CN116757465A - Line risk assessment method and device based on double training weight distribution model - Google Patents

Line risk assessment method and device based on double training weight distribution model Download PDF

Info

Publication number
CN116757465A
CN116757465A CN202310434199.8A CN202310434199A CN116757465A CN 116757465 A CN116757465 A CN 116757465A CN 202310434199 A CN202310434199 A CN 202310434199A CN 116757465 A CN116757465 A CN 116757465A
Authority
CN
China
Prior art keywords
data
load
target
prediction data
load prediction
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
Application number
CN202310434199.8A
Other languages
Chinese (zh)
Inventor
王晔辰
卢纯义
黄健
杨震
侯虎成
于津
吕默影
范彬彬
张元吉
刘晓谦
郑腾飞
吴敏航
宋俊杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jinhua Power Supply Co of State Grid Zhejiang Electric Power Co Ltd
Original Assignee
Jinhua Power Supply Co of State Grid Zhejiang Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jinhua Power Supply Co of State Grid Zhejiang Electric Power Co Ltd filed Critical Jinhua Power Supply Co of State Grid Zhejiang Electric Power Co Ltd
Priority to CN202310434199.8A priority Critical patent/CN116757465A/en
Publication of CN116757465A publication Critical patent/CN116757465A/en
Pending legal-status Critical Current

Links

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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • 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
    • G06N3/0499—Feedforward 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
    • 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
    • G06Q10/04—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
    • 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
    • G06Q10/20—Administration of product repair or maintenance
    • 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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Game Theory and Decision Science (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Development Economics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Educational Administration (AREA)
  • Primary Health Care (AREA)

Abstract

The invention discloses a line risk assessment method and device based on a double training weight distribution model, and relates to the technical field of power distribution network line risk prediction, wherein the method comprises the following steps: acquiring historical load data of an object under different working conditions, preprocessing the historical load data to obtain historical load characteristic data, inputting a target cyclic neural network model and a target autoregressive moving average model for training, and respectively obtaining first load prediction data and second load prediction data; performing weight optimization on the first load prediction data and the second load prediction data through a BP neural network algorithm to obtain a first target weight and a second target weight; and calculating target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, and determining a line risk index. The method and the device can capture more accurate target load prediction data, thereby improving the accuracy of the load prediction result and the accuracy of line risk assessment.

