CN110991723A - Application method of artificial intelligence in seasonal load prediction - Google Patents

Application method of artificial intelligence in seasonal load prediction Download PDF

Info

Publication number
CN110991723A
CN110991723A CN201911182133.4A CN201911182133A CN110991723A CN 110991723 A CN110991723 A CN 110991723A CN 201911182133 A CN201911182133 A CN 201911182133A CN 110991723 A CN110991723 A CN 110991723A
Authority
CN
China
Prior art keywords
load
temperature
formula
max
day
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.)
Withdrawn
Application number
CN201911182133.4A
Other languages
Chinese (zh)
Inventor
王冰
张秋桥
王敏
张鹏
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.)
Hohai University HHU
Original Assignee
Hohai University HHU
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 Hohai University HHU filed Critical Hohai University HHU
Priority to CN201911182133.4A priority Critical patent/CN110991723A/en
Publication of CN110991723A publication Critical patent/CN110991723A/en
Withdrawn legal-status Critical Current

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
    • 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
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

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

Abstract

The invention discloses an application method of artificial intelligence in seasonal load prediction, which comprises the steps of firstly obtaining meteorological load data based on a method class of eliminating trend loads, obtaining a linear trend load function by adopting least square fitting, and subtracting the trend loads from an original load sequence to obtain the meteorological load data; secondly, performing clustering analysis on the sample data by applying an FCM clustering algorithm to obtain three training samples of different types; then, establishing a temperature correction model, and considering the temperature accumulation effect of high-temperature weather in summer, wherein the temperature correction is necessary to be carried out on the highest temperature of the high-temperature weather; and finally, establishing a PSO-ELM load prediction model, and combining a particle swarm optimization algorithm (PSO) with an Extreme Learning Machine (ELM), so that the load prediction precision is effectively improved, and the method has important practical application value.

