CN112328847A - Transformer overload visualization method and system based on big data - Google Patents

Abstract

The invention discloses a transformer overload visualization method and system based on big data, belongs to the technical field of power grid operation, and aims to realize fault diagnosis and prediction of equipment states by detecting the equipment states and influencing parameter changes from a large amount of data. Firstly, collecting detection data of a power grid and/or a transformer in the power grid, uploading the detection data to a system background, and loading the detection data to a big data platform by the system background; then, cleaning the detection data in the big data platform by using spark, filtering out junk data, converting the detection data after filtering out the junk data into a PRPD or PRPD matrix form, and storing the detection data in the hbase; analyzing and predicting the detection data after filtering the garbage data by utilizing a network model algorithm to obtain the temperature trend of the transformer; and finally, performing data display on the diagnosis result through java web.

Description

Transformer overload visualization method and system based on big data

Technical Field

The invention belongs to the technical field of power grid operation, and relates to transformer overload visualization.

Background

Along with the development of various industries, the demand of power supply continuously meets the urban construction, the scale is rapidly enlarged, the power transmission and transformation plays a crucial role in the power transmission process, and in order to ensure the stability and timeliness of the power transmission and transformation, the load data of the power transmission and transformation is analyzed by using big data analysis through an advanced computer technology, so that the important guiding significance is achieved.

The current power transmission and transformation load analysis is based on traditional statistical analysis to perform data specification, data storage, data calculation and data presentation, and the traditional statistical analysis refers to research activities performed from the combination of quantification and qualification by using a statistical method and knowledge related to an analysis object. When a traditional statistical analysis method is used for analysis and application, the relationship between data distribution and variables needs to be assumed, the probability function is determined to describe the relationship between the variables, and the statistical significance of the parameters is checked to verify whether the assumption is established, so that the automatic searching of the hidden relationship or rule between the variables cannot be realized, and the traditional statistical analysis has low efficiency when processing massive, fuzzy and disordered data, and cannot well support the related applications such as analysis and mining of large data of a distribution transformer.

At present, data related to operation and maintenance of equipment such as transformers have big data characteristics, transformer overload analysis research based on big data is promoted, and overload diagnosis and prediction develop towards the direction based on panoramic visualization and comprehensive analysis. On the other hand, big data analysis technology oriented to data mining, machine learning and knowledge discovery is rapidly developed in recent years, and is widely applied to the fields of internet, social security, telecommunication, finance, commerce, medical treatment and the like. Under the background, it is necessary to fully mine various effective information associated with the equipment state by using an advanced big data analysis processing technology, and to detect the equipment state and the association relation and development rule influencing parameter change from a large amount of data, so as to provide a brand-new solution idea and technical means for fault diagnosis and prediction of the equipment state.

Disclosure of Invention

The invention aims to: the invention provides a transformer overload visualization method and system based on big data, which can realize fault diagnosis and prediction of equipment states by detecting the equipment states and influencing parameter changes from a large amount of data.

The technical scheme adopted by the invention is as follows:

a transformer overload visualization method based on big data comprises the following steps:

step S1, collecting detection data of the power grid and/or the transformer in the power grid, uploading the detection data to a system background, and loading the detection data to a big data platform by the system background; the system background preferably selects the HDFS system.

Step S2, cleaning the detection data in the big data platform by spark, filtering the garbage data, converting the detection data after filtering the garbage data into PRPD or PRPD matrix form and storing in hbase;

step S3, analyzing and predicting the detection data after filtering the garbage data by using a network model algorithm to obtain the temperature trend of the transformer;

the detection data includes: the method comprises the following steps of (1) transformer station name, transformer model, transformer manufacturer, transformer operation time, transformer phase, transformer discharge capacity, transformer discharge position, transformer discharge pulse number, transformer discharge waveform, transformer discharge pulse sequence map and transformer discharge pulse phase map;

in step S3, the network model algorithm specifically includes:

step S31, constructing network model

And step S32, inputting the detection data of the last three days as sample data into a network model, wherein the output data of the network model is a predicted value, and the temperature trend of the transformer is obtained.

