CN114444760A - Industry user electric quantity prediction method based on mode extraction and error adjustment - Google Patents
Industry user electric quantity prediction method based on mode extraction and error adjustment Download PDFInfo
- Publication number
- CN114444760A CN114444760A CN202111526907.8A CN202111526907A CN114444760A CN 114444760 A CN114444760 A CN 114444760A CN 202111526907 A CN202111526907 A CN 202111526907A CN 114444760 A CN114444760 A CN 114444760A
- Authority
- CN
- China
- Prior art keywords
- electric quantity
- value
- network
- user
- predicted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000000605 extraction Methods 0.000 title claims abstract description 17
- 241000287127 Passeridae Species 0.000 claims abstract description 51
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 238000003064 k means clustering Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 32
- 238000012549 training Methods 0.000 claims description 24
- 238000005457 optimization Methods 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 11
- 230000005611 electricity Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000002596 correlated effect Effects 0.000 claims description 8
- 239000000126 substance Substances 0.000 claims description 7
- 238000003491 array Methods 0.000 claims description 4
- 230000000875 corresponding effect Effects 0.000 claims description 4
- 230000005284 excitation Effects 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 244000062645 predators Species 0.000 claims description 4
- 238000012163 sequencing technique Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000005057 refrigeration Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Game Theory and Decision Science (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Entrepreneurship & Innovation (AREA)
- Probability & Statistics with Applications (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of industry user load prediction, in particular to an industry user electric quantity prediction method based on mode extraction and error adjustment, which comprises the following operation steps: s1, extracting a typical industry annual load curve by using a K-means clustering algorithm; s2, analyzing the correlation degree between external macroscopic factors such as holidays and meteorology and the electric quantity of the user to be predicted by using a maximum information coefficient method; s3, according to the typical industry load curve and external factors such as holidays, meteorology and the like, using sparrows to search for an optimized BP neural network for predicting the electric quantity; s4, obtaining the probability distribution of the residual error by using a nonparametric estimation method, and adjusting the prediction result to obtain the final predicted value of the electric quantity.
Description
Technical Field
The invention relates to the technical field of industry user load prediction, in particular to an industry user electric quantity prediction method based on mode extraction and error adjustment.
Background
The power load prediction is the basis for implementing various power consumption client-oriented applications, and is always concerned by power research personnel, and by means of a large amount of data, the power consumption modes and power consumption of users in different industries can be predicted more accurately, so that on one hand, the power consumption behaviors of different industries can be better known, and better power consumption service is provided; on the other hand, the method can help a power grid company to further perfect a power selling scheme and adjust the working mode of the power grid, so that energy and cost can be saved, efficiency can be improved, accurate prediction becomes more and more important at present when the economic society is rapidly developed, the more the power utilization modes of users are mastered, the more the power grid company is favorable for optimizing the power supply structure of the power grid company, and the core competitiveness of the power grid company is improved.
In recent years, the industrial structure of China is changed greatly, from agriculture to industry, the third industry is developed rapidly, and the electricity utilization characteristics in different periods are changed along with the change of the industrial structure and the development of new industries, so that the traditional prediction model taking all the electricity utilization industries as a whole has defects, and prediction errors can be generated by neglecting the internal characteristics of the electricity utilization industries for prediction.
In developed countries in the western world, due to the fact that the development of each industry is quite mature and stable, the electric quantity prediction is carried out by taking industry users as units, the prediction is quite accurate, and the electric quantity prediction method has targeted guiding significance, China is still in the preliminary stage of social connotation, the economic development is rapid, the development mode is often in the transition period, the industry pattern is changed violently, and a plurality of problems need to be researched and solved in the aspects of analysis of industry electric quantity historical data and selection of industry electric quantity prediction methods.
Therefore, for the improvement of the existing industry user load prediction method, a novel industry user electric quantity prediction method based on mode extraction and error adjustment is designed to change the technical defects, and the practicability of the whole industry user load prediction method is improved, which is very important.
Disclosure of Invention
The invention aims to provide an industry user electric quantity prediction method based on mode extraction and error adjustment, so as to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme:
an industry user electric quantity prediction method based on mode extraction and error adjustment comprises the following operation steps:
s1, extracting a typical industry annual load curve by using a K-means clustering algorithm;
s2, analyzing the correlation degree between external macroscopic factors such as holidays and meteorology and the electric quantity of the user to be predicted by using a maximum information coefficient method;
s3, according to the typical industry load curve and external factors such as holidays, weather and the like, the electric quantity is predicted by searching the optimized BP neural network with sparrows;
and S4, obtaining the probability distribution of the residual error by using a nonparametric estimation method, and adjusting the prediction result to obtain the final predicted value of the electric quantity.
As a preferred scheme of the present invention, the core of the K-means algorithm in S1 is to target a given data set to minimize a loss function, and the algorithm flow specifically includes the following contents:
s11, standardizing the original data to prevent the large number from eating the small number;
S13, defining a loss functionM is the number of users, μiThe electric quantity a of the ith useriThe cluster center of (a);
s14, let t be 0,1, 2.. for iteration steps, the following two steps are repeated until convergence:
s141, the electric quantity x of the ith user is calculatediWhich is assigned to the center closest thereto
Wherein the content of the first and second substances,for the electric quantity a after the t-th iterationiNearest clustering center, t is iteration step number;
s142, recalculating the initial position of the clustering center
Wherein, the first and the second end of the pipe are connected with each other,representing the kth new clustering center after the t-th iteration is finished;
in the algorithm, Euclidean distance is used as similarity measurement, and a loss function is the sum of squares of errors of the center points of clusters to which the distance of each group of electric quantity data belongs.
