CN108830405B - Real-time power load prediction system and method based on multi-index dynamic matching - Google Patents

Real-time power load prediction system and method based on multi-index dynamic matching Download PDF

Info

Publication number
CN108830405B
CN108830405B CN201810525917.1A CN201810525917A CN108830405B CN 108830405 B CN108830405 B CN 108830405B CN 201810525917 A CN201810525917 A CN 201810525917A CN 108830405 B CN108830405 B CN 108830405B
Authority
CN
China
Prior art keywords
load
index
data
representing
curve
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810525917.1A
Other languages
Chinese (zh)
Other versions
CN108830405A (en
Inventor
周铁华
王玲
马福涛
高雪
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.)
Northeast Electric Power University
Original Assignee
Northeast Dianli University
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 Northeast Dianli University filed Critical Northeast Dianli University
Priority to CN201810525917.1A priority Critical patent/CN108830405B/en
Publication of CN108830405A publication Critical patent/CN108830405A/en
Application granted granted Critical
Publication of CN108830405B publication Critical patent/CN108830405B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Engineering & Computer Science (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Educational Administration (AREA)
  • Game Theory and Decision Science (AREA)
  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention relates to a real-time power load prediction system and a method thereof based on multi-index dynamic matching, wherein the system comprises: the data acquisition module is in signal connection with the data analysis module, the data analysis module is in signal connection with the multi-index dynamic matching module, and the multi-index dynamic matching module is in signal connection with the real-time power load prediction module. The method can acquire multiple data indexes such as real-time power load data, meteorological data, economic data and the like; analyzing the correlation between the power load data and each factor index, and mining a dynamic key factor set influencing power load prediction in a self-adaptive manner; and establishing a comprehensive adaptive index system influencing the power load prediction according to the dependency relationship between the power load and the key indexes, and extracting a key influence factor set in real time to predict the short-term and ultra-short-term power load through the provided A-GPR prediction model.