The network model is a network structure including four layers of neurons, where the input vector is X ═ X₁，x₂，…，x_m]^TThe corresponding output vector is Y ═ Y₁，y₂，…，y_k]；

An input layer: the number of neurons is equal to the dimension m of the input vector, the input layer passes the variables directly forward to the mode layer:

mode layer: the number of neurons equals the number of input samples ", different neurons correspond to different learning samples, and the transfer function of the pattern layer neuron i is

Wherein, Pi Gaussian kernel function, X is a deprecated network input vector, Xi is a learning sample corresponding to the neuron i, and sigma is a smoothing factor of the network;

and a summation layer: the layer comprises two types of neurons, one is to perform arithmetic summation on the outputs of all mode layers, the summed value and the connection weight value between the neurons of the mode layers are 1, and the transfer function is as follows:

another neuron is the weighted sum of the outputs of all mode layers, where S_NjThe connection weight of is the ith output sample Y_iThe jth element y in (2)_ijThe transfer function is:

an output layer: the number of neurons is equal to the dimension k of the output vector, and the weighted summation output of the neurons in the summation layer is divided by the arithmetic summation output to obtain the output corresponding to each neuron, namely:

the step of determining the smoothing factor in the mode layer comprises the following steps:

step 1, after sample data is classified, normalization processing is carried out as required, and a normalization formula is as follows:

wherein x is_max、x_minThe formula normalizes the data to [ a, b ] for the maximum and minimum values in a set of data]；

Step 2, making the smoothing factor at [ min, max]Within the range, the step size is changed gradually to obtain a set of smoothing factors sigma₁，σ₂，…，σ_p；

Step 3, for each smoothing factor σ_i(i ═ 1, 2, …, P) corresponds to an n × k dimensional error matrix error, k is the dimension of the output vector, the mean square value is calculated to obtain n error mean square values P-error, the finally obtained P × n error mean square values are sorted from small to large, and the first (pxn)/2 are taken out;

step 4, giving a value range [ K ] of K_min，K_max]In which K is_min＞o，K_maxNot more than (p multiplied by n)/2, the K value is changed in an incremental way according to a certain step length, the first K error mean square values are taken out in each round, and the distribution condition is judged;

and 5, counting the smoothing factor with the maximum occurrence frequency.

Finally, in step S4, the diagnosis result is presented via java web.

A big data based transformer overload visualization system, comprising:

the data acquisition module is used for acquiring detection data of the power grid and/or a transformer in the power grid and uploading the detection data to the system background, and the system background loads the detection data to the big data platform;

the data processing module is used for cleaning the detection data in the big data platform by using spark, filtering out junk data, converting the detection data with the filtered junk data into a PRPD or PRPD matrix form and storing the detection data in the hbase;

the data analysis and prediction module is used for analyzing and predicting the detection data after the garbage data is filtered by utilizing a network model algorithm to obtain the temperature trend of the transformer;

and the data display module is used for displaying the diagnosis result through java web.

In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:

in the invention, the HDFS system is adopted as the system with the station, the system has high fault tolerance and can be deployed on a cheap machine; the method can provide high-throughput data access, is very suitable for application on large-scale data sets, realizes low-cost and high-throughput data access of a visualization method and a visualization system, and enables the visualization method and the visualization system to be provided; the HDFS system, the spark technology and the spark mlib deep learning and diagnosis algorithm are cooperatively used, java web is matched for data display of diagnosis results, various effective information related to equipment states is fully mined by utilizing the advanced big data analysis and processing technology, the association relation and the development rule of the equipment states and the influence parameter changes are explored from a large amount of data, and fault diagnosis and prediction of the running state of the transformer are achieved. When data is analyzed and predicted, a network model algorithm is adopted, so that the prediction performance is better, the prediction precision is high, the prediction speed is high, and the applicability is better and wider.

Drawings

The invention will now be described, by way of example, with reference to the accompanying drawings, in which:

FIG. 1 is a technical architecture and flow diagram of the present invention;

FIG. 2 is a schematic diagram of the overload display of the present invention;

FIG. 3 is a schematic diagram of the structure of a defect word cloud according to the present invention; the term cloud visually highlights the keywords with high frequency of occurrence in the data to form a keyword cloud layer or keyword rendering, so that a large amount of text information is filtered, and an application can draw the subject of the text as long as the application sweeps the text at a glance. The defect names are displayed through the defect word cloud, the more the number of the defects appears, the larger the name of the manufacturer is, and the purpose is to highlight the relation between the defects and the manufacturer;

FIG. 4 is a schematic bar diagram of a defect of the present invention; wherein, the percentage diagram can be displayed under the conditions of ensuring the aesthetic degree and strong visibility by using the stacked column diagram. In application, defect proportion of equipment model manufacturers is shown by stacking column diagrams, and the more the number of defective equipment models at the manufacturers is, the more the proportion is, the larger the height in the column diagrams is.

FIG. 5 is a graph of the relationship between data defects and manufacturers according to the present invention;

FIG. 6 is another graph of the relationship between data defects and manufacturers according to the present invention, wherein a percentage graph can be displayed under the condition of ensuring good aesthetic appearance and high visibility by using stacked column graphs. In application, defect proportion of equipment model manufacturers is shown by stacking column diagrams, and the more the number of defective equipment models at the manufacturers is, the more the proportion is, the larger the height in the column diagrams is.

Detailed Description

All of the features disclosed in this specification, or all of the steps in any method or process so disclosed, may be combined in any combination, except combinations of features and/or steps that are mutually exclusive.

