CN107506865A - A kind of load forecasting method and system based on LSSVM optimizations - Google Patents
A kind of load forecasting method and system based on LSSVM optimizations Download PDFInfo
- Publication number
- CN107506865A CN107506865A CN201710767106.8A CN201710767106A CN107506865A CN 107506865 A CN107506865 A CN 107506865A CN 201710767106 A CN201710767106 A CN 201710767106A CN 107506865 A CN107506865 A CN 107506865A
- Authority
- CN
- China
- Prior art keywords
- data
- particle
- value
- load
- particles
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of load forecasting method and system based on LSSVM optimizations, including：(1) after original historical load data is obtained, screen and correct abnormal data therein, construction feature vector, k mean clusters are carried out to characteristic vector, according to Clustering Effect, select input variable；(2) with the penalty factor of LSSVM models and kernel function width parameter σ, the particle as particle cluster algorithm is in the position coordinates of search space, output of the particle coordinate vector value [C, σ] for selecting to cause fitness value minimum as particle cluster algorithm；(3), using [C, the σ] obtained in step (2), load prediction regression equation is solved as the input and output of LSSVM models by the use of the input variable data after optimization processing, load prediction is carried out using regression equation.The present invention have modified abnormal data, find optimum penalty factor and kernel function width parameter, so as to improve the load prediction precision based on LSSVM.
Description
Technical Field
The invention belongs to the field of load prediction, and particularly relates to a load prediction method and system based on LSSVM optimization.
Background
In the existing load prediction technology, the LSSVM has the advantages of high calculation speed, high prediction precision and good popularization performance, and is very suitable for load prediction application scenes with nonlinearity, high dimensionality and small samples. By predicting the building load in real time and comparing the predicted value with the actual load value, the abnormal use of the building energy consumption and the equipment fault can be found in time, and remedial measures are taken to avoid possible loss; and analyzing the energy-saving state and energy-saving potential of the building, adjusting an energy-saving strategy and providing a decision basis for reasonably distributing energy.
When load prediction is performed based on the LSSVM technology, the quality of sample data and parameters of a model mainly have great influence on a prediction result. The key for improving the load prediction accuracy and speed of the LSSVM is to improve the quality of data and select a proper penalty factor and a proper kernel function width parameter. In the aspect of data quality, the traditional transverse comparison method utilizes the property that two continuous points in a load curve cannot generate mutation, if the difference between loads at two moments before and after exceeds a certain threshold value, the point is considered to have an abnormal point, and the average value of the two moments before and after is taken for correction, so that the traditional transverse comparison method cannot sufficiently correct the existing abnormal data due to the existence of continuous abnormal values in the load data; the training sample set cannot fully reflect the internal characteristics of the load; the model structure is complicated due to excessive dimensionality of input variables and correlation among the variables, the effect of reducing the dimension can be achieved by performing attribute reduction on the influence factors of the load, but information carried by other factors is abandoned, and the correlation problem among the variables cannot be solved.
In the aspect of parameter selection, the main disadvantage of the LSSVM load prediction model is that the selection of the penalty term parameter C and the kernel function width parameter has no certain basis, and only depends on the experience of a predictor or an experimental method. The method for selecting parameters based on experience has strong randomness, and the calculation amount is large based on experimental methods such as grid search, cross validation and the like, so that the method is time-consuming. The model parameters of the LSSVM can be optimized based on the standard PSO particle swarm algorithm, but in the process of searching the optimal parameters, the particles are easy to approach to local extreme values, so that the premature convergence problem is caused.
In summary, when load prediction is performed based on the LSSVM technology in the prior art, the real-time load prediction based on the LSSVM is difficult due to the low quality of model data samples and the difficulty in selecting the penalty parameter C and the kernel function width parameter σ.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to solve the technical problems that when the load prediction is carried out based on the LSSVM technology in the prior art, the quality of a model data sample is not high, and the selection of a punishment parameter C and a kernel function width parameter sigma is difficult, so that the real-time load prediction based on the LSSVM is difficult.
In order to achieve the above object, in one aspect, the present invention provides a load prediction method based on LSSVM optimization, including the following steps:
(1) After the original historical load data are obtained, processing the data by using a bidirectional comparison method, and screening and correcting abnormal data in the data; then constructing a characteristic vector by using the temperature, date type and load data, carrying out k-means clustering on the characteristic vector, and selecting data of a date with higher characteristic similarity with the date data to be predicted as an input variable of a training sample according to a clustering effect; and carrying out principal component analysis and dimension reduction processing on the input variable to remove redundant information.
(2) Taking a penalty factor C and a kernel function width parameter sigma of the LSSVM model as position coordinates of particles of a particle swarm algorithm in a search space, inputting a position vector [ C, sigma ] of each particle in each iteration process into the LSSVM model, and calculating errors of a predicted value and an actual value of a training set as fitness; selecting particles with good fitness as parent particles in an iteration process, performing cross operation, and then performing variation on position variables of the particles; the particle coordinate vector value [ C, σ ] that minimizes the fitness value is selected as an output of the particle swarm algorithm.
(3) And (2) using the input variable data after the optimization processing in the step (1) as the input and the output of the LSSVM model, using the coordinate value C of [ C, sigma ] obtained in the step (2) as a penalty factor of the LSSVM model, using the coordinate value sigma as a kernel function width parameter of the LSSVM model, solving a load prediction regression equation, and using the regression equation to predict the load.
Optionally, in step (2), the input variable data after being optimized in step (1) is used as the input and output of the training set in step (2).