Description

Application method of artificial intelligence in seasonal load prediction
Technical Field
The invention relates to the technical field of meteorological load prediction, in particular to an application method of artificial intelligence in seasonal load prediction,
background art:
a great deal of research shows that the relation between meteorological factors such as temperature, humidity and wind power and short-term load change of electric power is particularly important, wherein the most critical meteorological factor is temperature change, and the second is the meteorological factor such as humidity and wind power. Along with the development and improvement of the social and economic level of China, more and more refrigerating and heating equipment enter the lives of residents, the improvement of the proportion of the temperature sensitive load to the total load of electric power can obviously affect an electric power system, the cooling load in summer and the heating load in winter can be directly affected by meteorological factors, if rainfall can directly affect agricultural irrigation and drainage loads and other loads, and the agricultural machinery and irrigation and drainage loads are often greatly reduced after the rainfall.
With the increasing influence of meteorological factors on the power load, the load forecasting precision requirement cannot be met only by considering the influence of daily characteristic meteorological factors on the load. Meanwhile, real-time meteorological conditions which can be provided by related meteorological departments are more and more abundant, and short-term load prediction work enters a new stage by combining historical data and online prediction information of the meteorological departments. Meteorological factors include many aspects, and the factors with the greatest influence on the load prediction accuracy mainly include temperature, humidity, rainfall and the like. The influence of temperature is considered separately, and it can be classified into daily maximum temperature, daily minimum temperature, daily average temperature, temperature difference, and humidity and rainfall can also be classified into several grades. Considering that rainfall mainly affects temperature and humidity data are not complete, the content of the research of the invention is mainly the influence of temperature-related factors on load prediction. Meanwhile, the influence of the accumulated temperature effect of the summer high-temperature weather on the load prediction accuracy is large, the influence of the temperature accumulation effect of continuous high-temperature days on load change researched by a plurality of relevant documents is large, and a load correction model considering the summer temperature accumulation effect is provided. At the present stage, a lot of researches are related to the influence of meteorological factors on load prediction, and the research method mainly comprises the steps of establishing a regression model algorithm of load and temperature and modeling by using an artificial intelligence algorithm.
Disclosure of Invention
The invention aims to provide an application method of artificial intelligence in seasonal load prediction to overcome the defect that continuous high temperature is inaccurate in meteorological load calculation in the prior art.
A method of applying artificial intelligence in seasonal load prediction, the method comprising the steps of:
taking the acquired meteorological load data as sample data;
carrying out cluster analysis on the sample data;
inputting data obtained by clustering analysis into a pre-constructed temperature correction model;
and inputting the result of the temperature correction model into a pre-constructed load prediction model to obtain a prediction result.
Further, the method for acquiring meteorological load data comprises the following steps:
acquiring an original load record, and extracting an original load sequence;
obtaining the trend load according to the basic load and the linear trend coefficient,
the formula is as follows:
Ee＝at+b；
in the formula, t represents a time variable, a is a linear trend coefficient, and b is a basic load;
and subtracting the trend load from the original load sequence to obtain the meteorological load, wherein the meteorological load is obtained by the following formula:
Em＝E-Ee；
in the formula, EmIndicating meteorological load, EeFor trending loads, E represents the original load sequence.
Further, the method for performing cluster analysis on the sample data comprises the following steps:
setting initialization parameters, including:
initializing a clustering center V (0), an iteration counter T (0), and a fuzzy weight index m (2);
calculating a fuzzy membership matrix U (t);
can be obtained by the following formula:
Figure BDA0002291549800000031
in the formula uijIs a sample ziFuzzy membership degree of j class, 0 ≤ uij≤1，
Figure BDA0002291549800000032
dijRepresenting a sample ziTo class j center vjThe euclidean distance of (a) can be obtained by:
Figure BDA0002291549800000033
an objective function:
||V(t)-V(t+1)||<ε；
where ε represents an iteration stop condition, t represents the number of iterations, V(t)A matrix of the cluster centers is represented,
Figure BDA0002291549800000034
can be obtained by the following formula:
Figure BDA0002291549800000035
in the formula, N represents the number of samples,
Figure BDA0002291549800000036
indicating that the ith sample is subordinate to the jth class size in the tth iteration;
the values represented by U and V are output.
Further, the method for constructing the load prediction model comprises the following steps:
calculating to obtain a hidden layer output matrix of the neural network:
Figure BDA0002291549800000041
and calculating to obtain T:
Figure BDA0002291549800000042
in the formula βi＝[βi1,βi2,…,βim]Representing the connection weight between the ith neuron of the hidden layer and each output layer neuron, N being the number of training samples, L being the number of neurons of the hidden layer, x being the input of the neurons, W being the connection weight between the input layer and the hidden layer, β being the connection weight between the hidden layer and the output layer, biA threshold value representing the ith neuron of the hidden layer, g (x) being the excitation function;