As a preferred embodiment of the present invention, the specific operation steps in S2 are as follows:
s21, quantifying main influence factors such as holidays, weather and the like, and defining quantified values of workdays, weekends and other holidays as: 2. 1, 0; the quantized values for sunny days, cloudy days, light rain, medium rain, heavy rain and snow days are defined as: 1. 2, 3, 4, 5, 6, 7;
s22, solving the correlation coefficients of the holidays, the highest temperature, the lowest temperature, the weather, the wind level and the like and the electric quantity by using a maximum information coefficient algorithm, wherein the solving method of the maximum information coefficient comprises the following steps:
wherein m ═ 1,2, the highest temperature, the lowest temperature, the weather and the wind level, X (m) is a quantization array of the influence factors, Y is an electric quantity array of a user to be predicted, X and Y are elements in the arrays X and Y respectively, p (X) and p (Y) respectively represent the probability of X and Y, p (X, Y) represents the joint probability between the variables X and Y, and B is a constant, and is usually 0.6 th power of the data quantity;
in general, when Rmic(x, y) is greater than or equal to 0.8, and the two are considered to be highly correlated; when R ismic(x, y) is greater than or equal to 0.5, and the two are considered to be moderately correlated; when 0.5 > Rmic(x, y) is more than or equal to 0.3, and the two are considered to be relatively related; when R ismicThe relationship (x, y) < 0.3 is very weak and is considered irrelevant; when the maximum information coefficient of the external macroscopic factor and the electric quantity to be predicted does not exceed 0.3, the external macroscopic factor and the electric quantity to be predicted are considered not to influence the electric quantity of the user, so that the parameter is not used as an input characteristic of the prediction network.
As a preferred embodiment of the present invention, in S3, the connection weight between the neurons of the BP neural network is optimized by using a sparrow search optimization algorithmAn implicit layer threshold p and an output layer threshold q; taking an industry typical year load curve, external macroscopic factors influencing electric quantity and historical load data as network input, obtaining a prediction result through the training of a sparrow search optimized BP neural network,
s31, wherein the sparrow search optimization algorithm comprises the following steps:
s311, initializing a population consisting of n sparrows as follows
Wherein n sparrows form a population Z, n is usually 20,the position of the nth sparrow of the d-th dimension is represented, namely the initial value of the weight or the threshold to be optimized, and d represents the number of the weight and the threshold to be optimized;
s312, calculating fitness value and sequencing
Wherein, FzRepresenting an array of fitness values, fjRepresenting the training accuracy, also called a moderate value, obtained by bringing the weight or threshold value corresponding to the j-th sparrow into the network;
s313, updating the position of the predator
Wherein s is the current iteration number, C is the maximum iteration number, and ZjlIndicates the position of the jth sparrow in the l-dimension, alpha and R2The number is a random number between 0 and 1, the ST safety value is 0.8, Q is a random number which follows normal distribution, and U is a full 1 matrix;
s314, updating the position of the joiner
Wherein Z ispFor the optimal position of the current finder, i.e. the optimal value of the current weight or threshold, ZwFor the current worst position, i.e. the worst value of the current weight or threshold, the elements of matrix A are then assigned a value of 1 or-1, A+=AT(AAT)-1;
S315, updating the position of the alert person
Wherein Z isbbeta-N (0,1) is a random number for the current optimal position,is a random number between-1 and 1, fbAnd fwRespectively representing the current optimum and worst moderate values, wherein epsilon is a constant to avoid the denominator being 0;
and S316, calculating the fitness value, updating the sparrow position, judging whether to stop or not, and otherwise, repeatedly executing.
S32, wherein the BP neural network comprises the following steps:
s321, setting initial values, and determining the number V of nodes of the network input layer according to the data characteristics of the network input and output sequence (X, Y)ⅠNumber of hidden layer nodes VⅡAnd number of output layer nodes VⅢAssigning the optimal initial value of sparrow search to the connection weight between neuronsThe hidden layer threshold p and the output layer threshold q (initialization is to generate random numbers), given a learning rate, a Sigmoid function is selected as a neuron excitation function Sm, and the calculation formula is as follows:
s322, calculating the output of the hidden layer, and according to the input matrix X of the network (T, P) and the connection weight w of each layerij,wik,wimAnd a threshold value p of the hidden layer, calculating the hidden layer output H
Wherein T is influence factor data such as weather, holidays and the like, P is electric quantity data of a user to be predicted, and l is the number of nodes of a hidden layer; sm is expressed as a hidden layer stimulus function;
s323, calculating an output layer, and connecting weight w of the neuron according to an output value H of the hidden layerij,wik,wimAnd a threshold value q, calculating the output value O of the BP neural network
S324, calculating errors, and calculating a network prediction error e according to the network output value (electric quantity prediction value) O and the expected output Y
ek=Yk-Okk=1,2,...,m
S325, updating the weight value, updating the network connection weight value W according to the error e of the network output value (predicted value of electricity quantity)1,W2,W3
wij=wij+ek=Yk-Ok k=1,2,...,m
wjk=wjk+ηHjek j=1,2,...l k=1,2,...,m
Wherein η is the learning rate;
s326, updating the threshold values p and q among the network nodes according to the error e of the network output value (predicted value)
bk=bk+ek k=1,2,...,m
And S327, iterating, judging whether the iteration of the algorithm is finished, and returning to S322 if the iteration of the algorithm is not finished.