Description

Real-time power load prediction system and method based on multi-index dynamic matching
Technical Field
The invention belongs to the technical field of intelligent power grid power load prediction, and particularly relates to a real-time power load prediction system and method based on multi-index dynamic matching.
Background
The power load prediction is to predict a future load using historical data, taking into account the operating characteristics, capacity increase, and natural conditions of the power system. Load prediction is an important function of the power grid energy management system and is the basis for economic, safe and reliable operation of the intelligent power grid system. The load of the power system has an inherent periodic law and is influenced by a plurality of factors, such as climate conditions, economic factors and the like. Due to the difference of the load characteristics of each region, the load prediction work of the regions is combined with the local actual situation according to the difference of the load characteristics among different regions, the influence factors of the load are considered on the basis of the load characteristics, and then a proper method is selected for prediction, so that the prediction accuracy is improved. The accuracy of load prediction directly influences the safety, economy and power supply quality of a power system, has an important influence on the scientificity of investment, scheduling, layout and operation of a power grid, and is an important basis for development planning and real-time control. Therefore, how to improve the prediction accuracy is the focus of the current load prediction technology research.
The current common load prediction methods comprise two main categories of traditional prediction methods and modern prediction methods, wherein the traditional prediction methods comprise a time series method, a regression analysis method, a gray prediction method and the like. Among them, the time-series method is most widely used. The time sequence method is to deduce the load data of a future period of time according to the historical load data for one-dimensional time sequence data. The gray prediction method only needs a small amount of historical data, and finds the rule influencing the load from the historical data, so that the calculation amount is small. The load prediction model established on the basis is difficult to deal with the power load with large fluctuation. And when the historical data is small and the discrete degree is large, the prediction precision is poor. The modern prediction method mainly comprises an expert system method, a support vector machine, a neural network algorithm and the like. The neural network becomes an important method for load prediction due to the self-learning capability and the capability of processing complex nonlinearity. However, the structure and network parameters of the neural network mostly need to be determined by subjective experience. Therefore, it is difficult to ensure the accuracy of the prediction result, and some methods do not consider the influence of local meteorological factors and economic factors on load prediction during prediction, resulting in the loss of important information. Even if the influence relation between a single factor and the load data is considered, the influence degree of all factors on the load prediction cannot be reflected, and the analysis result is easy to generate errors, so that the accuracy of the load prediction is influenced. If various influencing factors are included in the input variables, the input variables are too many, the training load is increased, the accuracy cannot be improved, and the performance of model prediction is reduced. Therefore, it is a problem that the load prediction must be solved by appropriately compressing the input variables while taking into consideration various factors that affect the load prediction.
Aiming at the defects of the existing load prediction method, the invention aims to provide a real-time power load prediction system based on multi-index dynamic matching, wherein the model of the system comprehensively considers the influence of a plurality of indexes on a load, and the weight of each factor changes along with the change of time along with the difference of the influence factors of the load in the process of processing load data. The invention fully considers the influence of the attribute on the load directly and indirectly, so that the final weight distribution is more scientific and reasonable, thereby selecting key indexes to establish an index system for load prediction, effectively reducing the workload of load prediction and improving the accuracy and reliability of load prediction. Meanwhile, an Adjustable Gaussian Process Regression model (Adjustable-Gaussian Process Regression) is established, namely an A-GPR real-time power load prediction model, so that timeliness of power supply and effective utilization of new energy under the intelligent power grid are realized.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provide a real-time power load prediction system based on multi-index dynamic matching, which has reasonable structure and high accuracy and reliability of functional load prediction, and provide science and reasonability.
The technical scheme adopted for realizing one of the purposes of the invention is as follows: a real-time power load prediction system based on multi-index dynamic matching is characterized by comprising the following components: the data acquisition module is used for acquiring load, meteorological and economic data, performing classified management and constructing a system database; calculating the correlation between the influence factors and the load based on the MIDM algorithm, sorting, screening out key attributes, and constructing a data analysis module of a comprehensive index system; the multi-index dynamic matching module of the A-GPR model can be constructed by converting fixed inertia factors in the model into dynamically changed inertia factors according to random changes of different regions and time; the real-time power load prediction module can predict the real-time power load according to the constructed comprehensive index system and the A-GPR model; the data acquisition module is in signal connection with the data analysis module, the data analysis module is in signal connection with the multi-index dynamic matching module, and the multi-index dynamic matching module is in signal connection with the real-time power load prediction module.