obtaining a target function of the extreme learning machine model according to the hidden layer output matrix and the T of the neural network;
Figure BDA0002291549800000043
the improved particle swarm optimization algorithm has the following formula:
Figure BDA0002291549800000044
where i is 1,2, …, m, d denotes the dimension of the space, k denotes the number of iterations, and a non-negative constant c1And c2Called the learning factor, r1And r2Is a random number between (0,1), Mi＝(mi1,mi2,…,mid) Denotes the position of the ith particle in d-dimensional space, Vi＝(vi1,vi2,…,vid) Representing the velocity of the ith particle in d-dimensional space, ω is called the inertial weight factor and can be obtained by:
Figure BDA0002291549800000051
in the formula, DmaxIs the maximum number of iterations, ωmaxAnd ωminRespectively, the maximum and minimum values of the inertial weight, typically ωmaxValue of 0.9, omegaminThe value is 0.4.
Further, the operation method for inputting the data obtained by the cluster analysis into the pre-constructed temperature correction model comprises the following steps:
obtaining the highest temperature of the day to be measured in the data after the clustering analysis;
comparing the highest temperature of the day to be measured with a set threshold temperature;
if the highest temperature of the day to be measured is larger than the set threshold temperature, inputting the cluster analysis data into a temperature correction model;
otherwise, directly inputting the cluster analysis data into a pre-constructed load prediction model for prediction.
Further, the method for constructing the temperature correction model comprises the following steps:
determining a temperature correction function:
Figure BDA0002291549800000052
in the formula, TiThe highest temperature of day i to be predicted, Ti≥Tmin；Ti-jThe highest temperature on the jth day before the day to be predicted; t isminIs the limit temperature of the high temperature day; k is a radical ofij∈[0,1]Is a cumulative effect coefficient, where ki1>ki2>…>kip；p＝min(n,dmax) Day maximum temperature n days before day i continuously above the limit temperature, dmaxMaximum cumulative days;
Tmincan be obtained by the following formula:
Figure BDA0002291549800000053
Figure BDA0002291549800000061
where L ═ f (t) denotes the functional relationship between load and temperature, and is referred to as the temperature rise curve;
cumulative effect coefficient kijCan be obtained by the following formula:
Figure BDA0002291549800000062
s.t.0≤kij≤1；
ki1>ki2>…>kip；
in the above formula, TiDenotes the original maximum temperature, Ti' modified maximum temperature, kijFor cumulative effect coefficients, LiRepresenting the maximum load.
Further, the maximum cumulative days dmaxCan be obtained by the following steps:
selecting 60 data in 7 and 8 months in summer as samples, and using adjacent dmaxSample data of +1 day as a group, 60-d in totalmaxGroup (d);
each sample is taken to the following equation to find the correlation coefficient and then divided by 60-dmaxObtaining an average correlation coefficient;
then d is chosen to maximize the average correlation coefficientmaxAs maximum cumulative days:
Figure BDA0002291549800000063
in the above formula, ave _ R is an average correlation coefficient between T, L; t, L are the highest temperature vector and the highest load vector, respectively; cov (T, L) is the covariance of T and L;
Figure BDA0002291549800000064
the mean square error of T and L respectively;
Figure BDA0002291549800000071
the invention has the advantages that:
1) according to the method, the trend load is removed from an original load sequence to obtain the meteorological load, and a least square fitting trend load function is adopted; considering that the difference between seasonal loads is large, an FCM clustering algorithm is adopted to perform clustering analysis on samples, and correlation calculation shows that the correlation degree between the loads and the temperature is improved greatly after the loads are processed by the method;
2) the temperature correction model is established from three aspects of limit temperature determination, maximum accumulation day determination and cumulative effect coefficient solving, and correlation analysis shows that the correlation coefficient between the corrected temperature and the load is obviously improved.
Drawings
FIG. 1 is a general flow chart of seasonal load prediction in the present invention;
FIG. 2 is a daily maximum load variation curve and a trend load fitting function curve according to the present invention;
FIG. 3 is a flow chart of the FCM clustering algorithm of the present invention;
FIG. 4 is a flow chart of cumulative effect of air temperature in the present invention;
FIG. 5 is a flow chart of the PSO-ELM algorithm of the present invention;
FIG. 6 is a diagram of the summer cooling load prediction result of the present invention.
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further described with the specific embodiments.
As shown in fig. 1 to 6, a method for applying artificial intelligence to seasonal load prediction includes the following steps:
taking the acquired meteorological load data as sample data;
carrying out cluster analysis on the sample data;
inputting data obtained by clustering analysis into a pre-constructed temperature correction model;
and inputting the result of the temperature correction model into a pre-constructed load prediction model to obtain a prediction result.
In this embodiment, the method for acquiring meteorological load data includes the following steps:
acquiring an original load record, and extracting an original load sequence;
obtaining the trend load according to the basic load and the linear trend coefficient,
the formula is as follows:
Ee＝at+b；