As a preferable embodiment of the present invention, the specific calculation method in S4 is as follows:
s41, obtaining the probability distribution of the electric quantity residual error after the network training by using a nonparametric estimation method, and obtaining the probability distribution of the electric quantity residual error by using a kernel density estimation method
Wherein f (g) is a residual probability function of electric quantity, giOptimizing the electric quantity residual value of BP neural network training for a sparrow search optimization algorithm, wherein lambda is the quantity of electric quantity residual errors, h is 0.5 and is a smoothing parameter, and K ishIs a Gaussian kernel function;
s42, integrating the obtained electric quantity residual error according to the probability function f (g) of the electric quantity residual error to obtain the mathematical expectation w of the training residual errorsI.e. the adjustment amount
ws=∫f(g)dx
S43, overlapping to obtain the final user electric quantity predicted value
w=wr+ws
Wherein, wrFor the predicted value of the user's electric quantity, wsThe amount of adjustment for the amount of power of the user obtained in step S42.
Compared with the prior art, the invention has the beneficial effects that:
1. in the invention, the method of the maximum information coefficient is adopted to analyze the degree of influence of external factors on the electric quantity, so that the method has universality and fairness and can more accurately describe the degree of influence of different factors on the electric quantity.
2. In the invention, a sparrow search optimization algorithm is adopted to give an optimal initial value of the traditional BP network, and a typical load curve of the industry is used as characteristic input, thereby being beneficial to improving the prediction accuracy.
Drawings
FIG. 1 is a flow chart of an industry user power prediction method based on pattern extraction and error adjustment according to the present invention;
FIG. 2 is a typical load curve extraction result based on K-maens clustering according to the present invention;
FIG. 3 is a diagram illustrating predicted results according to the present invention;
FIG. 4 is a training set residual distribution diagram according to the present invention;
FIG. 5 is a residual probability density distribution diagram according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described below clearly and completely in conjunction with the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments, and all other embodiments obtained by those skilled in the art without making creative work based on the embodiments of the present invention belong to the protection scope of the present invention.
While several embodiments of the present invention will be described below in order to facilitate an understanding of the invention, with reference to the related description, the invention may be embodied in many different forms and is not limited to the embodiments described herein, but rather, these embodiments are provided so that this disclosure will be thorough and complete.
It will be understood that when an element is referred to as being "secured to" another element, it can be directly on the other element or intervening elements may also be present, that when an element is referred to as being "connected" to another element, it can be directly connected to the other element or intervening elements may also be present, and that the terms "vertical", "horizontal", "left", "right" and the like are used herein for descriptive purposes only.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, and the terms used herein in the specification of the present invention are for the purpose of describing particular embodiments only and are not intended to limit the present invention, and the term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.
Referring to fig. 1-5, the present invention provides a technical solution:
an industry user electric quantity prediction method based on mode extraction and error adjustment comprises the following operation steps:
s1, extracting a typical industry annual load curve by using a K-means clustering algorithm;
s2, analyzing the correlation degree between external macroscopic factors such as holidays and meteorology and the electric quantity of the user to be predicted by using a maximum information coefficient method;
s3, according to the typical industry load curve and external factors such as holidays, weather and the like, the electric quantity is predicted by searching the optimized BP neural network with sparrows;
and S4, obtaining the probability distribution of the residual error by using a nonparametric estimation method, and adjusting the prediction result to obtain the final predicted value of the electric quantity.
Further, the core of the K-means algorithm in S1 is to target a given data set to minimize a loss function, and the algorithm flow specifically includes the following contents:
s11, standardizing the original data to prevent the large number from eating the small number;
S13, defining a loss functionM is the number of users, μiThe electric quantity a of the ith useriThe cluster center of (a);
s14, let t equal to 0,1, 2.. for iteration steps, repeat the following two steps until convergence:
s141, the electric quantity x of the ith user is calculatediWhich is assigned to the center closest thereto
Wherein the content of the first and second substances,for the electric quantity a after the t-th iterationiNearest clustering center, t is iteration step number;
s142, recalculating the initial position of the clustering center
Wherein the content of the first and second substances,representing the kth new clustering center after the t-th iteration is finished;
in the algorithm, Euclidean distance is used as similarity measurement, and a loss function is the sum of squares of errors of the center points of clusters to which the distance of each group of electric quantity data belongs.
Further, the specific operation steps in S2 are as follows:
s21, quantifying main influence factors such as holidays, weather and the like, and defining quantified values of workdays, weekends and other holidays as: 2. 1, 0; the quantized values for sunny days, cloudy days, light rain, medium rain, heavy rain and snow days are defined as: 1. 2, 3, 4, 5, 6, 7;
s22, solving the correlation coefficients of the holidays, the highest temperature, the lowest temperature, the weather, the wind level and the like and the electric quantity by using a maximum information coefficient algorithm, wherein the solving method of the maximum information coefficient comprises the following steps:
wherein, m ═ 1,2,. and 5 represent five influence factors of holiday, maximum temperature, lowest temperature, weather and wind level, X (m) is a quantized array of the influence factors, Y is an electric quantity array of a user to be predicted, X and Y are elements in the arrays X and Y respectively, p (X) and p (Y) represent probabilities of X and Y respectively, p (X, Y) represents a joint probability between variables X and Y, and B is a constant, and the data quantity is generally taken to the power of 0.6;
in general, when Rmic(x, y) is greater than or equal to 0.8, and the two are considered to be highly correlated; when R ismic(x, y) is greater than or equal to 0.5, and the two are considered to be moderately correlated; when 0.5 > Rmic(x, y) is more than or equal to 0.3, and the two are considered to be relatively related; when R ismicThe relationship (x, y) < 0.3 is very weak and is considered irrelevant; when the maximum information coefficient of the external macroscopic factor and the electric quantity to be predicted does not exceed 0.3, the electric quantity of the user is not influenced, so that the parameter is not used as an input characteristic of the prediction network.