Description

Line risk assessment method and device based on double training weight distribution model
Technical Field
The invention relates to the technical field of power distribution network line risk prediction, in particular to a line risk assessment method and device based on a double training weight distribution model.
Background
The risk assessment of the distribution network line mainly depends on intelligent load prediction, including prediction according to social and economic development, population growth, distribution network operation characteristics and other factors. The main content of load prediction comprises the following four types: the power consumption and maximum load of the whole society, the maximum load of each voltage class network supply, the saturated load space time distribution of the power distribution network and the maximum load of power users. For a branch line in a certain power grid, the maximum load of a user is particularly important, and the main basis of the power supply voltage level, the selected electrical equipment and the conductor material is further determined.
In recent years, a machine learning training method related to a neural network is gradually raised, and particularly, the large data of a power distribution network are used for load prediction, electric automobile charging equipment requirements, power supply reliability analysis, user participation demand response potential analysis and other scenes, so that the values of supporting rules and operation of the power distribution network, improving an information system, improving user satisfaction and the like are realized. At present, some researches have been conducted on the application of large data of a power distribution network to various technical fields of the power distribution network. For the large data processing of the power distribution network, a learner provides a data cleaning method suitable for a distribution transformer according to the characteristics of distribution transformer load data, and performs parallel calculation of missing data recovery flow under a Spark calculation engine. In the field of load prediction, a learner performs parallelization transformation on a random forest algorithm in a MapReduce mode according to the characteristics of data on the power consumer side, and performs data storage and management in HBase and HDFS respectively so as to realize parallel load prediction under a Hadoop architecture. In the field of power distribution network state evaluation, scholars also propose how to use big data for power distribution network state detection and fault handling, operation evaluation on the power distribution network side, reliability evaluation, and the like. However, in general, the research about the application of the big data of the power distribution network in the power distribution network planning is still less at present, and the method for predicting the load is always single, so that the problem of low prediction accuracy of the load prediction result exists, and the risk assessment result of the line is affected.
Disclosure of Invention
The invention aims to solve the problem of low accuracy of a load prediction result when line risk assessment is carried out in a power distribution network, and provides a line risk assessment method based on a double training weight distribution model, which can be used for respectively predicting through double models based on big data, and capturing more accurate target load prediction data, so that the accuracy of the load prediction result and the accuracy of line risk assessment are improved, powerful data support is provided for making countermeasures, and the intelligent operation and maintenance efficiency of the power distribution network is improved.
In a first aspect, a technical solution provided in an embodiment of the present invention is a line risk assessment method based on a dual training weight distribution model, where the method includes the following steps:
acquiring historical load data of an object under different working conditions, and preprocessing the historical load data to obtain historical load characteristic data;
inputting the historical load characteristic data into a target cyclic neural network model for training to obtain first load prediction data, and inputting the historical load characteristic data into a target autoregressive moving average model for training to obtain second load prediction data; performing weight optimization on the first load prediction data and the second load prediction data through a BP neural network algorithm to obtain a first target weight and a second target weight;
Calculating target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, and determining a line risk index according to the target load prediction data.
Optionally, the obtaining historical load data of the object under different working conditions, preprocessing the historical load data to obtain historical load feature data, includes:
acquiring the historical load data of the object under different working conditions, wherein the historical load data comprises acquisition of line periodic load data and line special working condition load data, and the line special working condition load data comprises date node load data, weather node load data and customer complaint work order load data;
according to a preset preprocessing rule, the date node load data, the weather node load data and the customer complaint work order load data are standardized;
constructing a sample data set based on the line cycle load data, the standardized date node load data, the weather node load data and the customer complaint work order load data;
and carrying out feature extraction on the data in the sample data set through a convolutional neural network to obtain the historical load feature data.
Optionally, before the normalizing the date node load data, the weather node load data and the customer complaint work order load data according to the preset preprocessing rule, the method further includes:
acquiring the customer complaint work order load data in a preset time interval;
and according to the preset complaint content keywords, performing irrelevant data clearing on the customer complaint work order load data, and merging customer complaint work orders with the same complaint content.
Optionally, the step of inputting the historical load characteristic data into a target cyclic neural network model for training to obtain first load prediction data includes:
constructing a time sequence of the historical load characteristic data, wherein the time sequence comprises a plurality of time nodes, and each time node has corresponding historical load characteristic data;
taking the historical load characteristic data corresponding to the current time node as characteristic input data of the target cyclic neural network model, and initializing a weight coefficient of the target cyclic neural network model;
determining a reset gate and an update gate of the target cyclic neural network model at the current time node through a hyperbolic tangent function according to the characteristic input data of the current time node, the hidden state data of the previous time node and the weight coefficient;
And outputting the first load prediction data according to the reset gate and the update gate of the current time node.
Optionally, before the historical load characteristic data is input into the target autoregressive moving average model for training to obtain second load prediction data, the method further includes:
calculating an autocorrelation coefficient and a partial autocorrelation coefficient according to the time sequence of the historical load characteristic data, and determining the type of a target model based on the autocorrelation coefficient and tailing or truncating of the partial autocorrelation coefficient;
performing pre-estimation test on parameters in the target model, and optimizing time sequence stationarity of the historical load characteristic data, wherein the pre-estimation test comprises deterministic factor parameter analysis of the target model and random time sequence parameter analysis of the model;
fitting the target model according to the time sequence of the historical load characteristic data after stability optimization, and constructing to obtain the target autoregressive moving average model.
Optionally, the BP neural network algorithm includes an input layer, a hidden layer, and an output layer, and the performing weight optimization on the first load prediction data and the second load prediction data by the BP neural network algorithm includes:
Inputting the first load prediction data and the second load prediction data into an input layer of the BP neural network algorithm respectively, and constructing a BP neural network algorithm model by combining parameters from the input layer to the hidden layer and parameters from the hidden layer to an output layer, wherein the parameters comprise weights and bias items;
initializing weight and bias items of the BP neural network algorithm model, and iterating expected values of output and loss functions of each layer for a plurality of times through back propagation;
determining parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer in the BP neural network algorithm model according to expected values of the loss function;
and updating weights and bias items among layers in the BP neural network algorithm model according to parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer until a preset iteration condition is met, and stopping updating to obtain the first target weight, the second target weight, the first bias item and the second bias item.
Optionally, the calculating the target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, and determining the line risk index according to the target load prediction data includes:
If the expected value of the loss function meets the preset expected value, determining target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, wherein the sum of the first target weight and the second target weight is equal to 1;
and determining the line risk index according to the target load prediction data based on a preset corresponding relation between the load data and the risk index.
In a second aspect, a technical solution provided in an embodiment of the present invention is a line risk assessment device based on a dual training weight distribution model, where the device includes:
the acquisition module is used for acquiring historical load data of the object under different working conditions, and preprocessing the historical load data to obtain historical load characteristic data;
the training module is used for inputting the historical load characteristic data into a target cyclic neural network model for training to obtain first load prediction data, and inputting the historical load characteristic data into a target autoregressive moving average model for training to obtain second load prediction data;
the optimization module is used for carrying out weight optimization on the first load prediction data and the second load prediction data through a BP neural network algorithm to obtain a first target weight and a second target weight;
And the calculating module is used for calculating target prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight and determining a line risk index according to the target prediction data.