in the formula, t represents a time variable, a is a linear trend coefficient, and b is a basic load;
and subtracting the trend load from the original load sequence to obtain the meteorological load, wherein the meteorological load is obtained by the following formula:
Em＝E-Ee；
in the formula, EmIndicating meteorological load, EeFor trending loads, E represents the original load sequence.
In this embodiment, the method for performing cluster analysis on sample data includes the following steps:
setting initialization parameters, including:
initializing a cluster center V ═ V(0)The iteration counter T is 0, and the fuzzy weight index m is 2;
calculating a fuzzy membership matrix U(t)；
Can be obtained by the following formula:
Figure BDA0002291549800000091
in the formula uijIs a sample xiFuzzy membership degree of j class, 0 ≤ uij≤1，
Figure BDA0002291549800000092
dijRepresents a sample xiTo class j center vjThe euclidean distance of (a) can be obtained by:
Figure BDA0002291549800000093
an objective function:
||V(t)-V(t+1)||<ε；
where ε represents an iteration stop condition, t represents the number of iterations, V(t)Representing cluster centersThe matrix is a matrix of a plurality of matrices,
Figure BDA0002291549800000094
can be obtained by the following formula:
Figure BDA0002291549800000095
in the formula, N represents the number of samples,
Figure BDA0002291549800000096
indicating that the ith sample is subordinate to the jth class size in the tth iteration;
the values represented by U and V are output.
In this embodiment, the method for constructing the load prediction model includes the following steps:
calculating to obtain a hidden layer output matrix of the neural network:
Figure BDA0002291549800000097
and calculating to obtain T:
Figure BDA0002291549800000101
in the formula βi＝[βi1,βi2,…,βim]Representing the connection weight between the ith neuron of the hidden layer and each output layer neuron, N being the number of training samples, L being the number of neurons of the hidden layer, x being the input of the neurons, W being the connection weight between the input layer and the hidden layer, β being the connection weight between the hidden layer and the output layer, biA threshold value representing the ith neuron of the hidden layer, g (x) being the excitation function;
obtaining a target function of the extreme learning machine model according to the hidden layer output matrix and the T of the neural network;
Figure BDA0002291549800000102
the improved particle swarm optimization algorithm has the following formula:
Figure BDA0002291549800000103
where i is 1,2, …, m, d denotes the dimension of the space, k denotes the number of iterations, and a non-negative constant c1And c2Called the learning factor, r1And r2Is a random number between (0,1), Mi＝(mi1,mi2,…,mid) Denotes the position of the ith particle in d-dimensional space, Vi＝(vi1,vi2,…,vid) Representing the velocity of the ith particle in d-dimensional space, ω is called the inertial weight factor and can be obtained by:
Figure BDA0002291549800000104
in the formula, DmaxIs the maximum number of iterations, ωmaxAnd ωminRespectively, the maximum and minimum values of the inertial weight, typically ωmaxValue of 0.9, omegaminThe value is 0.4.
In this embodiment, the operation method of inputting data obtained by cluster analysis into a pre-constructed temperature correction model includes the following steps:
obtaining the highest temperature of the day to be measured in the data after the clustering analysis;
comparing the highest temperature of the day to be measured with a set threshold temperature;
if the highest temperature of the day to be measured is larger than the set threshold temperature, inputting the cluster analysis data into a temperature correction model;
otherwise, directly inputting the cluster analysis data into a pre-constructed load prediction model for prediction.
In this embodiment, the method for constructing the temperature correction model includes the following steps:
determining a temperature correction function:
Figure BDA0002291549800000111
in the formula, TiThe highest temperature of day i to be predicted, Ti≥Tmin；Ti-jThe highest temperature on the jth day before the day to be predicted; t isminIs the limit temperature of the high temperature day; k is a radical ofij∈[0,1]Is a cumulative effect coefficient, where ki1>ki2>…>kip；p＝min(n,dmax) Day maximum temperature n days before day i continuously above the limit temperature, dmaxMaximum cumulative days;
Tmincan be obtained by the following formula:
Figure BDA0002291549800000112
Figure BDA0002291549800000113
where L ═ f (t) denotes the functional relationship between load and temperature, and is referred to as the temperature rise curve;
cumulative effect coefficient kijCan be obtained by the following formula:
Figure BDA0002291549800000121
s.t.0≤kij≤1；
ki1>ki2>…>kip；
in the above formula, TiDenotes the original maximum temperature, Ti' modified maximum temperature, kijFor cumulative effect coefficients, LiRepresenting the maximum load.
In the present embodiment, the maximum cumulative number of days dmaxCan be obtained by the following steps:
selecting 60 data in 7 and 8 months in summer as samples, and using adjacent dmaxSample data of +1 day as a group, 60-d in totalmaxGroup (d);
each sample is taken to the following equation to find the correlation coefficient and then divided by 60-dmaxObtaining an average correlation coefficient;
then d is chosen to maximize the average correlation coefficientmaxAs maximum cumulative days:
Figure BDA0002291549800000122
in the above formula, ave _ R is an average correlation coefficient between T, L; t, L are the highest temperature vector and the highest load vector, respectively; cov (T, L) is the covariance of T and L;
Figure BDA0002291549800000123
the mean square error of T and L respectively;
Figure BDA0002291549800000124
based on the above, the method comprises the following specific implementation steps:
s1: acquiring meteorological load data based on a method class of eliminating trend loads, acquiring a linear trend load function by adopting least square fitting, and subtracting the trend loads from an original load sequence to obtain the meteorological load data;