Further, in S3, the connection weights between the neurons of the BP neural network are optimized by using a sparrow search optimization algorithmAn implicit layer threshold p and an output layer threshold q; taking an industry typical annual load curve, external macroscopic factors influencing electric quantity and historical load data as network input, obtaining a prediction result through the training of a sparrow search optimized BP neural network,
s31, wherein the sparrow search optimization algorithm comprises the following steps:
s311, initializing a population consisting of n sparrows as follows
Wherein n sparrows form a population Z, n is usually 20,the position of the nth sparrow of the d-th dimension is represented, namely the initial value of the weight or the threshold to be optimized, and d represents the number of the weight and the threshold to be optimized;
s312, calculating fitness value and sequencing
Wherein, FzRepresenting an array of fitness values, fjRepresenting the training accuracy obtained by bringing the weight or threshold value corresponding to the jth sparrow into the network, which is also called a moderate value;
s313, updating the position of the predator
Wherein s is the current iteration number, C is the maximum iteration number, and ZjlIndicating the position of the jth sparrow in the l-dimension, alpha and R2The number is a random number between 0 and 1, the ST safety value is 0.8, Q is a random number which follows normal distribution, and U is a full 1 matrix;
s314, updating the position of the joiner
Wherein Z ispFor the optimal position of the current finder, i.e. the optimal value of the current weight or threshold, ZwFor the current worst position, i.e. the worst value of the current weight or threshold, the elements of matrix A are then assigned a value of 1 or-1, A+=AT(AAT)-1;
S315, updating the position of the alert person
Wherein Z isbbeta-N (0,1) is a random number for the current optimal position,is a random number between-1 and 1, fbAnd fwRespectively representing the current optimum and worst moderate values, wherein epsilon is a constant to avoid the denominator being 0;
s316, calculating the fitness value, updating the sparrow position, judging whether to stop or not, and otherwise, repeatedly executing;
s32, wherein the BP neural network comprises the following steps:
s321, setting initial values, and determining the number V of nodes of the network input layer according to the data characteristics of the network input and output sequence (X, Y)ⅠNumber of hidden layer nodes VⅡAnd number of output layer nodes VⅢAssigning the optimal initial value of sparrow search to the connection weight between neuronsThe hidden layer threshold p and the output layer threshold q (initialization, i.e., generation of random numbers), given a learning rate, select the Sigmoid function as the neuron excitation function Sm, whose calculation formula is:
s322, calculating the output of the hidden layer, and according to the input matrix X of the network (T, P) and the connection weight w of each layerij,wik,wimAnd a threshold value p of the hidden layer, calculating the hidden layer output H
Wherein T is data of influence factors such as weather, holidays and the like, P is electric quantity data of a user to be predicted, and l is represented by the number of nodes of a hidden layer; sm is expressed as a hidden layer stimulus function;
s323, calculating an output layer, and connecting weight w of the neuron according to an output value H of the hidden layerij,wik,wimAnd a threshold value q, calculating the output value O of the BP neural network
S324, calculating errors, and calculating a network prediction error e according to the network output value (electric quantity prediction value) O and the expected output Y
ek=Yk-Okk=1,2,...,m
S325, updating the weight value, updating the network connection weight value W according to the error e of the network output value (predicted value of electricity quantity)1,W2,W3
wij=wij+ek=Yk-Qk k=1,2,...,m
wjk=wjk+ηHjek j=1,2,...,l k=1,2,...,m
Wherein η is the learning rate;
s326, updating the threshold values p and q among the network nodes according to the error e of the network output value (predicted value)
bk=bk+ekk=1,2,...,m
And S327, iterating, judging whether the iteration of the algorithm is finished, and returning to S322 if the iteration of the algorithm is not finished.
Further, the specific calculation method in S4 is as follows:
s41, obtaining the probability distribution of the electric quantity residual error after the network training by using a nonparametric estimation method, and obtaining the probability distribution of the electric quantity residual error by using a kernel density estimation method
Wherein f (g) is a residual probability function of electric quantity, giOptimizing the electric quantity residual value of BP neural network training for a sparrow search optimization algorithm, wherein lambda is the quantity of electric quantity residual errors, h is 0.5 and is a smoothing parameter, and KhIs a Gaussian kernel function;
s42, integrating the obtained electric quantity residual error according to the probability function f (g) of the electric quantity residual error to obtain the mathematical expectation w of the training residual errorsI.e. the adjustment amount
ws=∫f(g)dx
S43, overlapping to obtain the final user electric quantity predicted value
w=wr+ws
Wherein, wrFor the predicted value of the user's electric quantity, wsThe amount of adjustment for the amount of power of the user obtained in step S42. The implementation case is as follows:
the method comprises the following steps: a typical industry annual load curve is extracted by using a K-means clustering algorithm, and the specific algorithm flow is as follows:
1) carrying out standardization processing on original data to prevent 'large numbers from eating small numbers';
3) Defining a loss functionM is the number of users, μiIs the electricity quantity a of the ith useriThe cluster center of (a);
4) let t be 0,1, 2.. for iteration steps, repeat the following two steps until convergence:
(1) the electric quantity x of the ith useriWhich is assigned to the center closest thereto
Wherein the content of the first and second substances,for the electric quantity a after the t-th iterationiNearest clustering center, t is iteration step number;
(2) recalculating the initial position of the cluster center:
wherein the content of the first and second substances,representing the kth new clustering center after the t-th iteration is finished;
in the algorithm, Euclidean distance is used as similarity measurement, and a loss function is the sum of squares of errors of the distances between each group of electric quantity data and the center point of a cluster to which the distance belongs;
step two: the method for analyzing the degree of association between external macroscopic factors such as holidays and meteorology and the electric quantity of the user to be predicted by using the maximum information coefficient comprises the following specific operation steps:
1) quantifying main influence factors such as holidays, weather and the like, and defining quantified values of workdays, weekends and other holidays as follows: 2. 1, 0; the quantized values for sunny days, cloudy days, light rain, medium rain, heavy rain and snow days are defined as: 1. 2, 3, 4, 5, 6, 7;