In a third aspect, a technical solution provided in an embodiment of the present application is an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, which when executed implements the steps in a line risk assessment method based on a dual training weight distribution model as described in any of the embodiments.
In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, where a computer program is stored, where the computer program when executed by a processor implements the steps in a line risk assessment method based on a dual training weight distribution model according to any one of the embodiments.
The application has the beneficial effects that: according to the method, the obtained historical load data are preprocessed in advance, so that the historical load characteristic data which are suitable for the model input can be obtained; the historical load characteristic data is trained in a double-model mode through the target cyclic neural network model and the target autoregressive moving average model respectively, and first load prediction data and second load prediction data are output, wherein the target cyclic neural network model has few parameters and low time cost of data calculation, and the target autoregressive moving average model can realize regression analysis and short-term prediction of a large amount of data; and finally, carrying out weight distribution on the first load prediction data and the second load prediction data through a BP neural network algorithm, optimizing the network weight through repeated updating and iteration, finding out optimal first target weight and second target weight for the first load prediction data and the second load prediction data, finally, calculating target load prediction data by combining the first load prediction data, the first target weight, the second load prediction data and the second target weight, and determining a line risk index according to the target load prediction data. Therefore, the method and the system predict by combining the double models, calculate the target load prediction data after fitting the prediction results of the double models to the optimal weights respectively, overcome the limitation of a single training model, and can capture more accurate target load prediction data by the prediction of the double models, thereby improving the accuracy of the load prediction results and the accuracy of line risk assessment, further providing powerful data support for making countermeasures, and improving the efficiency of intelligent operation and maintenance of the distribution network.
The foregoing summary is merely an overview of the present invention, and is intended to be implemented in accordance with the teachings of the present invention in order that the same may be more fully understood, and in order that the same or additional objects, features and advantages of the present invention may be more fully understood.
Drawings
Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures.
FIG. 1 is a flow chart of a line risk assessment method based on a dual training weight distribution model according to an embodiment of the present invention;
fig. 2 is an input/output structure diagram of a target cyclic neural network model according to an embodiment of the present invention;
FIG. 3 is a flowchart of obtaining a target autoregressive moving average model according to an embodiment of the present invention;
fig. 4 is a specific workflow diagram of a BP neural network algorithm according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a line risk assessment device based on a dual training weight distribution model according to an embodiment of the present invention;
Fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings and examples, it being understood that the detailed description herein is merely a preferred embodiment of the present invention, which is intended to illustrate the present invention, and not to limit the scope of the invention, as all other embodiments obtained by those skilled in the art without making any inventive effort fall within the scope of the present invention.
Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart depicts operations (or steps) as a sequential process, many of the operations (or steps) can be performed in parallel, concurrently, or at the same time. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figures; the processes may correspond to methods, functions, procedures, subroutines, and the like.
The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. It should also be understood that, in various embodiments of the present invention, the sequence number of each process does not mean the order of execution, and the order of execution of each process should be determined by its functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
It should be understood that in the present invention, "plurality" means two or more. "and/or" is merely a variable relationship describing an associated object, meaning that there may be three relationships, e.g., and/or B, may represent: a exists alone, A and B exist together, and B exists alone. The character "/" generally indicates that the context-dependent object is an "or" relationship. "comprising A, B and C", "comprising A, B, C" means that all three of A, B, C comprise, "comprising A, B or C" means that one of the three comprises A, B, C, and "comprising A, B and/or C" means that any 1 or any 2 or 3 of the three comprises A, B, C.
It should be understood that in the present invention, "B corresponding to a", "a corresponding to B", or "B corresponding to a" means that B is associated with a, from which B can be determined. Determining B from a does not mean determining B from a alone, but may also determine B from a and/or other information. The matching of A and B is that the similarity of A and B is larger than or equal to a preset threshold value.
Example 1
Referring to fig. 1, fig. 1 is a flowchart of a line risk assessment method based on a dual training weight distribution model according to an embodiment of the present invention. A line risk assessment method based on a dual training weight distribution model, the method comprising the steps of:
s101, acquiring historical load data of an object under different working conditions, and preprocessing the historical load data to obtain historical load characteristic data.
The line risk assessment method based on the double-training weight distribution model can be used in scenes such as load prediction of power supply equipment of a power supply bureau, and the electronic equipment operated by the line risk assessment method based on the double-training weight distribution model can communicate with other electronic equipment through a network in a wired connection mode or a wireless connection mode, so that functions such as data transmission are realized. It should be noted that the wireless connection may include, but is not limited to, 3G/4G/5G connection, wiFi connection, bluetooth connection, wiMAX connection, zigbee connection, UWB (ultra wideband) connection, and other now known or later developed wireless connection.
Specifically, the object may refer to a main body to be predicted, which needs load prediction, and may include, but is not limited to, a power supply office, a transformer substation, a power supply transfer station, a power station, and the like in each region. The working conditions can respectively correspond to factors such as weather, external force, specific events, user complaints and the like, so that historical load data of an object in a circuit of the power distribution network in different factor environments within a certain time period can be obtained, and the load data can refer to power data of the circuit. The historical load data can refer to load data respectively generated under different working conditions in a certain past time period, and the load data under different working conditions in the same time period can form a model input sample. The preprocessing of the historical load data may be to normalize the disordered historical load data by a certain processing rule to obtain the historical load characteristic data so as to be convenient for predicting as the input data of the model.
S102, inputting the historical load characteristic data into a target cyclic neural network model for training to obtain first load prediction data, and inputting the historical load characteristic data into a target autoregressive moving average model for training to obtain second load prediction data.
Specifically, the target recurrent neural network model may refer to a recurrent neural network model after training is completed, and in this embodiment, the target GRU (Gated Recurrent Unit) model after training is completed may be specifically described. Compared with LSTM, the target GRU model has fewer internal gating and fewer parameters, and can effectively capture semantic association between long sequences and relieve gradient disappearance or explosion phenomenon. In this embodiment, because the historical load characteristic data is load data based on a time sequence, each time node corresponds to load data, the historical load characteristic data can be input into the target GRU model, semantic association between load data in the time sequence is captured through the target GRU model, so that load data prediction in a future period is realized, and finally first load prediction data is output from the target GRU model.
More specifically, the target autoregressive moving average model may refer to an autoregressive moving average model (target ARMA model) obtained after training is completed, and the target ARMA model may be used for regression analysis and short-term prediction of a large amount of data, and is composed of an Autoregressive (AR) model and a Moving Average (MA) model. And inputting the preprocessed historical load characteristic data into a target ARMA model for training, and finally outputting second load prediction data from the target ARMA model. Thus, the double-model realizes twice prediction on the same historical load data, and outputs two load prediction data.
And S103, carrying out weight optimization on the first load prediction data and the second load prediction data through a BP neural network algorithm to obtain a first target weight and a second target weight.
Specifically, the BP neural network is a multi-layer feedforward network trained according to error back propagation (error back propagation for short), the algorithm is called as BP algorithm (BP neural network algorithm), the basic idea is gradient descent method, and gradient search technology is utilized to minimize the expected value of the loss function of the network. And finally adjusting the weight of the network. The BP neural network comprises an input layer, a hidden layer and an output layer, wherein weight and bias parameters are included between the layers, each layer comprises a plurality of neurons, the number of layers of the hidden layer can be customized according to the needs, and in the embodiment, one layer is taken as an example.
More specifically, in order to improve the accuracy of the load prediction data, in this embodiment, the first load prediction data and the second load prediction data are used as inputs of a BP neural network algorithm, after multiple iterative computations, the parameter configuration of the BP neural network is optimized, the expected value of the loss function is minimum, and then the weights from the input layer to the hidden layer and the circles from the hidden layer to the output layer of the network when the corresponding parameter configuration is optimized and the expected value of the loss function is minimum are respectively used as the first target weight of the first load prediction data and the second target weight of the second load prediction data. The target weights are respectively configured after the first load prediction data and the second load prediction data are calculated through the double models, so that the prediction accuracy of the target load prediction data can be improved.
