp, and constructing a data quality sequence.

4. The abnormal electricity usage detection method according to claim 2, wherein the method of constructing a date type sequence includes the steps of:

and according to the attribute of the working day or the holiday of each natural day, giving one-per-hour date type data to the natural day to construct a date type sequence.

5. The abnormal electricity utilization detection method according to claim 2, wherein the data normalization processing is performed in a classified normalization manner;

the voltage data were normalized as follows:

the current data were normalized as follows:

the normalization of the total active power is as follows:

the data quality was normalized as follows:

6. the abnormal electricity usage detection method according to claim 1, wherein the training method of the abnormal electricity usage detection model includes the steps of:

acquiring intelligent electric meter data of normal electricity utilization cases and abnormal electricity utilization cases, and converting the intelligent electric meter data into multi-time sequence data samples;

calibrating the electricity type label of the multi-time sequence data sample, and dividing the electricity type label into a training sample set and a verification sample set;

randomly sampling a batch of data samples from a training sample set, inputting the data samples into a deep hybrid neural network, calculating loss and optimizing parameters of the deep hybrid neural network;

traversing the whole training sample set by using a parameter optimized deep hybrid neural network to complete a round of training;

after each training round is finished, evaluating the training effect of the deep hybrid neural network by using the verification sample set, and storing the abnormal power utilization detection model snapshot;

repeating the training to reach the preset number of rounds, so that the loss of the deep hybrid neural network tends to be stable, and finishing the training;

and determining the trained abnormal electricity utilization detection model according to the evaluation result after each round of training.

7. The abnormal electricity usage detection method according to claim 1, wherein the electricity usage type tag acquisition method includes the steps of:

extracting local features of the data of the intelligent electric meter;

extracting global features of the data of the intelligent electric meter;

and splicing and combining the local features and the global features, and converting the local features and the global features into power utilization type labels.

8. The abnormal electricity usage detection method according to any one of claims 1 to 7, wherein the data structure sequence of the multi-time series data includes A-phase voltage, B-phase voltage, C-phase voltage, A-phase current, B-phase current, C-phase current, total active power, total power factor, data quality, and date type.

9. An abnormal electricity utilization detection system, characterized in that the system comprises the following modules:

the data acquisition module is used for acquiring data of the intelligent ammeter to be detected;

the data conversion module is used for converting the data of the intelligent electric meter into multi-time sequence data;

the abnormal electricity utilization detection model is used for obtaining a predicted electricity utilization type label according to the input multi-time sequence data;

and the detection output module is used for taking the electricity utilization type with the highest probability in the predicted electricity utilization type labels as an abnormal electricity utilization detection result of the data of the intelligent electric meter to be detected.

10. The abnormal electricity utilization detection system of claim 9, wherein the abnormal electricity utilization detection model employs a deep hybrid neural network, the deep hybrid neural network comprising a local feature network, a global feature network, and a classification network;

the local feature network is used for extracting local features of the data of the intelligent electric meter;

the global feature network is used for extracting global features of the data of the intelligent electric meter;

and the classification network is used for splicing and combining the local features and the global features and converting the local features and the global features into electricity utilization type labels.

11. The abnormal electricity utilization detection system according to claim 10, wherein the local feature network comprises a plurality of cascaded one-dimensional convolution modules and a Flatten layer;

the one-dimensional convolution module is used for learning and extracting local features of the data of the intelligent ammeter; the Flatten layer is used for expanding the local features extracted by the one-dimensional convolution module and outputting the local features as one-dimensional vectors;

the one-dimensional convolution module comprises a one-dimensional convolution layer and a one-dimensional average pooling layer;

the one-dimensional convolution layer is used for keeping the dimension of the output time axis consistent with the dimension of the input time axis;

and the one-dimensional average pooling layer is used for reducing the dimension of the input time axis on the time axis by adopting an average operator.

12. The abnormal electricity utilization detection system according to claim 10, wherein the global feature network comprises a scatter layer and a plurality of cascaded fully-connected layers;

the Flatten layer is used for unfolding the data of the intelligent electric meter with multiple time sequences into one-dimensional vectors;

and the full connection layer is used for learning and extracting the global features of the intelligent electric meter data from the one-dimensional vectors output by the Flatten layer of the global feature network.

13. The abnormal electricity usage detection system of claim 10, wherein the classification network includes a splice layer and a plurality of cascaded fully-connected layers;

the splicing layer is used for splicing the local features and the global features output by the local feature network and the global feature network into a one-dimensional feature vector and inputting the one-dimensional feature vector into a plurality of stacked full-connection layers;

and the full connection layer is used for calculating and outputting type labels of normal electricity utilization and abnormal electricity utilization.

14. An abnormal electricity utilization detection system is characterized by comprising a processor and a storage medium;

the storage medium is used for storing instructions;

the processor is configured to operate in accordance with the instructions to perform the steps of the method according to any one of claims 1 to 8.

15. Computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 8.

Images