The second technical scheme for realizing the purpose of the invention is as follows: a real-time power load prediction method based on multi-index dynamic matching is characterized by comprising the following contents:
1) the data acquisition module is used for acquiring data by utilizing respective functions of a meteorological data unit, a load data unit and an economic data unit of the data acquisition module, and the meteorological data unit is used for acquiring temperature, humidity, precipitation, visibility, wind direction, wind speed and weather condition data of a required area in real time to integrate the data and transmitting the data to a system database to generate a meteorological data table; the load data unit is responsible for collecting industrial power loads, agricultural power loads, municipal power loads, post and post electricity power loads, traffic power loads, life power loads and commercial power loads in real time for data integration and transmitting the data integration to a system database to generate a load data table; the economic data unit is responsible for collecting economic data in the annual economic report of the government in real time for data integration and transmitting the economic data to the system database to generate an economic data table;
2) the method comprises the following steps of utilizing the function of a data analysis module to adopt a Multi-Index Dynamic Matching algorithm (Multi-Index Dynamic Matching), namely, carrying out correlation analysis on each Index by using an MIDM algorithm, and analyzing the correlation degree between load data and corresponding meteorological data and economic data so as to determine key indexes influencing a required area, wherein the Multi-Index Dynamic Matching algorithm comprises the following steps:
(1) calculating the projection distance of each index curve to the load:
Figure GDA0003104932750000031
wherein xqiRepresenting the projected abscissa, k, of a point on the influence index curve on the load curveqiThe slope of this point of the load curve, b1(2) Representing the abscissa of the point on the influence index curve,b1(1) ordinate representing a point on the curve of the influence indicator, bqiRepresenting the intercept of the load curve at this point, yqiRepresents the projection ordinate of the point on the influence index on the load curve, q represents 24 points on the curve 24 hours a day of the influence index, i represents a certain influence index, rqiRepresenting the projected distance of the impact indicator to the load, xqi2Abscissa, x, representing a point subsequent to the point of projection of the influence index curve on the load curveqi1Abscissa, y, representing a point preceding the point of projection of the influence index curve on the load curveqi2Longitudinal coordinate, y, of the point of influence index curve after the projection point of the load curveqi1Representing the ordinate, R, of the influence index curve at a point preceding the projection point of the load curvetRepresenting the fluctuation distance of the load between two points, t representing 24 points on the curve at 24 hours a day, yt2Ordinate, y, representing the latter point on the load curvet1Ordinate, x, representing the previous point on the load curvet2Abscissa, x, representing the latter point on the load curvet1The ordinate represents the previous point on the load curve;
(2) and (3) calculating the weight of each index:
Figure GDA0003104932750000032
wherein wiRepresenting the influence weight of each influence factor on the load, i represents a certain influence index, rqiRepresenting the projected distance of the influencing factor to the load, RtRepresenting the fluctuation distance of the load between two points;
(3) and (3) calculating the weight of the set index:
Figure GDA0003104932750000033
i, j ≠ 1,2
Figure GDA0003104932750000041
Wherein n represents the number of indices, wi,jIndirect weights representing combined indicators, i, j representing an influence indicator, wi' represents the weight of the set index, and fully expresses the direct and indirect influence of each index on the load;
3) establishing a comprehensive dynamic index system according to the real-time correlation of the load data and the key indexes of different regions by using the function of a multi-index dynamic matching module; an A-GPR model is established by utilizing the selected load data and a comprehensive index system, and the method comprises the following steps:
(1) initializing population size N, maximum number of iterations Tmax；
(2)
Figure GDA0003104932750000042
A vector of the current position of the particle;
(3) fitness, the Fitness value of vector x;
(4)
Figure GDA0003104932750000043
velocity of particle, its dimension and vector
Figure GDA0003104932750000044
Are the same as (a);
(5)pbestthe fitness value corresponding to the best position encountered by each particle in the flight process;
(6)
Figure GDA0003104932750000045
for recording the best position encountered during the flight of the particles, dimensions and vectors of pbest
Figure GDA0003104932750000046
Are the same in dimension;
(7) updating formula of particle position:
Figure GDA0003104932750000047
xi＝xi+vi