2) and solving the correlation coefficients of the holidays, the highest temperature, the lowest temperature, the weather, the wind level and the like and the electric quantity by using a maximum information coefficient algorithm, wherein the solving method of the maximum information coefficient comprises the following steps:
wherein, m ═ 1,2,. and 5 represent five influence factors of holiday, maximum temperature, lowest temperature, weather and wind level, X (m) is a quantized array of the influence factors, Y is an electric quantity array of a user to be predicted, X and Y are elements in the arrays X and Y respectively, p (X) and p (Y) represent probabilities of X and Y respectively, p (X, Y) represents a joint probability between variables X and Y, and B is a constant, and the data quantity is generally taken to the power of 0.6;
in general, when Rmic(x, y) is greater than or equal to 0.8, and the two are considered to be highly correlated; when R ismic(x, y) is greater than or equal to 0.5, and the two are considered to be moderately correlated; when 0.5 > Rmic(x, y) is more than or equal to 0.3, and the two are considered to be relatively related; when R ismicThe relationship (x, y) < 0.3 is very weak and is considered irrelevant; when the maximum information coefficient of the external macroscopic factor and the electric quantity to be predicted does not exceed 0.3, the external macroscopic factor and the maximum information coefficient of the electric quantity to be predicted do not influence the electric quantity of the user, so that the parameter is not used as an input characteristic of the prediction network;
step three: according to typical industry load curves and external factors such as holidays, meteorology and the like, the electric quantity is predicted by utilizing a sparrow search optimized BP neural network, and specifically, the connection weight between neurons of the BP neural network is optimized by utilizing a sparrow search optimization algorithmAn implicit layer threshold p and an output layer threshold q; taking an industry typical year load curve, external macroscopic factors influencing electric quantity and historical load data as network input, obtaining a prediction result through the training of a sparrow search optimized BP neural network,
1) the sparrow search optimization algorithm comprises the following steps:
(1) initializing a population of n sparrows, as follows
Wherein n sparrows form a population Z, n is usually 20,the position of the nth sparrow of the d-th dimension is represented, namely the initial value of the weight or the threshold to be optimized, and d represents the number of the weight and the threshold to be optimized;
(2) calculating fitness value, and sequencing
Wherein, FzRepresenting an array of fitness values, fjRepresenting the training accuracy obtained by bringing the weight or threshold value corresponding to the jth sparrow into the network, which is also called a moderate value;
(3) updating predator positions
Wherein s is the current iteration number, C is the maximum iteration number, and ZjlIndicating the position of the jth sparrow in the l-dimension, alpha and R2The number is a random number between 0 and 1, the ST safety value is 0.8, Q is a random number which follows normal distribution, and U is a full 1 matrix;
(4) updating the location of an enrollee
Wherein Z ispFor the optimal position of the current finder, i.e. the optimal value of the current weight or threshold, ZwFor the current worst position, i.e. the worst value of the current weight or threshold, the elements of matrix A are then assigned a value of 1 or-1, A+=AT(AAT)-1;
(5) Updating the position of the alert
Wherein Z isbFor the current optimum positionSetting beta-N (0,1) as random number,is a random number between-1 and 1, fbAnd fwRespectively representing the current optimum and worst moderate values, wherein epsilon is a constant to avoid the denominator being 0;
(6) calculating a fitness value, updating the position of the sparrow, judging whether to stop or not, and otherwise, repeatedly executing;
2) the BP neural network comprises the following steps:
(1) given an initial value, determining the number V of network input layer nodes based on the data characteristics of the network input and output sequences (X, Y)ⅠNumber of hidden layer nodes VⅡAnd number of output layer nodes VⅢAssigning the optimal initial value of sparrow search to the connection weight between neuronsThe hidden layer threshold p and the output layer threshold q (initialization is to generate random numbers), given a learning rate, a Sigmoid function is selected as a neuron excitation function Sm, and the calculation formula is as follows:
(2) calculating the output of the hidden layer according to the input matrix X of the network as (T, P) and the connection weight w of each layerij,wik,wimAnd a threshold value p of the hidden layer, calculating the hidden layer output H
Wherein T is influence factor data such as weather, holidays and the like, P is electric quantity data of a user to be predicted, and l is the number of nodes of a hidden layer; sm is expressed as a hidden layer stimulus function;
(3) calculating output layer according to output value H of hidden layer and connection weight w of neuronij,wik,wimAnd a threshold value q, calculating the output value O of the BP neural network
(4) Calculating error, and calculating network prediction error e according to the network output value (predicted value of electric quantity) O and the expected output Y
ek=Yk-Okk=1,2,...,m
(5) Updating the weight value, namely updating the network connection weight value W according to the error e of the network output value (electric quantity predicted value)1,W2,W3
wij=wij+ek=Yk-Ok k=1,2,...,m
wjk=wjk+ηHjek j=1,2...,l k=1,2,...,m
Wherein η is the learning rate;
(6) updating the threshold values, updating the threshold values p, q between the network nodes according to the error e of the network output value (predicted value)
bk=bk+ek k=1,2,...,m
(7) Iteration, judging whether the algorithm iteration is finished or not, and if not, returning to the step (2);
step four: obtaining the probability distribution of residual errors by using a nonparametric estimation method, and adjusting the prediction result to obtain a final predicted value of the electric quantity, wherein the specific calculation method comprises the following steps:
1) obtaining the probability distribution of the electric quantity residual after network training by using a nonparametric estimation method, and obtaining the probability distribution of the electric quantity residual by using a kernel density estimation method
Wherein f (g) is a residual probability function of electric quantity, giOptimizing the electric quantity residual value of BP neural network training for a sparrow search optimization algorithm, wherein lambda is the quantity of electric quantity residual errors, h is 0.5 and is a smoothing parameter, and K ishIs a Gaussian kernel function;
2) integrating the obtained electric quantity residual error according to the probability function f (g) of the electric quantity residual error to obtain the mathematical expectation w of the training residual errorsI.e. the adjustment amount
ws=∫f(g)dx
3) Overlapping to obtain a final user electric quantity predicted value
w=wr+ws
Wherein, wrFor the predicted value of the user's electric quantity, wsThe user electric quantity adjustment quantity obtained in the step 2).