A transformer overload visualization method based on big data comprises the following steps:

step S1, collecting detection data of the power grid and/or the transformer in the power grid, uploading the detection data to a system background, and loading the detection data to a big data platform by the system background;

the background of the system adopts the HDFS which is a high fault-tolerant system and is suitable for being deployed on a cheap machine. HDFS provides high throughput data access and is well suited for application on large-scale data sets. Which in this application serves as the bottom layer of storage.

Step S2, cleaning the detection data in the big data platform by spark, filtering the garbage data, converting the detection data after filtering the garbage data into PRPD or PRPD matrix form and storing in hbase;

HBase is a distributed, column-oriented open-source database that provides Bigtable-like capabilities. The method and the device can be used as a storage library of mass data, and can provide rapid data reading and writing service. The Spark does not need to read and write the HDFS any more, so the Spark can be better suitable for MapReduce algorithms which need iteration, such as data mining, machine learning and the like. Which in the present application serves as a background data etl tool and a computational engine for intelligent diagnostics and matching algorithms.

Step S3, analyzing and predicting the detection data after filtering the garbage data by using a network model algorithm to obtain the temperature trend of the transformer;

the detection data includes: the method comprises the following steps of (1) transformer station name, transformer model, transformer manufacturer, transformer operation time, transformer phase, transformer discharge capacity, transformer discharge position, transformer discharge pulse number, transformer discharge waveform, transformer discharge pulse sequence map and transformer discharge pulse phase map;

in step S3, the network model algorithm specifically includes:

step S31, constructing network model

And step S32, inputting the detection data of the last three days as sample data into a network model, wherein the output data of the network model is a predicted value, and the temperature trend of the transformer is obtained.

The network model is a network structure including four layers of neurons, where the input vector is X ═ X₁，x₂，…，x_m]^TThe corresponding output vector is Y ═ Y₁，y₂，…，y_k]；

An input layer: the number of neurons is equal to the dimension m of the input vector, the input layer passes the variables directly forward to the mode layer:

mode layer: the number of neurons equals the number of input samples "Different neurons correspond to different learning samples, and the transfer function of the pattern layer neuron i is

Wherein, Pi Gaussian kernel function, X is a deprecated network input vector, Xi is a learning sample corresponding to the neuron i, and sigma is a smoothing factor of the network;

and a summation layer: the layer comprises two types of neurons, one is to perform arithmetic summation on the outputs of all mode layers, the summed value and the connection weight value between the neurons of the mode layers are 1, and the transfer function is as follows:

another neuron is the weighted sum of the outputs of all mode layers, where S_NjThe connection weight of is the ith output sample Y_iThe jth element y in (2)_ijThe transfer function is:

an output layer: the number of neurons is equal to the dimension k of the output vector, and the weighted summation output of the neurons in the summation layer is divided by the arithmetic summation output to obtain the output corresponding to each neuron, namely:

the step of determining the smoothing factor in the mode layer comprises the following steps:

step 1, after sample data is classified, normalization processing is carried out as required, and a normalization formula is as follows:

wherein x is_max、x_minThe formula normalizes the data to [ a, b ] for the maximum and minimum values in a set of data]；

Step 2, making the smoothing factor at [ min, max]Within the range, the step size is changed gradually to obtain a set of smoothing factors sigma₁，σ₂，…，σ_p；

Step 3, for each smoothing factor σ_i(i ═ 1, 2, …, P) corresponds to an n × k dimensional error matrix error, k is the dimension of the output vector, the mean square value is calculated to obtain n error mean square values P-error, the finally obtained P × n error mean square values are sorted from small to large, and the first (pxn)/2 are taken out;

step 4, giving a value range [ K ] of K_min，K_max]In which K is_min＞o，K_maxNot more than (p multiplied by n)/2, the K value is changed in an incremental way according to a certain step length, the first K error mean square values are taken out in each round, and the distribution condition is judged;

and 5, counting the smoothing factor with the maximum occurrence frequency.

And step S4, displaying the diagnosis result through java web.

The transformer overload visualization display based on big data utilizes a multi-dimensional comprehensive data display mode to carry out comprehensive statistical analysis on transformer defects, manufacturers and equipment, and a multi-angle transformer overload visualization system is established for users. Including overload displays, defect word clouds, transformer-to-manufacturer relationship diagrams, bubble diagrams, etc.

In addition, the application also provides a transformer overload visualization system based on big data, and the system comprises a data acquisition module, a data processing module, a data analysis and prediction module and a data display module;

the data acquisition module is used for acquiring detection data of the power grid and/or a transformer in the power grid and uploading the detection data to the system background, and the system background loads the detection data to the big data platform;

the data processing module is used for cleaning the detection data in the big data platform by using spark, filtering out junk data, converting the detection data with the filtered junk data into a PRPD or PRPD matrix form and storing the detection data in the hbase;

the data analysis and prediction module is used for analyzing and predicting the detection data after the garbage data is filtered by utilizing a network model algorithm to obtain the temperature trend of the transformer;

and the data display module is used for displaying the diagnosis result through java web.