wherein v isiRepresenting the velocity, x, of the particleiRepresenting the current position of the particle, i representing the current number of iterations per particle update, gbestiRepresenting the optimal position of the population as a whole, pbestiRepresents the optimal position of the individual in the population, w represents an inertial weight factor, phi1Represents the individual cognitive learning rate, phi2Representing social learning rate, and rand () is the interval [0, 1%]Random number in between, K represents the convergence factor, K can be described as:
Figure GDA0003104932750000048
the standard particle swarm algorithm generally sets an inertia weight factor: w is at+1＝wtSince the inertial weight factor is a variable affecting the current particle velocity, a larger value is favorable for global search, a smaller value is favorable for local search, and in order to better balance the search capability, an a-GPR model is proposed:
Figure GDA0003104932750000049
wherein: w is atAn inertial weight factor, w, representing the number of current iterationsmaxIs the maximum value of inertia, wminIs the inertia minimum value, T is the current iteration time, T is the maximum iteration time,
therefore, the update formula of the particle position is:
Figure GDA0003104932750000051
xi＝xi+vi
because the inertia weight factor is changed all the time, the speed and the position of the particles are continuously updated all the time in a large range and a small range of the population, thereby balancing the capability of the algorithm in global search and local search;
(8) recalculating the fitness function E of the particle according to the updated position of the particle, where the fitness function may be described as:
Figure GDA0003104932750000052
where m is the number of input samples, yiRepresenting the process output value, y, of the current A-GPR model trainingi-1I represents the current iteration number for the output value of the previous generation model;
(9) judging whether the progress requirement is met or whether the iteration times are reached, and if the progress requirement is met, terminating the search; otherwise, carrying out next search;
(10) outputting the information of the final particles, namely the parameters of the A-GPR model;
4) the method comprises the steps of acquiring meteorological data and economic data of a moment to be predicted by utilizing the function of a real-time power load prediction module, establishing a comprehensive index system, inputting the comprehensive index system and corresponding load data into an A-GPR model as input vectors, and outputting the input vectors, namely the load value of the moment to be predicted.
The multi-index dynamic matching real-time power load prediction system and the method thereof have the advantages that:
1. the multi-index dynamic matching real-time power load forecasting system is reasonable in structure, has the functions of data acquisition, data analysis, multi-index dynamic matching, real-time power load forecasting and the like, and can realize the real-time extraction of key influence factor sets to carry out short-term and ultra-short-term power load forecasting;
2. because the change of the load is influenced by weather conditions, temperature, humidity, atmospheric pressure, visibility, wind direction, wind speed and GDP, the change has obvious correlation, and the factors influencing the load have diversity, the method of the invention can fully consider multiple indexes influencing the load, accurately master the change condition of the load, calculate the correlation of each index according to the MIDM algorithm, sort according to the correlation between the indexes, screen out the key indexes influencing the load, make the change characteristic of the load easier to grasp, and have high accuracy;
3. according to the A-GPR method in the real-time power load prediction module, the originally fixed model parameters can be subjected to self-adaptive real-time change according to the input of different indexes, the adaptability and the robustness of the model are improved, the method is a brand new development direction of power load prediction of a power system in future, and the development of an intelligent power grid and an energy internet is facilitated;
4. the method is scientific, high in applicability and good in effect.
Drawings
FIG. 1 is a block diagram of a multi-index dynamically matched real-time power load forecasting system;
fig. 2 is a flowchart of the operation of fig. 1.
Detailed Description
Referring to fig. 1 and 2, a real-time power load prediction system based on multi-index dynamic matching includes: the data acquisition module is used for acquiring load, meteorological and economic data, performing classified management and constructing a system database; calculating the correlation between the influence factors and the load based on the MIDM algorithm, sorting, screening out key attributes, and constructing a data analysis module of a comprehensive index system; the multi-index dynamic matching module of the A-GPR model can be constructed by converting fixed inertia factors in the model into dynamically changed inertia factors according to random changes of different regions and time; the real-time power load prediction module can predict the real-time power load according to the constructed comprehensive index system and the A-GPR model; the data acquisition module is in signal connection with the data analysis module, the data analysis module is in signal connection with the multi-index dynamic matching module, and the multi-index dynamic matching module is in signal connection with the real-time power load prediction module.
A real-time power load prediction method based on multi-index dynamic matching comprises the following contents:
1) the data acquisition module is used for acquiring data by utilizing respective functions of a meteorological data unit, a load data unit and an economic data unit of the data acquisition module, and the meteorological data unit is used for acquiring temperature, humidity, precipitation, visibility, wind direction, wind speed and weather condition data of a required area in real time to integrate the data and transmitting the data to a system database to generate a meteorological data table; the load data unit is responsible for collecting industrial power loads, agricultural power loads, municipal power loads, post and post electricity power loads, traffic power loads, life power loads and commercial power loads in real time for data integration and transmitting the data integration to a system database to generate a load data table; the economic data unit is responsible for collecting economic data in the annual economic report of the government in real time for data integration and transmitting the economic data to the system database to generate an economic data table;
2) and performing correlation analysis on each index by using the function of the data analysis module and adopting an MIDM algorithm, and analyzing the correlation degree between the load data and the corresponding meteorological data and economic data so as to determine key indexes influencing the required area, wherein the multi-index dynamic matching algorithm comprises the following steps of:
(1) calculating the projection distance of each index curve to the load:
Figure GDA0003104932750000071
wherein xqiRepresenting the projected abscissa, k, of a point on the influence index curve on the load curveqiRepresenting the slope of the load curve at this point, b1(2) Abscissa representing point on the influence index curve, b1(1) Ordinate representing a point on the curve of the influence indicator, bqiRepresenting the intercept of the load curve at this point. y isqiRepresents the projected ordinate of the point on the influence index on the load curve, q represents 24 points on the curve of the influence index 24 hours a day, and i represents a certain influence index. r isqiRepresenting the projected distance of the impact indicator to the load, xqi2Abscissa, x, representing a point subsequent to the point of projection of the influence index curve on the load curveqi1Representing influence index curveThe line projects on the load curve the abscissa, y, of a point preceding the pointqi2Longitudinal coordinate, y, of the point of influence index curve after the projection point of the load curveqi1Representing the ordinate of the impact indicator curve at a point preceding the projected point of the load curve. RtRepresenting the fluctuation distance of the load between two points, t representing 24 points on the curve at 24 hours a day, yt2Ordinate, y, representing the latter point on the load curvet1Ordinate, x, representing the previous point on the load curvet2Abscissa, x, representing the latter point on the load curvet1Representing the ordinate of the previous point on the load curve.
(2) And (3) calculating the weight of the index:
Figure GDA0003104932750000072
wherein wiRepresenting the weight of the impact indicator on the load, i representing a certain impact indicator, rqiRepresenting the projected distance of the influencing factor to the load, RtRepresenting the fluctuating distance of the load between two points.
(3) And (3) calculating the weight of the set index:
Figure GDA0003104932750000073
i, j ≠ 1,2
Figure GDA0003104932750000081
Wherein n represents the number of indices, WijIndirect weights representing combined indices, i and j representing certain influence indices, wi' represents the weight of the set index, and fully expresses the direct and indirect influence of each index on the load;
3) establishing a comprehensive dynamic index system according to the real-time correlation of the load data and the key indexes of different regions by using the function of a multi-index dynamic matching module; an A-GPR model is established by utilizing the selected load data and a comprehensive index system, and the method comprises the following steps:
(1) initializing population size N, maximum number of iterations Tmax；
(2)
Figure GDA0003104932750000082
A vector of the current position of the particle;
(3) fitness, the Fitness value of vector x;
(4)
Figure GDA0003104932750000083
velocity of particle, its dimension and vector
Figure GDA0003104932750000084
Are the same as (a);
(5)pbestthe fitness value corresponding to the best position encountered by each particle in the flight process;
(6)
Figure GDA0003104932750000085
for recording the best position encountered during the flight of the particles, dimensions and vectors of pbest
Figure GDA0003104932750000086
Are the same in dimension;
(7) updating formula of particle position:
Figure GDA0003104932750000087
xi＝xi+vi
wherein v isiRepresenting the velocity, x, of the particleiRepresenting the current position of the particle, i representing the current number of iterations per particle update, gbestiRepresenting the optimal position of the population as a whole, pbestiRepresents the optimal position of the individual in the population, w represents an inertial weight factor, phi1Represents the individual cognitive learning rate, phi2Representing social learning rate, and rand () is the interval [0, 1%]With the followingThe number of machines, K representing the convergence factor, K can be described as:
Figure GDA0003104932750000088
the standard particle swarm algorithm generally sets an inertia weight factor: w is at+1＝wtSince the inertial weight factor is a variable affecting the current particle velocity, a larger value is favorable for global search, a smaller value is favorable for local search, and in order to better balance the search capability, an a-GPR model is proposed:
Figure GDA0003104932750000089
wherein: w is atRepresenting the inertial weight factor, w, of each iterationmaxIs the maximum value of inertia, wminIs the inertia minimum value, T is the current iteration time, T is the maximum iteration time,
therefore, the update formula of the particle position is:
Figure GDA0003104932750000091
xi＝xi+vi
because the inertia weight factor is changed all the time, the speed and the position of the particles are continuously updated all the time in a large range and a small range of the population, thereby balancing the capability of the algorithm in global search and local search;