The method takes the power data set of 28 users in the same industry in Shenzhen city, 12 months in 2019 to 2020 and 12 months as an example to verify the practicability of the power prediction method based on the mode extraction and the error adjustment for the industry users. Firstly, the electric quantity data of 27 users are taken, a typical annual load curve of the industry is obtained through clustering, and external macroscopic factors influencing the electric quantity of the user to be predicted are analyzed through a maximum information coefficient; the typical annual load curve of the industry and external factors are used as characteristic input for optimizing the BP network, the electric quantity of the user in the last week of 2020 is predicted, and finally the actual effect of the method is compared;
a) extracting an industry typical load curve
And selecting a clustering center, and clustering and extracting 27 users in the industry by using a K-means clustering algorithm, wherein the result is shown in figure 2.
As can be seen from FIG. 2, the power consumption of the industry is periodically strong, the power consumption is greatly reduced before and after the spring festival and before and after No. 10 month 1, the power consumption is greatly influenced by the holidays of the festival, and the power consumption in summer is higher than that in winter.
b) Analysis of influence of external macroscopic factors on electric quantity
The influence of meteorological factors on the power consumption is large, for example, when the temperature is high (higher than 26 ℃) or low (lower than 26 ℃), the use frequency of refrigeration or heating equipment such as an air conditioner and the like is increased suddenly, the power consumption and the load of densely-populated indoor places such as residential districts or shopping malls and the like are increased remarkably, the method for analyzing the association degree by adopting the maximum information coefficient can analyze not only the linear association degree but also the nonlinear association degree, and has universality and fairness, and the results are as follows:
TABLE 1 results of analysis of degree of association
As can be seen from the table, the correlation between the highest temperature and the lowest temperature and the electricity consumption of the industry is the largest, and the load of the industry user is strongly correlated with meteorological factors, so that the prediction accuracy can be improved to a certain extent by considering the meteorological factors which have great influence during load prediction.
c) Network setup and prediction results
According to the analysis result, an industry typical load curve, holidays, highest temperature, lowest temperature and weather are taken as 5 characteristic inputs of the network, the number of network hidden nodes is 2, the number of evolutionary times is 30, the population scale is 20, and the prediction result is shown in fig. 3.
d) Result adjustment and analysis
In order to evaluate the accuracy of the results of the industrial user electric quantity prediction method based on mode extraction and error adjustment, three indexes of a standardized absolute average error, a relative average error and a root mean square error are adopted to carry out error estimation on the prediction results and actual data values, the prediction error of the method is calculated, and the results are shown in table 2.
TABLE 2 prediction error
It can be seen that the normalized absolute average error and relative average error of the method of the present invention are between 1% and 2%, and have high accuracy.
To highlight the superiority of the present invention, the average error is shown in table 3 by comparing it with the conventional BP network prediction, the conventional BP network prediction without considering the power consumption mode of the user, and the sparrow search optimization BP network prediction without considering the power consumption mode of the user:
TABLE 3 comparison of prediction errors for different methods
Therefore, compared with a method for predicting a BP (back propagation) network without considering a user power consumption mode, a traditional BP neural network and a long-short term memory neural network, the method for extracting the user power consumption mode and optimizing and improving the BP network can obtain a better prediction effect.
Further, the predicted value is adjusted according to the training residual, the residual of the training set is collected firstly, the distribution of the residual is shown in fig. 4, the probability distribution condition of the residual of the training set is obtained by using a non-parameter estimation kernel density estimation method, and the probability density graph is shown in fig. 5; the probability density function is then integrated to obtain the adjustment amount, the error after adjustment is shown in table 4, and the comparison before and after adjustment is shown in the figure.
TABLE 4 error contrast before and after adjustment
Therefore, the prediction error can be reduced by 0.2-0.3% through adjustment, and the adjustment method has a good effect and can further reduce the prediction error.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (5)
1. An industry user electric quantity prediction method based on mode extraction and error adjustment is characterized by comprising the following operation steps:
s1, extracting a typical industry annual load curve by using a K-means clustering algorithm;
s2, analyzing the correlation degree between external macroscopic factors such as holidays and meteorology and the electric quantity of the user to be predicted by using a maximum information coefficient method;
s3, according to the typical industry load curve and external factors such as holidays, weather and the like, the electric quantity is predicted by searching the optimized BP neural network with sparrows;
and S4, obtaining the probability distribution of the residual error by using a nonparametric estimation method, and adjusting the prediction result to obtain the final predicted value of the electric quantity.