S104, calculating target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, and determining a line risk index according to the target load prediction data.
Specifically, the target load prediction data may be finally obtained by multiplying the first load prediction data by the first target weight, multiplying the second load prediction data by the second target weight, and then summing the multiplied second load prediction data. The target load prediction data and the line risk index have a preset corresponding relation, for example: when the target load prediction data is in the sequentially increasing ranges a, b and c, the line risk levels corresponding to the ranges a, b and c are respectively low, medium and high; or the corresponding line risk level is 1 level, 2 level and 3 level with sequentially increased risk. Of course, other representations are possible and are not limited solely herein. Therefore, the line risk index can be judged according to the target load prediction data, and the line risk assessment is realized, so that the line control strategy can be formulated in advance according to the judgment result.
In the embodiment of the application, the obtained historical load data is preprocessed in advance, so that the historical load characteristic data which is suitable for the model input can be obtained; the historical load characteristic data are trained in a double-model mode through the target GRU model and the target ARMA model respectively, and first load prediction data and second load prediction data are output, wherein the target GRU model has few parameters and low time cost of data calculation, and the target ARMA model can realize regression analysis and short-term prediction of a large amount of data; and finally, carrying out weight distribution on the first load prediction data and the second load prediction data through a BP neural network algorithm, optimizing the network weight through repeated updating and iteration, finding out optimal first target weight and second target weight for the first load prediction data and the second load prediction data, finally, calculating target load prediction data by combining the first load prediction data, the first target weight, the second load prediction data and the second target weight, and determining a line risk index according to the target load prediction data. Therefore, the method and the system predict by combining the double models, calculate the target load prediction data after fitting the prediction results of the double models to the optimal weights respectively, overcome the limitation of a single training model, and can capture more accurate target load prediction data by the prediction of the double models, thereby improving the accuracy of the load prediction results and the accuracy of line risk assessment, further providing powerful data support for making countermeasures, and improving the efficiency of intelligent operation and maintenance of the distribution network.
Optionally, the step S101 includes:
s1011, acquiring the historical load data of the object under different working conditions, wherein the historical load data comprises acquisition of line periodic load data and line special working condition load data, and the line special working condition load data comprises date node load data, weather node load data and customer complaint work order load data.
Specifically, the line cycle load data may include load data of the line under a normal working condition and a fault working condition within a preset time range, where the preset time range may be one month, half year or year. The date node load data may include load data generated by a change in line load due to a date factor of an important activity. The weather node load data may include load data generated due to a change in line load due to weather factors, for example: plum rainy season, three-volt day, winter snowy season, etc. The customer complaint work order load data may refer to load data provided on a customer complaint work order. In addition, load data of the line under random events and extreme events can be obtained, for example: equipment damage, overload outage, load data when natural disaster accidents occur, and the like.
S1012, according to a preset preprocessing rule, the date node load data, the weather node load data and the customer complaint work order load data are standardized.
Specifically, because the working condition types of the load data are different, in order to realize standardization, a preprocessing rule can be set, standard processing is performed on different working condition types based on the preprocessing rule, including standardized processing is performed on date node load data, weather node load data and customer complaint work order load data, for example: calibrating the date factor into a digital format; the weather factors are calibrated by numbers, the good weather is calibrated to be 1, the general weather is calibrated to be 2, and the bad weather is calibrated to be 3, wherein the good weather represents sunny days and cloudy days, the general weather represents light rain and light snow, and the bad weather represents severe weather such as heavy rain, heavy snow, heavy wind, heavy rain and the like; extracting text comprising lines from a customer complaint work order, setting priority according to data of complaint lines in the complaint work order, and processing lines with high priority preferentially.
S1013, constructing a sample data set based on the line cycle load data, the standardized date node load data, the weather node load data and the customer complaint work order load data.
And S1014, carrying out feature extraction on the data in the sample data set through a convolutional neural network to obtain the historical load feature data.
Specifically, the load data of the line under the normal working condition and the fault working condition in the preset time interval, the standardized date node load data, the standardized weather node load data and the standardized customer complaint work order load data can be combined, a characteristic sample data set is constructed based on a time sequence, characteristic extraction is carried out on each characteristic sample data in the characteristic sample data set through a convolutional neural network, so that a high-dimensional prediction characteristic vector is formed in a high-dimensional space, namely, the historical load characteristic data expressed by the vector is finally input into a target GRU model and a target ARMA model for prediction.
In the embodiment, load data can be predicted in a more azimuth manner by collecting the load data under different working conditions, and the standardization of the data can be ensured by preprocessing the historical load data, so that the target GRU model and the target ARMA model can be identified quickly.
In one possible embodiment, before the step S1012, the method further includes:
s1015, acquiring the customer complaint work order load data in a preset time interval.
S1016, according to the preset complaint content keywords, irrelevant data clearing is carried out on the customer complaint work order load data, and customer complaint work orders with the same complaint content are combined.
Specifically, the preset time interval may be one year, two years, three years, or the like. The customer complaint work orders can comprise complaint contents, complaint lines, load data and the like, and can be actively submitted by customers or obtained by actively conducting satisfaction investigation to the customers through a subject to be predicted. Because not all complaint content is related to the self state of the equipment of the distribution network, the obtained irrelevant data in the customer complaint work order can be cleared. The complaint content related to the self-state of the equipment of the power distribution network mainly comprises the following aspects: 1. reliable power supply complaints, such as: in summer, the power consumption peak period is often the power consumption peak period, and under overload state, the phenomenon of frequent power failure easily appears, influences the normal production of customer, life power consumption, and power enterprises is carrying out the prearrangement power failure, because of factors such as insufficient reasonable scientific arrangement work or meet thunderstorm weather, leads to the time delay to reply the electricity, and does not inform in time to the customer, probably arouses customer complaints. 2. Power quality complaints, such as: the problems of voltage fluctuation and the like caused by load fluctuation and untimely voltage regulation are easy to cause customer complaints. 3. Power supply safety complaints, such as: the occurrence of unexpected events causes overload of load, potential safety hazards such as equipment residue, line drop and the like, or customer complaints caused by equipment noise, heat dissipation and various construction problems.
More specifically, in order to extract a complaint work order related to the state of the power distribution network device itself from a large number of complaint work orders, screening may be performed according to predetermined complaint content keywords, for example: the predetermined complaint content keywords are quality, safety, reliability, overload, equipment damage, power failure, unstable voltage and the like. After screening based on preset complaint content keywords, customer complaint worksheets of irrelevant data can be eliminated, and relevant customer complaint worksheets can be extracted. The complaint worksheets of customers with the same complaint content can be combined because of the worksheets with the same complaint content, the combined data volume of each relevant customer complaint worksheet is recorded, the complaint quantity of each line can be judged based on the combined quantity, the priority can be configured for the line according to the complaint quantity, and the priority treatment with high priority can be carried out.
Optionally, referring to fig. 2, in step S102, the step of inputting the historical load characteristic data into a target recurrent neural network model for training to obtain first load prediction data includes: s1021, constructing a time sequence of the historical load characteristic data, wherein the time sequence comprises a plurality of time nodes, and each time node has corresponding historical load characteristic data.
S1022, taking the historical load characteristic data corresponding to the current time node as characteristic input data of the target cyclic neural network model, and initializing a weight coefficient of the target cyclic neural network model.
S1023, determining a reset gate and an update gate of the target cyclic neural network model at the current time node through a hyperbolic tangent function according to the characteristic input data of the current time node, the hidden state data of the previous time node and the weight coefficient.
S1024, outputting the first load prediction data according to the reset gate and the update gate of the current time node.