(8) recalculating the fitness function E of the particle according to the updated position of the particle, where the fitness function may be described as:
Figure GDA0003104932750000092
where m is the number of input samples, yiRepresenting the process output value, y, of the current A-GPR model trainingi-1I represents the current iteration number for the output value of the previous generation model;
(9) judging whether the progress requirement is met or whether the iteration times are reached, and if the progress requirement is met, terminating the search; otherwise, carrying out next search;
(10) outputting the information of the final particles, namely the parameters of the A-GPR model;
4) the method comprises the steps of acquiring meteorological data and economic data of a moment to be predicted by utilizing the function of a real-time power load prediction module, establishing a comprehensive index system, inputting the comprehensive index system and corresponding load data into an A-GPR model as input vectors, and outputting the input vectors, namely the load value of the moment to be predicted.
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
According to the method, an actual power grid of a certain region is taken as an embodiment, historical load data of the region in 2014-2017 and corresponding historical meteorological data are collected, wherein the influence factor data comprise temperature, humidity, atmospheric pressure, rainfall, wind direction, wind speed, visibility and GDP. And analyzing by using a multi-index dynamic matching method.
TABLE 1 correlation coefficient of load with various influencing factors
Figure GDA0003104932750000093
Figure GDA0003104932750000101
Correlation analysis is performed by using 2016 data, the correlation degree of each meteorological factor is changed all the year round, the correlation degree of the weather condition and the GDP is the largest in spring, the correlation degree of rainfall and humidity is the largest in summer, the correlation degree of wind speed and temperature is the largest in autumn, and the correlation degree of temperature and the GDP is the largest in winter. And selecting influence factors according to the date of the forecast day, and establishing a comprehensive index system.
The GPR prediction model was:
Figure GDA0003104932750000102
wherein: y' represents the output of the test sample, y represents the output of the training sample, and K (X, X) is the training input variable X and the test input variable X*Is the n X1 order covariance function matrix, K (X, X) is the n X n order covariance function matrix of the training input variable X, I is the unit order matrix, deltan 2Is a hyper-parameter of the model.
The difficulty of establishing the model lies in solving the hyper-parameters of the model, and the hyper-parameters of the model mainly exist in a covariance function and white noise; therefore, to solve the model hyper-parameters, a specific form of the covariance function is first determined. The covariance function of the gaussian process is expressed in the form:
Figure GDA0003104932750000103
δijis a Kronecher constant, δp 2，l，δn 2Being a hyper-parameter, δ, of the modelp 2Is the variance of the kernel function of the model, l is the feature width, and i, j represents a certain column of the matrix.
In order to optimize the parameters of the model, the invention provides an A-GPR model, which comprises the following steps:
step 1: and initializing particle swarm information such as flight speed, current position and the like by taking the variance as fitness according to the hyper-parameter constraint condition.
Step 2: inputting the hyper-parameter speed and position information, the training sample and the test sample of the influence factor into a GPR regression model according to the formula
Figure GDA0003104932750000104
And performing rolling prediction on the test sample of the load value, and calculating the fitness of the particles.
And step 3: and updating the optimal values of the individual positions of the particles and the optimal values of the group positions, and judging whether the optimal fitness of the group meets the requirements. If yes, the iteration is finished, otherwise, the step 4 is carried out.
And 4, step 4: according to the formula
Figure GDA0003104932750000105
And xi＝xi+viAnd (5) updating the speed and position information of the particles, and turning to the step 2.
Setting the population size N as 200, the maximum iteration number T as 150, wmax＝0.9,wmin0.4. And after the algorithm is finished, obtaining the optimal information of the model parameters, and then testing the trained GPR model to establish an A-GPR load prediction model.
By establishing an A-GPR real-time load prediction model, comprehensive meteorological factors of historical load data and prediction time are used as improved A-GPR model input quantity, a load prediction value of the prediction time is used as an output quantity of the model, and 2016, 4, 20 and 2016 of the area are selected as prediction days. And training and testing the A-GPR model by using data from 1 month and 1 day to 4 months and 19 days as a training sample set, wherein the table 2 is an error analysis table of the predicted load value and the predicted value and the real value of the GPR model for predicting the A-GPR model.
TABLE 2 error analysis table for each time point
Figure GDA0003104932750000111
Figure GDA0003104932750000121
TABLE 3 average error analysis Table
Categories Mean error
GPR 5.7％
A-GPR 1.4％
As can be seen from tables 2 and 3, the real-time load prediction model based on multi-index dynamic matching has good prediction accuracy and can obtain good prediction results.
The above description is a specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily change or replace the present invention within the scope of the present invention, and therefore, the present invention shall be subject to the protection scope of the claims.