s2: clustering and analyzing the sample data by applying an FCM clustering algorithm to obtain three training samples of different types, and analyzing the cooling load in summer and the influence factors thereof;
s3: establishing a temperature correction model, wherein the temperature cumulative effect of high-temperature weather in summer is considered, the highest temperature of the high-temperature weather needs to be corrected, and the establishment of the model mainly comprises three processes of determining the limit temperature, determining the maximum cumulative days and solving the cumulative effect coefficient;
s4: establishing a PSO-ELM load prediction model, combining a Particle Swarm Optimization (PSO) algorithm with an Extreme Learning Machine (ELM), taking relevant meteorological factors as the input of the model, taking meteorological load as the output of the model, comparing a prediction result with an actual value, and evaluating the performance of the PSO-ELM load prediction model through a root mean square error and an average absolute percentage error.
The PSO is an optimization algorithm for optimizing the ELM model, that is, selecting the optimal parameters for the ELM model, and the optimization process is the construction process of the PSO-ELM model, that is: firstly, the root mean square error is used as an objective function of PSO optimization, and the root mean square error is calculated through an actual value and a predicted value of an ELM (the process is a training process of a model); and then, selecting a parameter of the corresponding ELM model when the root mean square error is minimized through the updating principle of the PSO, wherein the parameter is the optimal parameter. The PSO-ELM model has been constructed so far, and the following is an input to the model to complete the prediction.
In step S1, the meteorological load data is obtained based on the method class of eliminating the trend load, the linear trend load function is obtained by using least square fitting, and the trend load is subtracted from the original load sequence to obtain the meteorological load, that is:
Em＝E-Ee； (1)
in the formula (1), EmIndicating meteorological load, EeFor trend loads, E represents the original load sequence, where the trend loads EeCan be obtained by the formula (2):
Ee＝at+b； (2)
in the formula (2), t represents a time variable, a is a linear trend coefficient, and b is a base load.
In step S2, the FCM clustering algorithm first sets Z ═ Z for given N training samples1,z2,…,zNV, which is classified into C class, V ═ V1,v2,…,vcIs C cluster centers. In practical problems, C is generally given manually, an iteration stop condition epsilon is set, and a clustering center V ═ V is initialized at the same time(0)Setting an iteration counter T as 0 and a fuzzy weight index m as 2; then calculating a membership matrix U(t)This is obtained by the formula (3):
Figure BDA0002291549800000141
in the formula (3), uijIs a sample ziFuzzy membership to class j,0≤uij≤1，
Figure BDA0002291549800000142
dijrepresenting a sample ziTo class j center vjThe euclidean distance of (a) can be obtained by equation (4):
Figure BDA0002291549800000143
in step S2, the objective function in the FCM clustering algorithm is:
||V(t)-V(t+1)||<ε (5)
in the formula (5), ε represents an iteration stop condition, t represents the number of iterations, and V(t)A matrix of the cluster centers is represented,
Figure BDA0002291549800000144
can be obtained by the formula (6):
Figure BDA0002291549800000145
in the formula (6), N represents the number of samples,
Figure BDA0002291549800000146
indicating that the ith sample is subject to the jth class size in the tth iteration.
In step S3, temperature correction needs to be performed on the highest temperature of the high-temperature weather, the establishment of the model mainly includes three processes of determining the limit temperature, determining the maximum cumulative days, and solving the cumulative effect coefficient, and the temperature correction function is:
Figure BDA0002291549800000151
in the formula (7), TiThe highest temperature of day i to be predicted, Ti≥Tmin；Ti-jThe highest temperature on the jth day before the day to be predicted; t isminIs the limit temperature of the high temperature day; k is a radical ofij∈[0,1]In order to accumulate the coefficients of effect,and satisfies the condition k according to the principle of' big-end-up and small-end-upi1>ki2>…>kip；p＝min(n,dmax) Day maximum temperature n days before day i continuously above the limit temperature, dmaxThe maximum number of days accumulated. T isminCan be obtained by the following formulas (8) and (9):
Figure BDA0002291549800000152
Figure BDA0002291549800000153
in the formula (8), L ═ f (t) represents a functional relationship between load and temperature, and is referred to as a temperature rise curve;
maximum cumulative days d in formula (7)maxCan be obtained by the formula (10):
Figure BDA0002291549800000154
in the formula (10), R is a correlation coefficient between T, L; t, L are the highest temperature vector and the highest load vector, respectively; cov (T, L) is the covariance of T and L;
Figure BDA0002291549800000155
the mean square error of T and L respectively;
Figure BDA0002291549800000156
the cumulative effect coefficient in the formula (7) can be obtained by the following formula:
Figure BDA0002291549800000161
s.t. 0≤kij≤1 (12)
ki1>ki2>…>kip(13)
in formula (11), TiDenotes the original maximum temperature, Ti' modified maximum temperature, kijFor cumulative effect coefficients, LiRepresenting the maximum load.
In step S4, the objective function of the extreme learning machine model is:
Figure BDA0002291549800000162
in the formula (14), H is a hidden layer output matrix of the neural network and can be obtained by the formula (15), T' is a transpose of the matrix T, T is an output of the extreme learning machine and can be obtained by the formula (16), βi＝[βi1,βi2,…,βim]Representing the connection weight between the ith neuron of the hidden layer and each neuron of the output layer;