Specifically, the time sequence may be t= (t-n,, t-2, t-1, t), where n is a positive integer less than t. The time sequence comprises a plurality of time nodes, and corresponding historical load characteristic data exists corresponding to each time node. If t is the current time node, when there is a current input load characteristic data x t And the hidden state h transferred from the last node t-1 Hidden state h t-1 Including information about previous nodes. Binding x t And h t-1 Calculating, wherein the historical load characteristic data corresponding to the current time node t is used as characteristic input data of a target GRU model, and the output y of the current hidden node is obtained t And hidden state h passed to the next node t Y of output t I.e. the predicted first load prediction data. The input-output structure of the target GRU model is shown in fig. 2. More specifically, first, the hidden state h transmitted from the previous node t-1 And input load characteristic data x of the current node t To obtain two gating states in the target GRU model as shown in the following formulas (1) and (2):
r t ＝tanh(W r ·[h t-1 ，x t ]) (1)
z t ＝tanh(W z ·[h t-1 ，x t ]) (2)
wherein r is t Representing a reset gate, z t Representing an update gate, W r 、W z Respectively represent weight matrices (weight coefficients), and tanh is a hyperbolic tangent function.
By combining x t And h t-1 The splicing is subjected to linear transformation, and then the value of the reset gate is acted on the hidden state h through activation of hyperbolic tangent function t-1 The data of the reset gate and the update gate of the target GRU model can be converted into [ -1,1]And in the range, normalization processing is realized. Will r t The value of (2) is used in equation (3) for the hidden state as follows:
h′ t ＝tanh(W·[r t *h t-1 ，x t ]) (3)
where w represents the weight in the hidden state.
Combining the formulas (3) and h t-1 、z t The hidden state h of the current node can be obtained t The following formula (4) shows:
h t ＝(1-z t )*h t-1 +z t *h′ t (4)
specifically, the reset gate r t Is useful for capturing short-term dependencies in a time series, and resetting gate r t The smaller the value of (2), the more h t-1 The smaller the product of h t-1 The less information is added to the candidate state. Updating door z t The degree to which state information for controlling the previous node is brought into the current state can help the target GRU model decide how much load characteristic data transferred by the previous node to transfer into the current node, when (1-z t ) The larger the load signature data is, the more load signature data is retained. The target GRU model does not clear previous information over time, and it retains relevant information and passes it on to the next node. Finally calculating the hidden state h of the current node t And the first load prediction data is finally output by the target GRU model.
In some embodiments, in the step S102, before the step of inputting the historical load characteristic data into the target autoregressive moving average model for training, the method further includes:
s1025, calculating an autocorrelation coefficient and a partial autocorrelation coefficient according to the time sequence of the historical load characteristic data, and determining the type of the target model based on the autocorrelation coefficient and tailing or truncating of the partial autocorrelation coefficient.
S1026, performing pre-estimation test on parameters in the target model, and optimizing time sequence stationarity of the historical load characteristic data, wherein the pre-estimation test comprises deterministic factor parameter analysis of the target model and random time sequence parameter analysis of the model.
S1027, fitting the target model according to the time sequence of the historical load characteristic data after stability optimization, and constructing to obtain the target autoregressive moving average model.
Specifically, the above-mentioned autocorrelation coefficients (ACF, autocorrelation Function) may represent the effect of measuring the historical load characteristic data on the current generation, and the partial autocorrelation coefficients (PACF, partial Autocorrelation Function) may refer to the indirect effect of removing intermediate variables in calculating the correlation on the basis of the autocorrelation coefficients. The time sequence of the historical load characteristic data is taken as input, an autocorrelation calculation result can be obtained based on an autocorrelation coefficient calculation mode and a partial autocorrelation coefficient calculation mode, an ACF image and a PACF image can be determined according to the autocorrelation calculation result, then a proper AR (p) model, MA (q) model and ARMA (p, q) model are established according to trailing or truncating of the ACF image and the PACF image, and a target model corresponding to the time sequence of the historical load characteristic data can be determined according to the AR (p) model, the MA (q) model and the ARMA (p, q) model. And further estimating and checking unknown parameters in the target model, wherein the checking is completed after the checking is passed, and the construction of the target ARMA model is shown. In most cases, the obtained historical load characteristic data is non-stationary data, so that analysis of the non-stationary historical load characteristic data is required. Wherein the analysis of the time series of non-stationary historical load signature data can be divided into two categories: deterministic factor parameter analysis: it may be referred to as attributing all time series changes in the historical load signature data to the combined effects of 3 factors of long term trends, cyclic variations and random fluctuations. Where long term trends are generally reasonably presumed based on historical load data experience, cyclic fluctuations may include the effects of date node factors and weather node factors, and random fluctuations may include the effects of some random events and extreme events. The randomness of the cyclic variation and the random fluctuation is large, and the cyclic variation and the random fluctuation are difficult to determine and analyze, so that the fitting effect of the target model is not optimal.
Second, random timing parameter analysis: to remedy the deficiencies of deterministic parameter analysis, random sequence models can be used for analysis. In this embodiment, the specific manner is to analyze the non-stationary time sequence by establishing an ARIMA model, convert the non-stationary time sequence into a stationary time sequence, and then perform modeling of the time sequence. Wherein ARIMA is a combination of differential operation and ARMA model. After subtracting the values of two time sequences with a certain distance, the non-stationary sequence shows the characteristic of stationary data, the sequence is called as a differential stationary sequence, and then a target model is used for fitting, and finally a target ARMA model is constructed. The differential stable sequence is a time sequence of historical load characteristic data after stability optimization. The specific flow diagram is shown in fig. 3.
Optionally, as shown in fig. 4, the step S103 includes:
s1031, respectively inputting the first load prediction data and the second load prediction data into an input layer of the BP neural network algorithm, and constructing a BP neural network algorithm model by combining parameters from the input layer to the hidden layer and parameters from the hidden layer to an output layer, wherein the parameters comprise weights and bias items.
S1032, initializing the weight and bias items of the BP neural network algorithm model, and iterating the expected values of the output and loss functions of each layer for a plurality of times through back propagation.
S1033, determining parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer in the BP neural network algorithm model according to the expected value of the loss function.
S1034, updating weights and bias items among layers in the BP neural network algorithm model according to parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer until a preset iteration condition is met, and stopping updating to obtain the first target weight, the second target weight, the first bias item and the second bias item.
Specifically, the BP neural network algorithm comprises an input layer, a hidden layer and an output layer, wherein weight and bias items are respectively arranged between the input layer and the hidden layer and between the hidden layer and the output layer. If the weights and bias terms of the input layer and the hidden layer are set to be w and b respectively 1 The weights and bias terms from the hidden layer to the output layer are v and b respectively 2 The activation function is g 1 And g is equal to 2 . At this time, the first load prediction data output by the target GRU model and the second load prediction data (ARMA sample) output by the target ARMA model are taken as input data of the input layer, and the calculation formulas from the input layer to the hidden layer and from the hidden layer to the output layer in the BP neural network are shown in formulas (5) and (6):
net 1 ＝w T x+b 1 ，h＝g 1 (net 1 ) (5)
Wherein x is a vector formed by the first load prediction data and the second load prediction data, h represents a value after the activation function is activated, and y represents an output value of the BP neural network algorithm model.
And (3) converting by combining the formulas (5) and (6), and constructing a BP neural network algorithm model as shown in the formula (7):
then respectively initializing weights and bias items in the BP neural network algorithm model, and respectively marking as:
/>
by activating forward propagation, the output value of each layer and the expected value of the loss function in the BP neural network algorithm model can be obtained, and the error can be calculated according to the output value and the expected value of each layer. Wherein the loss function is as shown in formula (8):
after back propagation and multiple iterations are performed according to the loss function, an error term of an output layer and an error term of a hidden layer of the BP neural network algorithm model can be calculated, wherein: the error term of the output layer may specifically be a gradient value or partial derivative of the calculated loss function with respect to the output layer, as in equation (9), according to the chain law:
the error term of the hidden layer may specifically be a gradient value or a partial derivative of the calculated loss function with respect to the hidden layer, as in equation (10), according to the chain law:
then, updating weights and bias terms of the output layer and the hidden layer in the BP neural network algorithm model according to the error terms of the output layer and the error terms of the hidden layer, wherein the weights and bias terms are shown in the following formulas (11) and (12):
Where η represents a learning rate, k=1, 2,3,..n, k represents the number of updates or iterations, k=1 represents the first update, and so on.
Specifically, the preset iteration conditions may include an update frequency threshold, an iteration frequency threshold, an error threshold, and the like, and after each iteration, the update frequency, the iteration frequency, and the error calculated by the loss function may be compared with the preset iteration conditions, so that the update may be stopped when the preset iteration conditions are satisfied. At this time, a first target weight and the second target weight, and a first bias term and a second bias term of the BP neural network algorithm model are obtained, and the two weights obtained by updating finally are respectively used as a first target weight of the first load prediction data and a second target weight of the second load data.