2. The industry user power prediction method based on pattern extraction and error adjustment as claimed in claim 1, wherein: the core of the K-means algorithm in S1 is to target a given data set to minimize a loss function, and the algorithm flow specifically includes the following contents:
s11, standardizing the original data to prevent the large number from eating the small number;
S13, defineLoss functionM is the number of users, μiIs the electricity quantity a of the ith useriThe cluster center of (a);
s14, let t be 0,1, 2.. for iteration steps, the following two steps are repeated until convergence:
s141, the electric quantity x of the ith user is calculatediWhich is assigned to the center closest thereto
Wherein the content of the first and second substances,for the electric quantity a after the t-th iterationiNearest clustering center, t is iteration step number;
s142, recalculating the initial position of the clustering center:
wherein the content of the first and second substances,representing the kth new clustering center after the t-th iteration is finished;
in the algorithm, Euclidean distance is used as similarity measurement, and a loss function is the sum of squares of errors of the distances between each group of electric quantity data and the center point of a cluster to which the group of electric quantity data belongs.
3. The industry user power prediction method based on pattern extraction and error adjustment as claimed in claim 1, wherein: the specific operation steps in the step S2 are as follows:
s21, quantifying main influence factors such as holidays, weather and the like, and defining quantified values of workdays, weekends and other holidays as: 2. 1, 0; the quantized values for sunny days, cloudy days, light rain, medium rain, heavy rain and snow days are defined as: 1. 2, 3, 4, 5, 6, 7;
s22, solving the correlation coefficients of the holidays, the highest temperature, the lowest temperature, the weather, the wind level and the like and the electric quantity by using a maximum information coefficient algorithm, wherein the solving method of the maximum information coefficient comprises the following steps:
wherein, m ═ 1,2,. and 5 represent five influence factors of holiday, maximum temperature, lowest temperature, weather and wind level, X (m) is a quantized array of the influence factors, Y is an electric quantity array of a user to be predicted, X and Y are elements in the arrays X and Y respectively, p (X) and p (Y) represent probabilities of X and Y respectively, p (X, Y) represents a joint probability between variables X and Y, and B is a constant, and usually the data amount is taken to the power of 0.6;
in general, when Rmic(x, y) is greater than or equal to 0.8, and the two are considered to be highly correlated; when R ismic(x, y) is greater than or equal to 0.5, and the two are considered to be moderately related; when 0.5 > Rmic(x, y) is more than or equal to 0.3, and the two are considered to be relatively related; when R ismicThe relationship (x, y) < 0.3 is very weak and is considered irrelevant; when the maximum information coefficient of the external macroscopic factor and the electric quantity to be predicted does not exceed 0.3, the external macroscopic factor and the electric quantity to be predicted do not influence the electric quantity of the user, so that the parameter is not used as the input characteristic of the prediction network.
4. The industry user power prediction method based on pattern extraction and error adjustment as claimed in claim 1, wherein: in S3, the method specifically comprises the step of optimizing the connection weight between the neurons of the BP neural network by utilizing a sparrow search optimization algorithmAn implicit layer threshold p and an output layer threshold q; taking an industry typical year load curve, external macroscopic factors influencing electric quantity and historical load data as network input, obtaining a prediction result through the training of a sparrow search optimized BP neural network,
s31, wherein the sparrow search optimization algorithm comprises the following steps:
s311, initializing a population consisting of n sparrows as follows
Wherein n sparrows form a population Z, n is usually 20,the position of the nth sparrow of the d-th dimension is represented, namely the initial value of the weight or the threshold to be optimized, and d represents the number of the weight and the threshold to be optimized;
s312, calculating fitness value and sequencing
Wherein, FzRepresenting an array of fitness values, fjRepresenting the training accuracy obtained by bringing the weight or threshold value corresponding to the jth sparrow into the network, which is also called a moderate value;
s313, updating the position of the predator
Wherein s is the current iteration number, C is the maximum iteration number, and ZjlIndicating the position of the jth sparrow in the l-dimension, alpha and R2The number is a random number between 0 and 1, the ST safety value is 0.8, Q is a random number which follows normal distribution, and U is a full 1 matrix;
s314, updating the position of the joiner
Wherein Z ispFor the optimal position of the current finder, i.e. the optimal value of the current weight or threshold, ZwFor the current worst position, i.e. the worst value of the current weight or threshold, the elements of matrix A are then assigned a value of 1 or-1, A+=AT(AAT)-1;
S315, updating the position of the alert person
Wherein Z isbbeta-N (0,1) is a random number for the current optimal position,is a random number between-1 and 1, fbAnd fwRespectively representing the current optimum and worst moderate values, wherein epsilon is a constant to avoid the denominator being 0;
s316, calculating a fitness value, updating the sparrow position, judging whether to stop, and otherwise, repeatedly executing;
s32, wherein the BP neural network comprises the following steps:
s321, setting initial values, and determining the number V of nodes of the network input layer according to the data characteristics of the network input and output sequence (X, Y)ⅠNumber of hidden layer nodes VⅡAnd number of output layer nodes VⅢAssigning the optimal initial value of sparrow search to the connection weight between neuronsThe hidden layer threshold p and the output layer threshold q (initialization is to generate random numbers), given a learning rate, a Sigmoid function is selected as a neuron excitation function Sm, and the calculation formula is as follows:
s322, calculating the output of the hidden layer, and determining the connection weight w of each layer according to the input matrix X ═ T, P of the networkij,wik,wimAnd the threshold value p of the hidden layer, calculating the output H of the hidden layer
Wherein T is influence factor data such as weather, holidays and the like, P is electric quantity data of a user to be predicted, and l is the number of nodes of a hidden layer; sm is expressed as a hidden layer stimulus function;
s323, calculating an output layer, and connecting weight w of the neuron according to an output value H of the hidden layerij,wik,wimAnd a threshold value q, calculating the output value O of the BP neural network
S324, calculating errors, and calculating a network prediction error e according to the network output value (electric quantity prediction value) O and the expected output Y
ek=Yk-Ok k=1,2,...,m
S325, updating the weight value, updating the network connection weight value W according to the error e of the network output value (predicted value of electricity quantity)1,W2,W3
wij=wij+ek=Yk-Ok k=1,2,...,m
wjk=wjk+ηHjek j=1,2,...,l k=1,2,...,m
Wherein η is the learning rate;
s326, updating the threshold value p and q between the network nodes according to the error e of the network output value (predicted value)
bk=bk+ek k=1,2,...,m
And S327, iterating, judging whether the iteration of the algorithm is finished, and returning to S322 if the iteration of the algorithm is not finished.