Claims (7)

1. A transformer overload visualization method based on big data is characterized by comprising the following steps

Step S1, collecting detection data of the power grid and/or the transformer in the power grid, uploading the detection data to a system background, and loading the detection data to a big data platform by the system background;

step S2, cleaning the detection data in the big data platform by spark, filtering the garbage data, converting the detection data after filtering the garbage data into PRPD or PRPD matrix form and storing in hbase;

step S3, analyzing and predicting the detection data after filtering the garbage data by using a network model algorithm to obtain the temperature trend of the transformer;

and step S4, displaying the diagnosis result through java web.

2. The big-data-based transformer overload visualization method according to claim 1, wherein in step S1, the HDFS system is selected as the system background.

3. The big data based transformer overload visualization method according to claim 1,

the detecting data includes: the method comprises the following steps of transformer station name, transformer model, transformer manufacturer, transformer operation time, transformer phase, transformer discharge amount, transformer discharge position, transformer discharge pulse number, transformer discharge waveform, transformer discharge pulse sequence map, transformer discharge pulse phase map and transformer temperature.

4. The big-data-based transformer overload visualization method according to claim 1, wherein in the step S3, the network model algorithm specifically includes:

step S31, constructing network model

And step S32, inputting the detection data of the last three days as sample data into a network model, wherein the output data of the network model is a predicted value, and the temperature trend of the transformer is obtained.

5. The big-data-based transformer overload visualization method according to claim 4, wherein the network model is a network structure including four layers of neurons, and an input vector is X ═ X₁，x₂，…，x_m]^TThe corresponding output vector is Y ═ Y₁，y₂，…，y_k]；

An input layer: the number of the neurons is equal to the dimension m of the input vector, and the input layer directly forwards transmits each variable to the mode layer;

mode layer: the number of neurons equals the number of input samples ", different neurons correspond to different learning samples, and the transfer function of the pattern layer neuron i is

Wherein, Pi Gaussian kernel function, X is a deprecated network input vector, Xi is a learning sample corresponding to the neuron i, and sigma is a smoothing factor of the network;

and a summation layer: the layer comprises two types of neurons, one is to perform arithmetic summation on the outputs of all mode layers, the summed value and the connection weight value between the neurons of the mode layers are 1, and the transfer function is as follows:

another neuron is the weighted sum of the outputs of all mode layers, where S_NjThe connection weight value of is ithAn output sample Y_iThe jth element y in (2)_ijThe transfer function is:

an output layer: the number of neurons is equal to the dimension k of the output vector, and the weighted summation output of the neurons in the summation layer is divided by the arithmetic summation output to obtain the output corresponding to each neuron, namely:

6. the big-data-based transformer overload visualization method according to claim 5, wherein the step of determining the smoothing factor in the mode layer comprises:

step 1, after sample data is classified, normalization processing is carried out as required, and a normalization formula is as follows:

wherein x is_max、x_minThe formula normalizes the data to [ a, b ] for the maximum and minimum values in a set of data]；

Step 2, making the smoothing factor at [ min, max]Within the range, the step size is changed gradually to obtain a set of smoothing factors sigma₁，σ₂，…，σ_p；

Step 3, for each smoothing factor σ_i(i ═ 1, 2, …, P) corresponds to an n × k dimensional error matrix error, k is the dimension of the output vector, the mean square value is calculated to obtain n error mean square values P-error, the finally obtained P × n error mean square values are sorted from small to large, and the first (P × n)/2 error mean square values are taken out;

step 4, giving a value range [ K ] of K_min，K_max]In which K is_min＞o，K_maxNot more than (p multiplied by n)/2, the K value is changed in an incremental way according to a certain step length, the first K error mean square values are taken out in each round, and the distribution condition is judged;

and 5, counting the smoothing factor with the maximum occurrence frequency.

7. A big data based transformer overload visualization system, comprising:

the data acquisition module is used for acquiring detection data of the power grid and/or a transformer in the power grid and uploading the detection data to the system background, and the system background loads the detection data to the big data platform;

the data processing module is used for cleaning the detection data in the big data platform by using spark, filtering out junk data, converting the detection data with the filtered junk data into a PRPD or PRPD matrix form and storing the detection data in the hbase;

the data analysis and prediction module is used for analyzing and predicting the detection data after the garbage data is filtered by utilizing a network model algorithm to obtain the temperature trend of the transformer;

and the data display module is used for displaying the diagnosis result through java web.

Images