Optionally, the step S104 includes:
s1041, if the expected value of the loss function meets the preset expected value, determining the target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, wherein the sum of the first target weight and the second target weight is equal to 1.
S1042, determining the line risk index according to the target load prediction data based on the corresponding relation between the preset load data and the risk index.
Specifically, the target load prediction data Y is calculated t The calculation formula of (2) is shown in the following formula (13):
/>
wherein f it { i=1, 2} includes first load prediction data and second load prediction data, w i { i=1, 2} includes the first target weight of the first load prediction data and the second target weight of the second load prediction data.
Wherein, the liquid crystal display device comprises a liquid crystal display device,
calculating to obtain the target load predicted numberAccording to Y t After that, the route risk index corresponding to the route can be determined according to the preset corresponding relation between the load data and the risk index, so as to realize the route risk assessment, and provide powerful data support for making countermeasures in advance, so that operation and maintenance personnel can make a route control strategy in advance according to the route risk index, for example: load supply, line service, etc. are increased.
Example two
As shown in fig. 5, fig. 5 is a line risk assessment device based on a dual training weight distribution model, which is proposed by a line risk assessment method based on a dual training weight distribution model, and the device 50 includes:
the acquisition module 501 is configured to acquire historical load data of an object under different working conditions, and perform preprocessing on the historical load data to obtain historical load characteristic data;
The training module 502 is configured to input the historical load characteristic data into a target recurrent neural network model for training to obtain first load prediction data, and input the historical load characteristic data into a target autoregressive moving average model for training to obtain second load prediction data;
an optimization module 503, configured to perform weight optimization on the first load prediction data and the second load prediction data through a BP neural network algorithm, so as to obtain a first target weight and a second target weight;
a calculating module 504, configured to calculate target prediction data according to the first load prediction data, the first target weight, the second load prediction data, and the second target weight, and determine a line risk index according to the target prediction data.
Optionally, the obtaining module 501 is specifically configured to:
acquiring the historical load data of the object under different working conditions, wherein the historical load data comprises acquisition of line periodic load data and line special working condition load data, and the line special working condition load data comprises date node load data, weather node load data and customer complaint work order load data;
according to a preset preprocessing rule, the date node load data, the weather node load data and the customer complaint work order load data are standardized;
Constructing a sample data set based on the line cycle load data, the standardized date node load data, the weather node load data and the customer complaint work order load data;
and carrying out feature extraction on the data in the sample data set through a convolutional neural network to obtain the historical load feature data.
Optionally, the apparatus 50 further includes:
the work order acquisition module is used for acquiring the customer complaint work order load data in a preset time interval;
and the screening module is used for clearing irrelevant data of the customer complaint work order load data according to preset complaint content keywords and merging customer complaint work orders with the same complaint content.
Optionally, the training module 502 is specifically configured to:
constructing a time sequence of the historical load characteristic data, wherein the time sequence comprises a plurality of time nodes, and each time node has corresponding historical load characteristic data;
taking the historical load characteristic data corresponding to the current time node as characteristic input data of the target cyclic neural network model, and initializing a weight coefficient of the target cyclic neural network model;
determining a reset gate and an update gate of the target cyclic neural network model at the current time node through a hyperbolic tangent function according to the characteristic input data of the current time node, the hidden state data of the previous time node and the weight coefficient;
And outputting the first load prediction data according to the reset gate and the update gate of the current time node.
Optionally, the apparatus 50 further includes:
the model determining module is used for calculating an autocorrelation coefficient and a partial autocorrelation coefficient according to the time sequence of the historical load characteristic data and determining the type of a target model based on the autocorrelation coefficient and tailing or truncating of the partial autocorrelation coefficient;
the stability analysis module is used for carrying out pre-estimation test on parameters in the target model and optimizing the time sequence stability of the historical load characteristic data, wherein the analysis module comprises deterministic factor parameter analysis on the target model and random time sequence parameter analysis on the model;
and the fitting module is used for fitting according to the time sequence of the historical load characteristic data after the stability optimization through the target model, and constructing and obtaining the target autoregressive moving average model.
Optionally, the optimizing module 503 is specifically configured to:
inputting the first load prediction data and the second load prediction data into an input layer of the BP neural network algorithm respectively, and constructing a BP neural network algorithm model by combining parameters from the input layer to the hidden layer and parameters from the hidden layer to an output layer, wherein the parameters comprise weights and bias items;
Initializing weight and bias items of the BP neural network algorithm model, and iterating expected values of output and loss functions of each layer for a plurality of times through back propagation;
determining parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer in the BP neural network algorithm model according to expected values of the loss function;
and updating weights and bias items among layers in the BP neural network algorithm model according to parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer until a preset iteration condition is met, and stopping updating to obtain the first target weight, the second target weight, the first bias item and the second bias item.
Optionally, the calculating module 504 is specifically configured to:
if the expected value of the loss function meets the preset expected value, determining target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, wherein the sum of the first target weight and the second target weight is equal to 1;
and determining the line risk index according to the target load prediction data based on a preset corresponding relation between the load data and the risk index.
The line risk assessment device based on the double training weight distribution model provided by the embodiment of the invention can realize each process realized by the line risk assessment method based on the double training weight distribution model in the method embodiment, can achieve the same beneficial effects, and is not repeated here.
Referring to fig. 6, fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, as shown in fig. 6, including: memory 602, processor 601 and a computer program of a data management method stored on memory 602 and executable on processor 601. The electronic device provided by the embodiment of the invention can realize each process realized by the line risk assessment method based on the double training weight distribution model in the method embodiment, can achieve the same beneficial effects, and is not repeated here for avoiding repetition.
It should be noted that, as those skilled in the art will appreciate, the electronic device herein is a device capable of automatically performing numerical calculation and/or information processing according to a predetermined or stored instruction, and its hardware includes, but is not limited to, a microprocessor, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a Programmable gate array (FPGA), a digital processor (Digital Signal Processor, DSP), an embedded device, and the like. The electronic device may be a computing device such as a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The electronic equipment can perform man-machine interaction in a mode of a keyboard, a mouse, a remote controller, a touch pad or voice control equipment and the like.
The embodiment of the invention also provides a computer readable storage medium, and a computer program is stored on the computer readable storage medium, and when the computer program is executed by a processor, the process of the line risk assessment method based on the double training weight distribution model provided by the embodiment of the invention is realized, and the same technical effect can be achieved, so that repetition is avoided, and the description is omitted here.
The readable storage medium includes flash memory, a hard disk, a multimedia card, a card memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, and the like. In some embodiments, the memory may be an internal storage unit of the electronic device, such as a hard disk or a memory of the electronic device. In other embodiments, the memory may also be an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Card (Flash Card) or the like. Of course, the memory may also include both internal storage units of the electronic device and external storage devices. In this embodiment, the memory is generally used to store an operating system and various application software installed in the electronic device, for example, a program code of a line risk assessment method based on a dual training weight distribution model. In addition, the memory can be used to temporarily store various types of data that have been output or are to be output.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program to instruct the associated hardware and that the program may be stored on a computer readable storage medium, which when executed may include the steps of the embodiments of the methods described above.
The above embodiments are preferred embodiments of the line risk assessment method based on the dual training weight distribution model, and are not intended to limit the scope of the present invention, which includes but is not limited to the embodiments, and equivalent changes of shape and structure according to the present invention are all within the scope of the present invention.