5. The industry user power prediction method based on pattern extraction and error adjustment as claimed in claim 1, wherein: the specific calculation method in S4 is as follows:
s41, obtaining the probability distribution of the electric quantity residual after the network training by using a nonparametric estimation method, and obtaining the probability distribution of the electric quantity residual by using a kernel density estimation method
Wherein f (g) is a power residual probability function, giOptimizing the electric quantity residual value of BP neural network training for a sparrow search optimization algorithm, wherein lambda is the quantity of electric quantity residual errors, h is 0.5 and is a smoothing parameter, and K ishIs a Gaussian kernel function;
s42, integrating the obtained electric quantity residual error according to the probability function f (g) of the electric quantity residual error to obtain the mathematical expectation w of the training residual errorsI.e. the adjustment amount
ws=∫f(g)dx
S43, overlapping to obtain the final user electric quantity predicted value
w=wr+ws
Wherein, wrFor the predicted value of the user's electric quantity, wsThe amount of adjustment for the amount of power of the user obtained in step S42.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111526907.8A CN114444760A (en) | 2021-12-14 | 2021-12-14 | Industry user electric quantity prediction method based on mode extraction and error adjustment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111526907.8A CN114444760A (en) | 2021-12-14 | 2021-12-14 | Industry user electric quantity prediction method based on mode extraction and error adjustment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114444760A true CN114444760A (en) | 2022-05-06 |
Family
ID=81364163
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111526907.8A Pending CN114444760A (en) | 2021-12-14 | 2021-12-14 | Industry user electric quantity prediction method based on mode extraction and error adjustment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114444760A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115629354A (en) * | 2022-11-30 | 2023-01-20 | 北京志翔科技股份有限公司 | Method and device for identifying out-of-tolerance electric energy meter based on power consumption adjustment amplitude |
-
2021
- 2021-12-14 CN CN202111526907.8A patent/CN114444760A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115629354A (en) * | 2022-11-30 | 2023-01-20 | 北京志翔科技股份有限公司 | Method and device for identifying out-of-tolerance electric energy meter based on power consumption adjustment amplitude |
CN115629354B (en) * | 2022-11-30 | 2023-04-07 | 北京志翔科技股份有限公司 | Method and device for identifying out-of-tolerance electric energy meter based on power consumption adjustment amplitude |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111199016B (en) | Daily load curve clustering method for improving K-means based on DTW | |
CN106485262B (en) | Bus load prediction method | |
CN111735178A (en) | Air conditioner energy-saving system based on elite meaning and SVR regression algorithm and optimization method | |
CN112734135B (en) | Power load prediction method, intelligent terminal and computer readable storage medium | |
CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
CN113762387B (en) | Multi-element load prediction method for data center station based on hybrid model prediction | |
CN111461921B (en) | Load modeling typical user database updating method based on machine learning | |
CN110163444A (en) | A kind of water demand prediction method based on GASA-SVR | |
CN115907195A (en) | Photovoltaic power generation power prediction method, system, electronic device and medium | |
CN111428766B (en) | Power consumption mode classification method for high-dimensional mass measurement data | |
CN116629428A (en) | Building energy consumption prediction method based on feature selection and SSA-BiLSTM | |
CN113743538A (en) | Intelligent building energy consumption prediction method, equipment and medium based on IPSO-BP neural network | |
CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
CN115879602A (en) | Ultra-short-term photovoltaic output prediction method based on transient weather | |
CN114444760A (en) | Industry user electric quantity prediction method based on mode extraction and error adjustment | |
CN114707422A (en) | Intelligent power check method based on load prediction | |
CN112836876B (en) | Power distribution network line load prediction method based on deep learning | |
CN111914900B (en) | User electricity utilization mode classification method | |
CN117390550A (en) | Low-carbon park carbon emission dynamic prediction method and system considering emission training set | |
CN117034762A (en) | Composite model lithium battery life prediction method based on multi-algorithm weighted sum | |
CN116167465A (en) | Solar irradiance prediction method based on multivariate time series ensemble learning | |
CN114091782B (en) | Medium-long term power load prediction method | |
CN114234392B (en) | Air conditioner load fine prediction method based on improved PSO-LSTM | |
CN109359671A (en) | A kind of classification intelligent extract method of Hydropower Station Reservoir Dispatching rule | |
CN112712213A (en) | Method and system for predicting energy consumption of deep migration learning of centralized air-conditioning house |
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 |