Figure BDA0002291549800000163
Figure BDA0002291549800000164
in the formulas (15) and (16), N is the number of training samples, L is the number of neurons in the hidden layer, W is the connection weight between the input layer and the hidden layer, β is the connection weight between the hidden layer and the output layer, biThe threshold representing the ith neuron of the hidden layer, g (x), is the excitation function.
In step S4, the improved particle swarm optimization algorithm has an update formula as follows:
Figure BDA0002291549800000171
in equation (8), i is 1,2, …, m, d represents the dimension of space, k represents the number of iterations, and a non-negative constant c1And c2Called the learning factor, r1And r2Is a random number between (0,1), Mi＝(mi1,mi2,…,mid) Denotes the position of the ith particle in d-dimensional space, Vi＝(vi1,vi2,…,vid) Representing the velocity of the ith particle in d-dimensional space, ω being called the inertial weightThe factor can be obtained by the formula (9):
Figure BDA0002291549800000172
in the formula (9), TmaxIs the maximum number of iterations, ωmaxAnd ωminRespectively, the maximum and minimum values of the inertial weight, typically ωmaxValue of 0.9, omegaminThe value is 0.4.
In step S4, an extreme learning machine is used to establish a space load prediction model, and parameters of the model are optimized by using a particle swarm optimization, thereby improving the accuracy of load prediction. The number of neurons of an input layer, the number of neurons of a hidden layer and the number of neurons of an output layer of the extreme learning machine are preset, unknown parameter variables of the model are that the connection weight between the input layer and the hidden layer is W and the neuron threshold value b of the hidden layer, an excitation function adopts a Sigmoid function, and an objective function is to enable the number of neurons of the input layer, the hidden layer and the output layer to be preset, the unknown parameter variables of the model
Figure BDA0002291549800000173
And minimum. Therefore, the PSO optimization process is to find the optimal connection weight between the input layer and the hidden layer of the extreme learning machine as W and the hidden layer neuron threshold b, so that the objective function is minimized. And through continuous iteration of the particle swarm, the target function of the model is used as the evaluation index of the particle swarm, and the optimal parameters of the model are output after the accuracy requirement is finally met.
The solution according to the invention is further illustrated below in a specific example:
step 1: collecting hourly power load data, daily maximum load, minimum load, power consumption and other data from 2014 to 2018 Shanghai city, drawing daily maximum power load change curve, and obtaining a trend load function E by adopting a least square fitting method as shown in FIG. 2eThe trend loads were subtracted from the original load sequence to obtain meteorological load data, and the correlation coefficients were used to represent the linear correlation between temperature and load, with the results shown in table 1.
TABLE 1 correlation coefficient comparison
Figure BDA0002291549800000181
Step 2: the meteorological load has typical seasonal variation, generally, the cooling load in high-temperature weather in summer is large, the heating load in winter is large, and the load generated by empty days in spring and autumn is small, so the sample data is clustered into three types, and the clustering result is shown in table 2.
TABLE 2 clustering center of three types of sample data
Figure BDA0002291549800000191
And step 3: and (3) selecting the summer high-temperature weather in the step (2) as a simulation experiment, correcting the temperature of the high-temperature weather, and correcting the influence of the cumulative effect of the weather exceeding the limit temperature on the temperature of the day in two days before the consideration, wherein the cumulative effect coefficient is shown in a table 3.
TABLE 3 cumulative Effect coefficients
T0℃ ki1 ki2
<31 0 0
[31,32) 0.28 0
[32,33) 0.41 0.12
[33,34) 0.52 0.17
[34,35) 0.65 0.26
[35,36) 0.75 0.44
[36,37) 0.68 0.30
[37,38) 0.24 0
≥38 0 0
After simulation, the limit temperature of the temperature correction model is 31 ℃, the maximum cumulative days are 2 days, and 17 sample data in 2018 summer of Shanghai city are subjected to temperature correction by combining the cumulative effect coefficients in the table 3. The correction results are shown in table 4.
TABLE 4 temperature correction results
Date Maximum temperature/. degree.C Temperature correction/. degree.C
7.15 35 36.38
7.16 37.7 42.024
7.17 36.1 38.208
7.18 36.3 38.544
7.19 36.2 39.336
7.20 36.6 40.128
7.21 35.2 37.356
7.22 34.8 36.69
7.23 34.5 36.135
7.24 35.3 37.393
7.25 35.2 37.422
7.26 35.1 37.599
7.27 35.8 39.132
7.28 35.6 38.694
7.29 36.8 41.322
7.30 35.3 37.876
7.31 34.5 36.075
And 4, step 4: fig. 6 shows the predicted results of the trend-eliminated load and the temperature-corrected load, the root-mean-square error and the average absolute percentage error are used as the result evaluation indexes, and the predicted results are shown in table 5.
TABLE 5 load prediction results
Figure BDA0002291549800000211
The above prediction results are counted, the absolute percentage error of 11 points is less than 1%, the absolute percentage error of 6 points is less than 5%, the maximum percentage error is 8.55%, and the average absolute percentage error is 1.51%. The prediction accuracy is much higher than that of the traditional load prediction, and the method has important practical application value.
It will be appreciated by those skilled in the art that the invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The embodiments disclosed above are therefore to be considered in all respects as illustrative and not restrictive. All changes which come within the scope of or equivalence to the invention are intended to be embraced therein.