Claims (10)

1. A line risk assessment method based on a double training weight distribution model, the method comprising the steps of: acquiring historical load data of an object under different working conditions, and preprocessing the historical load data to obtain historical load characteristic data;
inputting the historical load characteristic data into a target cyclic neural network model for training to obtain first load prediction data, and inputting the historical load characteristic data into a target autoregressive moving average model for training to obtain second load prediction data; performing weight optimization on the first load prediction data and the second load prediction data through a BP neural network algorithm to obtain a first target weight and a second target weight;
Calculating target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, and determining a line risk index according to the target load prediction data.
2. The line risk assessment method based on the double training weight distribution model according to claim 1, wherein the obtaining the historical load data of the object under different working conditions, preprocessing the historical load data to obtain the historical load characteristic data, comprises:
acquiring the historical load data of the object under different working conditions, wherein the historical load data comprises acquisition of line periodic load data and line special working condition load data, and the line special working condition load data comprises date node load data, weather node load data and customer complaint work order load data;
according to a preset preprocessing rule, the date node load data, the weather node load data and the customer complaint work order load data are standardized;
constructing a sample data set based on the line cycle load data, the standardized date node load data, the weather node load data and the customer complaint work order load data;
And carrying out feature extraction on the data in the sample data set through a convolutional neural network to obtain the historical load feature data.
3. The line risk assessment method based on a dual training weight distribution model according to claim 2, wherein before the normalizing the date node load data, the weather node load data, and the customer complaint work order load data according to a preset preprocessing rule, the method further comprises:
acquiring the customer complaint work order load data in a preset time interval;
and according to the preset complaint content keywords, performing irrelevant data clearing on the customer complaint work order load data, and merging customer complaint work orders with the same complaint content.
4. The line risk assessment method based on the double training weight distribution model as claimed in claim 1, wherein the step of inputting the historical load characteristic data into a target cyclic neural network model for training to obtain first load prediction data comprises the following steps:
constructing a time sequence of the historical load characteristic data, wherein the time sequence comprises a plurality of time nodes, and each time node has corresponding historical load characteristic data;
Taking the historical load characteristic data corresponding to the current time node as characteristic input data of the target cyclic neural network model, and initializing a weight coefficient of the target cyclic neural network model;
determining a reset gate and an update gate of the target cyclic neural network model at the current time node through a hyperbolic tangent function according to the characteristic input data of the current time node, the hidden state data of the previous time node and the weight coefficient;
and outputting the first load prediction data according to the reset gate and the update gate of the current time node.
5. The line risk assessment method based on a dual training weight distribution model according to claim 4, wherein before said inputting said historical load characteristic data into a target autoregressive moving average model for training, said method further comprises:
calculating an autocorrelation coefficient and a partial autocorrelation coefficient according to the time sequence of the historical load characteristic data, and determining the type of a target model based on the autocorrelation coefficient and tailing or truncating of the partial autocorrelation coefficient;
performing pre-estimation test on parameters in the target model, and optimizing time sequence stationarity of the historical load characteristic data, wherein the pre-estimation test comprises deterministic factor parameter analysis of the target model and random time sequence parameter analysis of the model;
Fitting the target model according to the time sequence of the historical load characteristic data after stability optimization, and constructing to obtain the target autoregressive moving average model.
6. The line risk assessment method based on a dual training weight distribution model according to any one of claims 1 to 5, wherein the BP neural network algorithm includes an input layer, a hidden layer, and an output layer, and the weighting optimization of the first load prediction data and the second load prediction data by the BP neural network algorithm includes:
inputting the first load prediction data and the second load prediction data into an input layer of the BP neural network algorithm respectively, and constructing a BP neural network algorithm model by combining parameters from the input layer to the hidden layer and parameters from the hidden layer to an output layer, wherein the parameters comprise weights and bias items;
initializing weight and bias items of the BP neural network algorithm model, and iterating expected values of output and loss functions of each layer for a plurality of times through back propagation;
determining parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer in the BP neural network algorithm model according to expected values of the loss function;
And updating weights and bias items among layers in the BP neural network algorithm model according to parameter error items from an input layer to a hidden layer and parameter error items from the hidden layer to an output layer until a preset iteration condition is met, and stopping updating to obtain the first target weight, the second target weight, the first bias item and the second bias item.
7. The line risk assessment method based on a dual training weight distribution model according to claim 6, wherein said calculating target load prediction data from said first load prediction data, said first target weight, said second load prediction data, and said second target weight, and determining a line risk index from said target load prediction data, comprises: if the expected value of the loss function meets the preset expected value, determining target load prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight, wherein the sum of the first target weight and the second target weight is equal to 1;
and determining the line risk index according to the target load prediction data based on a preset corresponding relation between the load data and the risk index.
8. A line risk assessment device based on a dual training weight distribution model, the device comprising:
the acquisition module is used for acquiring historical load data of the object under different working conditions, and preprocessing the historical load data to obtain historical load characteristic data;
the training module is used for inputting the historical load characteristic data into a target cyclic neural network model for training to obtain first load prediction data, and inputting the historical load characteristic data into a target autoregressive moving average model for training to obtain second load prediction data;
the optimization module is used for carrying out weight optimization on the first load prediction data and the second load prediction data through a BP neural network algorithm to obtain a first target weight and a second target weight;
and the calculating module is used for calculating target prediction data according to the first load prediction data, the first target weight, the second load prediction data and the second target weight and determining a line risk index according to the target prediction data.
9. An electronic device, comprising: memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in a line risk assessment method based on a dual training weight distribution model according to any one of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of a line risk assessment method based on a dual training weight distribution model according to any of claims 1 to 7.
CN202310434199.8A 2023-04-21 2023-04-21 Line risk assessment method and device based on double training weight distribution model Pending CN116757465A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310434199.8A CN116757465A (en) 2023-04-21 2023-04-21 Line risk assessment method and device based on double training weight distribution model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310434199.8A CN116757465A (en) 2023-04-21 2023-04-21 Line risk assessment method and device based on double training weight distribution model