Claims (2)

1. A real-time power load prediction method based on multi-index dynamic matching is characterized by comprising the following contents:
1) the data acquisition module is used for acquiring data by utilizing respective functions of a meteorological data unit, a load data unit and an economic data unit of the data acquisition module, and the meteorological data unit is used for acquiring temperature, humidity, precipitation, visibility, wind direction, wind speed and weather condition data of a required area in real time to integrate the data and transmitting the data to a system database to generate a meteorological data table; the load data unit is responsible for collecting industrial power loads, agricultural power loads, municipal power loads, post and post electricity power loads, traffic power loads, life power loads and commercial power loads in real time for data integration and transmitting the data integration to a system database to generate a load data table; the economic data unit is responsible for collecting economic data in the annual economic report of the government in real time for data integration and transmitting the economic data to the system database to generate an economic data table;
2) the data analysis module adopts a Multi-Index Dynamic Matching algorithm, namely, the MIDM algorithm carries out correlation analysis on each Index, and analyzes the correlation degree between the load data and the corresponding meteorological data and economic data, thereby determining the key indexes influencing the required area, wherein the Multi-Index Dynamic Matching algorithm comprises the following steps:
(1) calculating the projection distance of each index curve to the load:
Figure FDA0003242403710000011
wherein xqiRepresenting the projected abscissa, k, of a point on the influence index curve on the load curveqiThe slope of this point of the load curve, b1(2) Abscissa representing point on the influence index curve, b1(1) Ordinate representing a point on the curve of the influence indicator, bqiRepresenting the intercept of the load curve at this point, yqiRepresents the projection ordinate of the point on the influence index on the load curve, q represents 24 points on the curve 24 hours a day of the influence index, i represents a certain influence index, rqiRepresenting the projected distance of the impact indicator to the load, xqi2Abscissa, x, representing a point subsequent to the point of projection of the influence index curve on the load curveqi1Abscissa, y, representing a point preceding the point of projection of the influence index curve on the load curveqi2Longitudinal coordinate, y, of the point of influence index curve after the projection point of the load curveqi1Representing the ordinate, R, of the influence index curve at a point preceding the projection point of the load curvetRepresenting the fluctuation distance of the load between two points, t representing 24 points on the curve at 24 hours a day, yt2Ordinate, y, representing the latter point on the load curvet1Ordinate, x, representing the previous point on the load curvet2Abscissa, x, representing the latter point on the load curvet1The ordinate represents the previous point on the load curve;
(2) and (3) calculating the weight of each index:
Figure FDA0003242403710000021
whereinwiRepresenting the influence weight of each influence factor on the load, i represents a certain influence index, rqiRepresenting the projected distance of the influencing factor to the load, RtRepresenting the fluctuation distance of the load between two points;
(3) and (3) calculating the weight of the set index:
Figure FDA0003242403710000022
i, j ≠ 1,2
Figure FDA0003242403710000023
Wherein n represents the number of indices, wi,jIndirect weights representing combined indicators, i, j representing an influence indicator, wi' represents the weight of the set index, and fully expresses the direct and indirect influence of each index on the load;
3) establishing a comprehensive dynamic index system according to the real-time correlation of the load data and the key indexes of different regions by using the function of a multi-index dynamic matching module; establishing an Adjustable Gaussian Process Regression model (A-GPR) by using the selected load data and a comprehensive index system, wherein the Adjustable Gaussian Process Regression model comprises the following steps:
(1) initializing population size N, maximum number of iterations Tmax；
(2)
Figure FDA0003242403710000024
A vector of the current position of the particle;
(3) fitness, the Fitness value of vector x;
(4)
Figure FDA0003242403710000025
velocity of particle, its dimension and vector
Figure FDA0003242403710000026
Are the same as (a);
(5)pbestthe fitness value corresponding to the best position encountered by each particle in the flight process;
(6)
Figure FDA0003242403710000027
for recording the best position encountered during the flight of the particles, dimensions and vectors of pbest
Figure FDA0003242403710000028
Are the same in dimension;
(7) updating formula of particle position:
Figure FDA00032424037100000211
xi＝xi+vi
wherein v isiRepresenting the velocity, x, of the particleiRepresenting the current position of the particle, i representing the current number of iterations per particle update, gbestiRepresenting the optimal position of the population as a whole, pbestiRepresents the optimal position of the individual in the population, w represents an inertial weight factor,
Figure FDA0003242403710000029
represents the cognitive learning rate of the individual,
Figure FDA00032424037100000210
representing social learning rate, and rand () is the interval [0, 1%]Random number in between, K represents the convergence factor, K can be described as:
Figure FDA0003242403710000031
the standard particle swarm algorithm generally sets an inertia weight factor: w is at+1＝wtSince the inertial weight factor is a variable that affects the current particle velocityIn order to better balance the search capability, an A-GPR model is provided:
Figure FDA0003242403710000032
wherein: w is atAn inertial weight factor, w, representing the number of current iterationsmaxIs the maximum value of inertia, wminIs the inertia minimum value, T is the current iteration time, T is the maximum iteration time,
therefore, the update formula of the particle position is:
Figure FDA0003242403710000034
xi＝xi+vi
because the inertia weight factor is changed all the time, the speed and the position of the particles are continuously updated all the time in a large range and a small range of the population, thereby balancing the capability of the algorithm in global search and local search;
(8) recalculating the fitness function E of the particle according to the updated position of the particle, where the fitness function may be described as:
Figure FDA0003242403710000033
where m is the number of input samples, yiRepresenting the process output value, y, of the current A-GPR model trainingi-1I represents the current iteration number for the output value of the previous generation model;
(9) judging whether the progress requirement is met or whether the iteration times are reached, and if the progress requirement is met, terminating the search; otherwise, carrying out next search;
(10) outputting the information of the final particles, namely the parameters of the A-GPR model;
4) the method comprises the steps of acquiring meteorological data and economic data of a moment to be predicted by utilizing the function of a real-time power load prediction module, establishing a comprehensive index system, inputting the comprehensive index system and corresponding load data into an A-GPR model as input vectors, and outputting the input vectors, namely the load value of the moment to be predicted.
2. A real-time power load prediction system based on multi-index dynamic matching, which realizes the real-time power load prediction method of claim 1, and is characterized by comprising the following steps: the data acquisition module is used for acquiring load, meteorological and economic data, performing classified management and constructing a system database; calculating the correlation between the influence factors and the load based on the MIDM algorithm, sorting, screening out key attributes, and constructing a data analysis module of a comprehensive index system; the multi-index dynamic matching module of the A-GPR model can be constructed by converting fixed inertia factors in the model into dynamically changed inertia factors according to random changes of different regions and time; the real-time power load prediction module can predict the real-time power load according to the constructed comprehensive index system and the A-GPR model; the data acquisition module is in signal connection with the data analysis module, the data analysis module is in signal connection with the multi-index dynamic matching module, and the multi-index dynamic matching module is in signal connection with the real-time power load prediction module.
CN201810525917.1A 2018-05-29 2018-05-29 Real-time power load prediction system and method based on multi-index dynamic matching Active CN108830405B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810525917.1A CN108830405B (en) 2018-05-29 2018-05-29 Real-time power load prediction system and method based on multi-index dynamic matching