Claims (7)

1. A method for applying artificial intelligence to seasonal load prediction, the method comprising the steps of:
taking the acquired meteorological load data as sample data;
carrying out cluster analysis on the sample data;
inputting data obtained by clustering analysis into a pre-constructed temperature correction model;
and inputting the result of the temperature correction model into a pre-constructed load prediction model to obtain a prediction result.
2. The method of claim 1, wherein the artificial intelligence is applied to seasonal load prediction: the method for acquiring the meteorological load data comprises the following steps:
acquiring an original load record, and extracting an original load sequence;
obtaining the trend load according to the basic load and the linear trend coefficient,
the formula is as follows:
Ee＝at+b；
in the formula, t represents a time variable, a is a linear trend coefficient, and b is a basic load;
and subtracting the trend load from the original load sequence to obtain the meteorological load, wherein the meteorological load is obtained by the following formula:
Em＝E-Ee；
in the formula, EmIndicating meteorological load, EeFor trending loads, E represents the original load sequence.
3. The method of claim 1, wherein the artificial intelligence is applied to seasonal load prediction: the method for carrying out cluster analysis on the sample data comprises the following steps:
setting initialization parameters, including:
initializing a cluster center V ═ V(0)The iteration counter T is 0, and the fuzzy weight index m is 2;
calculating a fuzzy membership matrix U(t)；
Can be obtained by the following formula:
Figure FDA0002291549790000021
in the formula uijIs a sample ziFuzzy membership degree of j class, 0 ≤ uij≤1，
Figure FDA0002291549790000022
dijRepresenting a sample ziTo class j center vjThe euclidean distance of (a) can be obtained by:
Figure FDA0002291549790000023
an objective function:
||V(t)-V(t+1)||<ε；
in which ε represents the iteration stopCondition, t denotes the number of iterations, V(t)A matrix of the cluster centers is represented,
Figure FDA0002291549790000024
can be obtained by the following formula:
Figure FDA0002291549790000025
in the formula, N represents the number of samples,
Figure FDA0002291549790000026
indicating that the ith sample is subordinate to the jth class size in the tth iteration;
the values represented by U and V are output.
4. The method of claim 1, wherein the artificial intelligence is applied to seasonal load prediction: the construction method of the load prediction model comprises the following steps:
calculating to obtain a hidden layer output matrix of the neural network:
Figure FDA0002291549790000027
and calculating to obtain T:
Figure FDA0002291549790000031
in the formula βi＝[βi1,βi2,…,βim]Representing the connection weight between the ith neuron of the hidden layer and each output layer neuron, N being the number of training samples, L being the number of neurons of the hidden layer, x being the input of the neurons, W being the connection weight between the input layer and the hidden layer, β being the connection weight between the hidden layer and the output layer, biA threshold value representing the ith neuron of the hidden layer, g (x) being the excitation function;
obtaining a target function of the extreme learning machine model according to the hidden layer output matrix and the T of the neural network;
Figure FDA0002291549790000032
the improved particle swarm optimization algorithm has the following formula:
Figure FDA0002291549790000033
where i is 1,2, …, m, d denotes the dimension of the space, k denotes the number of iterations, and a non-negative constant c1And c2Called the learning factor, r1And r2Is a random number between (0,1), Mi＝(mi1,mi2,…,mid) Denotes the position of the ith particle in d-dimensional space, Vi＝(vi1,vi2,…,vid) Representing the velocity of the ith particle in d-dimensional space, ω is called the inertial weight factor and can be obtained by:
Figure FDA0002291549790000034
in the formula, DmaxIs the maximum number of iterations, ωmaxAnd ωminRespectively, the maximum and minimum values of the inertial weight, typically ωmaxValue of 0.9, omegaminThe value is 0.4.
5. The method of claim 1, wherein the artificial intelligence is applied to seasonal load prediction: the operation method for inputting data obtained by clustering analysis into a pre-constructed temperature correction model comprises the following steps:
obtaining the highest temperature of the day to be measured in the data after the clustering analysis;
comparing the highest temperature of the day to be measured with a set threshold temperature;
if the highest temperature of the day to be measured is larger than the set threshold temperature, inputting the cluster analysis data into a temperature correction model;
otherwise, directly inputting the cluster analysis data into a pre-constructed load prediction model for prediction.
6. The method of claim 1, wherein the artificial intelligence is applied to seasonal load prediction: the construction method of the temperature correction model comprises the following steps:
determining a temperature correction function:
Figure FDA0002291549790000041
in the formula, TiThe highest temperature of day i to be predicted, Ti≥Tmin；Ti-jThe highest temperature on the jth day before the day to be predicted; t isminIs the limit temperature of the high temperature day; k is a radical ofij∈[0,1]Is a cumulative effect coefficient, where ki1>ki2>…>kip；p＝min(n,dmax) Day maximum temperature n days before day i continuously above the limit temperature, dmaxMaximum cumulative days;
Tmincan be obtained by the following formula:
Figure FDA0002291549790000042
Figure FDA0002291549790000043
where L ═ f (t) denotes the functional relationship between load and temperature, and is referred to as the temperature rise curve;
cumulative effect coefficient kijCan be obtained by the following formula:
Figure FDA0002291549790000051
s.t.0≤kij≤1；
ki1>ki2>…>kip；
the above formulaIn, TiDenotes the original maximum temperature, Ti' modified maximum temperature, kijFor cumulative effect coefficients, LiRepresenting the maximum load.
7. The method of claim 6, wherein the artificial intelligence is applied to seasonal load prediction: the maximum cumulative days dmaxCan be obtained by the following steps:
selecting 60 data in 7 and 8 months in summer as samples, and using adjacent dmaxSample data of +1 day as a group, 60-d in totalmaxGroup (d);
each sample is taken to the following equation to find the correlation coefficient and then divided by 60-dmaxObtaining an average correlation coefficient;
then d is chosen to maximize the average correlation coefficientmaxAs maximum cumulative days:
Figure FDA0002291549790000052
in the above formula, ave _ R is an average correlation coefficient between T, L; t, L are the highest temperature vector and the highest load vector, respectively; cov (T, L) is the covariance of T and L;
Figure FDA0002291549790000053
the mean square error of T and L respectively;
Figure FDA0002291549790000054
CN201911182133.4A 2019-11-27 2019-11-27 Application method of artificial intelligence in seasonal load prediction Withdrawn CN110991723A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911182133.4A CN110991723A (en) 2019-11-27 2019-11-27 Application method of artificial intelligence in seasonal load prediction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911182133.4A CN110991723A (en) 2019-11-27 2019-11-27 Application method of artificial intelligence in seasonal load prediction