Publications (1)

Publication Number Publication Date
CN116757465A true CN116757465A (en) 2023-09-15

Family

ID=87952104

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310434199.8A Pending CN116757465A (en) 2023-04-21 2023-04-21 Line risk assessment method and device based on double training weight distribution model

Country Status (1)

Country Link
CN (1) CN116757465A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117726959A (en) * 2024-02-09 2024-03-19 国网安徽省电力有限公司巢湖市供电公司 Unmanned aerial vehicle power line safety inspection system and method based on intelligent image recognition
CN117933499A (en) * 2024-03-22 2024-04-26 中国铁建电气化局集团有限公司 Invasion risk prediction method, device and storage medium for high-speed railway catenary

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117726959A (en) * 2024-02-09 2024-03-19 国网安徽省电力有限公司巢湖市供电公司 Unmanned aerial vehicle power line safety inspection system and method based on intelligent image recognition
CN117726959B (en) * 2024-02-09 2024-05-10 国网安徽省电力有限公司巢湖市供电公司 Unmanned aerial vehicle power line safety inspection system and method based on intelligent image recognition
CN117933499A (en) * 2024-03-22 2024-04-26 中国铁建电气化局集团有限公司 Invasion risk prediction method, device and storage medium for high-speed railway catenary

Similar Documents

Publication Publication Date Title
CN113962364B (en) Multi-factor power load prediction method based on deep learning
Kazemzadeh et al. A hybrid data mining driven algorithm for long term electric peak load and energy demand forecasting
CN110610280B (en) Short-term prediction method, model, device and system for power load
Zhang et al. Short term electricity load forecasting using a hybrid model
Ahmad et al. A data-driven deep sequence-to-sequence long-short memory method along with a gated recurrent neural network for wind power forecasting
Kani et al. Very short-term wind speed prediction: A new artificial neural network–Markov chain model
Wang et al. A compound framework for wind speed forecasting based on comprehensive feature selection, quantile regression incorporated into convolutional simplified long short-term memory network and residual error correction
CN116757465A (en) Line risk assessment method and device based on double training weight distribution model
CN111027772A (en) Multi-factor short-term load prediction method based on PCA-DBILSTM
Chen et al. Short-term load forecasting and associated weather variables prediction using ResNet-LSTM based deep learning
Mokarram et al. Net-load forecasting of renewable energy systems using multi-input LSTM fuzzy and discrete wavelet transform
Zhang et al. Load Prediction Based on Hybrid Model of VMD‐mRMR‐BPNN‐LSSVM
CN114548586A (en) Short-term power load prediction method and system based on hybrid model
CN110837934B (en) Smart grid short-term residential load prediction method based on deep learning
Kosana et al. Hybrid wind speed prediction framework using data pre-processing strategy based autoencoder network
Rizvi Time series deep learning for robust steady-state load parameter estimation using 1D-CNN
CN116663746A (en) Power load prediction method and device, computer equipment and storage medium
Mishra et al. Performance evaluation of prophet and STL-ETS methods for load forecasting
Gao et al. A two-layer SSA-XGBoost-MLR continuous multi-day peak load forecasting method based on hybrid aggregated two-phase decomposition
Ibrahim et al. LSTM neural network model for ultra-short-term distribution zone substation peak demand prediction
Li et al. Short-term probabilistic load forecasting method based on uncertainty estimation and deep learning model considering meteorological factors
CN116865254A (en) Power load index prediction method, system, equipment and medium
Zhang Short‐Term Power Load Forecasting Based on SAPSO‐CNN‐LSTM Model considering Autocorrelated Errors
Wang et al. Short-term probability density function forecasting of industrial loads based on ConvLSTM-MDN
Shen et al. An interval analysis scheme based on empirical error and MCMC to quantify uncertainty of wind speed

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