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810525917.1A CN108830405B (en) 2018-05-29 2018-05-29 Real-time power load prediction system and method based on multi-index dynamic matching

Publications (2)

Publication Number Publication Date
CN108830405A CN108830405A (en) 2018-11-16
CN108830405B true CN108830405B (en) 2021-11-30

Family

ID=64146506

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810525917.1A Active CN108830405B (en) 2018-05-29 2018-05-29 Real-time power load prediction system and method based on multi-index dynamic matching

Country Status (1)

Country Link
CN (1) CN108830405B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111275240B (en) * 2019-12-27 2023-06-09 华北电力大学 Load prediction method based on multi-energy coupling scene
CN111582619B (en) * 2020-01-22 2023-09-26 汕头大学 Adaptive design method based on correlation and dependency analysis
CN113283659B (en) * 2021-06-03 2022-11-22 上海分未信息科技有限公司 Power load response task allocation method for virtual power plant

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102694800A (en) * 2012-05-18 2012-09-26 华北电力大学 Gaussian process regression method for predicting network security situation
CN105701572A (en) * 2016-01-13 2016-06-22 国网甘肃省电力公司电力科学研究院 Photovoltaic short-term output prediction method based on improved Gaussian process regression
CN106709261A (en) * 2017-01-10 2017-05-24 辽宁工程技术大学 Method for evaluating mine disaster
CN106779129A (en) * 2015-11-19 2017-05-31 华北电力大学（保定） A kind of Short-Term Load Forecasting Method for considering meteorologic factor
CN106971240A (en) * 2017-03-16 2017-07-21 河海大学 The short-term load forecasting method that a kind of variables choice is returned with Gaussian process
CN106997495A (en) * 2017-04-13 2017-08-01 云南电网有限责任公司电力科学研究院 A kind of Methods of electric load forecasting

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102694800A (en) * 2012-05-18 2012-09-26 华北电力大学 Gaussian process regression method for predicting network security situation
CN106779129A (en) * 2015-11-19 2017-05-31 华北电力大学（保定） A kind of Short-Term Load Forecasting Method for considering meteorologic factor
CN105701572A (en) * 2016-01-13 2016-06-22 国网甘肃省电力公司电力科学研究院 Photovoltaic short-term output prediction method based on improved Gaussian process regression
CN106709261A (en) * 2017-01-10 2017-05-24 辽宁工程技术大学 Method for evaluating mine disaster
CN106971240A (en) * 2017-03-16 2017-07-21 河海大学 The short-term load forecasting method that a kind of variables choice is returned with Gaussian process
CN106997495A (en) * 2017-04-13 2017-08-01 云南电网有限责任公司电力科学研究院 A kind of Methods of electric load forecasting

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
基于变量选择与高斯过程回归的短期负荷预测;梁智 等;《电力建设》;20170201;第38卷(第2期);122-128 *
数据挖掘技术用于负荷与负荷影响因素的相关性分析;李莉 等;《华北电力大学学报》;20061130;第33卷(第6期);15-19 *
粒子群优化算法简介;diaokeng7360;《https://blog.csdn.net/diaokeng7360/article/details/102222272》;20161020;网页全文 *

Also Published As

Publication number Publication date
CN108830405A (en) 2018-11-16

Similar Documents

Publication Publication Date Title
CN112949945B (en) Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network
CN106650767B (en) Flood forecasting method based on cluster analysis and real-time correction
CN108898251B (en) Offshore wind farm power prediction method considering meteorological similarity and power fluctuation
CN110728411B (en) High-low altitude area combined rainfall prediction method based on convolutional neural network
CN110648014B (en) Regional wind power prediction method and system based on space-time quantile regression
CN110782658B (en) Traffic prediction method based on LightGBM algorithm
CN108388962B (en) Wind power prediction system and method
CN112288164B (en) Wind power combined prediction method considering spatial correlation and correcting numerical weather forecast
CN110555989B (en) Xgboost algorithm-based traffic prediction method
CN106600037B (en) Multi-parameter auxiliary load prediction method based on principal component analysis
CN113205207A (en) XGboost algorithm-based short-term power consumption load fluctuation prediction method and system
CN108830405B (en) Real-time power load prediction system and method based on multi-index dynamic matching
CN114676822B (en) Multi-attribute fusion air quality forecasting method based on deep learning
CN115270965A (en) Power distribution network line fault prediction method and device
CN112285807A (en) Meteorological information prediction method and device
CN105678406A (en) Short-term load prediction method based on cloud model
CN114202103A (en) Machine learning-based holiday scenic spot passenger flow prediction method
CN108334988A (en) A kind of short-term Load Forecasting based on SVM
CN113554466A (en) Short-term power consumption prediction model construction method, prediction method and device
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
CN115907822A (en) Load characteristic index relevance mining method considering region and economic influence
CN112288157A (en) Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning
CN115907228A (en) Short-term power load prediction analysis method based on PSO-LSSVM
CN116561569A (en) Industrial power load identification method based on EO feature selection and AdaBoost algorithm
CN114169226A (en) Short-term power load prediction method, computer device, and storage medium

Legal Events

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