Publications (1)

Publication Number Publication Date
CN110991723A true CN110991723A (en) 2020-04-10

Family

ID=70087344

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911182133.4A Withdrawn CN110991723A (en) 2019-11-27 2019-11-27 Application method of artificial intelligence in seasonal load prediction

Country Status (1)

Country Link
CN (1) CN110991723A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113268913A (en) * 2021-06-24 2021-08-17 广州鼎泰智慧能源科技有限公司 Intelligent building air conditioner cooling machine system operation optimization method based on PSO-ELM algorithm
CN117371243A (en) * 2023-11-03 2024-01-09 国网河北省电力有限公司 Temperature sensing correction method considering temperature accumulation effect in short-term load prediction

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113268913A (en) * 2021-06-24 2021-08-17 广州鼎泰智慧能源科技有限公司 Intelligent building air conditioner cooling machine system operation optimization method based on PSO-ELM algorithm
CN117371243A (en) * 2023-11-03 2024-01-09 国网河北省电力有限公司 Temperature sensing correction method considering temperature accumulation effect in short-term load prediction

Similar Documents

Publication Publication Date Title
CN108022001B (en) Short-term load probability density prediction method based on PCA (principal component analysis) and quantile regression forest
CN110414788B (en) Electric energy quality prediction method based on similar days and improved LSTM
CN112288164B (en) Wind power combined prediction method considering spatial correlation and correcting numerical weather forecast
WO2018045642A1 (en) A bus bar load forecasting method
CN107730031B (en) Ultra-short-term peak load prediction method and system
CN107967542B (en) Long-short term memory network-based electricity sales amount prediction method
CN111428926B (en) Regional power load prediction method considering meteorological factors
CN110070282B (en) Low-voltage transformer area line loss influence factor analysis method based on comprehensive relevance
CN111105104A (en) Short-term power load prediction method based on similar day and RBF neural network
CN111260136A (en) Building short-term load prediction method based on ARIMA-LSTM combined model
CN113554466B (en) Short-term electricity consumption prediction model construction method, prediction method and device
CN111028100A (en) Refined short-term load prediction method, device and medium considering meteorological factors
CN106600037B (en) Multi-parameter auxiliary load prediction method based on principal component analysis
CN109389238B (en) Ridge regression-based short-term load prediction method and device
CN111915092A (en) Ultra-short-term wind power prediction method based on long-time and short-time memory neural network
CN111144440A (en) Method and device for analyzing daily power load characteristics of special transformer user
CN114862032B (en) XGBoost-LSTM-based power grid load prediction method and device
CN111178585A (en) Fault reporting amount prediction method based on multi-algorithm model fusion
CN111401657A (en) Transformer hot spot temperature time sequence prediction method based on data mining algorithm
CN116316599A (en) Intelligent electricity load prediction method
CN112288157A (en) Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning
CN110991723A (en) Application method of artificial intelligence in seasonal load prediction
CN113361202A (en) Distribution transformer heavy overload rapid early warning method based on characteristic load prediction
CN115358437A (en) Power supply load prediction method based on convolutional neural network
CN116205508A (en) Distributed photovoltaic power generation abnormality diagnosis method and system

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20200410