Optionally, the LSSVM load prediction model completes the step (1) by using a Correct-Cluster-PCA algorithm optimization model, and the step (1) specifically includes the following steps:
step 1-1, for the original n days history load data, starting from the 2 nd time point of the D day, calculating the average value of the load variation D (D, t) = L (D, t) -L (D, t-1) of the t time point relative to the t-1 time point and the n days load variationJudging whether the daily load variation exceeds a threshold value, namely D (D, t)&And if yes, jumping to the step 1-2, otherwise jumping to the step 1-4, wherein lambda is a correction factor constant, d is more than or equal to 1 and less than or equal to n, t is more than or equal to 2 and less than or equal to infinity, and L (d, t) is the load capacity at the t moment on the day d.
Step 1-2, calculating the transverse threshold value of the bidirectional comparison method If the load data value to be determined satisfiesAnd (4) judging the load data to be judged as abnormal data, jumping to the step 1-3, otherwise jumping to the step 1-4.
And 1-3, screening abnormal data and counting the number of the abnormal data.
And 1-4, counting the data judged to be normal, and calculating the average value of the normal data to be used as the correction value of the abnormal data.
And 1-5, correcting abnormal data.
And 1-6, judging whether the data correction of the day d is finished, if so, jumping to the step 1-6, otherwise, jumping to the step 1-1 at t + 1.
And 1-7, judging whether the data of n days are corrected, if so, ending the Correct part of the algorithm, otherwise, d +1, and jumping to the step 1-1.
And 1-8, forming similar day feature vectors by using the load data, the daily average temperature data and the date type data after the correction in the steps 1-7, and carrying out k-means clustering on the similar day feature vectors of n days to complete the Cluster part of the algorithm.
And 1-9, selecting data of the same date as the date to be predicted as a training sample according to the clustering result in the step 1-8, and performing dimensionality reduction on the input vector by utilizing a PCA (principal component analysis) algorithm, namely a principal component analysis method, so as to complete the PCA part of the algorithm.
Optionally, the LSSVM model finds the most suitable penalty factor parameter C and kernel function width parameter σ of the model by using an Improved-PSO-LSSVM parameter optimization algorithm, and step (2) specifically includes the following steps:
step 2-1, initializing the particle swarm scale of a particle swarm PSO algorithm, the search range of the parameter C and the search range of the parameter sigma, the maximum iteration number maxgen, an inertia factor parameter w, acceleration parameters C1 and C2, a speed upper limit parameter vmax, a cross probability pc and a variation probability pm.
Step 2-2, initializing the positions of the particles in the search ranges of C and sigma, and taking the current position of each particle as the optimal position P of the individual history id 。
Step 2-3, initializing position vectors [ C, sigma ] of the particles in the step 2-2]The coordinate values are respectively used as a penalty factor C and a kernel function width parameter sigma of the LSSVM model, data obtained in the steps 1-9 and processed by the Correct-Cluster-PCA algorithm are used as input, historical load data are used as output, and the average absolute error between the predicted load value and the actual predicted value is calculatedThe difference is taken as a fitness value; finding out the position with the minimum fitness value from the individual historical optimal positions of each particle as the group optimal position P gd 。
Step 2-4, the speed updating formula of the particles in the (n + 1) th iteration is as follows: the location update formula is The positions of the particles are updated according to a velocity update formula and a position update formula, respectively, wherein,is the velocity of the particles at the nth iteration,is the position of the particle at the nth iteration, w is the inertia factor constant of the particle at the nth iteration,for the individual optimal position of the particle at the nth iteration,for the optimal position of the particle group in the nth iteration, c1 and c2 are acceleration constant factors, and r1 and r2 are value ranges of [0, 1%]The random number of (2).
And 2-5, sorting the particles according to the size of the fitness value.
And 2-6, selecting particles with good fitness to form a particle pool, selecting the particles in the particle pool as parent particles according to the probability pc, and performing cross operation, wherein the speed updating formula of the child particles is as follows:
the location update formula is:wherein p is a value in the range of [0,1 ]]Random number, parent 1 (V i )、parent 1 (X i ) Parent is the velocity and position of parent particle 1 2 (V i )、parent 2 (X i ) Respectively, the velocity and position of the parent particle 2, child 1 (V i )、child 1 (X i ) For the velocity and position of the daughter particle 1, child 2 (V i )、child 2 (X i ) For the velocity and position of the daughter particle 2, i is the number of iterations of the microparticle.
Step 2-7, performing variation operation on the child particles obtained in the step 2-6 according to the probability pm, wherein the variation formula is X i ′＝X i + rand × η, where rand is a value in the range of [0,1]Eta is equal to X i Homologized random vectors, X, following a (0, 1) standard normal distribution i ' position of particle in space after performing mutation operation, X i The position of the particle in space before performing the mutation operation.
Step 2-8, calculating the fitness value of the selected, crossed and mutated particles, and updating the individual optimal position P id And the population optimal position P gd 。
And 2-9, judging whether the maximum iteration number is reached or whether the optimal position adaptability value searched by the group so far is smaller than a preset threshold value, if so, ending the algorithm, and if not, returning to the step 2-4 to continue executing.
In another aspect, the present invention provides a load prediction system based on LSSVM optimization, including:
the training sample processing unit is used for processing data by using a bidirectional comparison method after acquiring original historical load data, and screening and correcting abnormal data in the data; then constructing a characteristic vector by using the temperature, the date type and the load data, carrying out k-means clustering on the characteristic vector, and selecting data of a date with higher characteristic similarity with the date data to be predicted as an input variable of a training sample according to a clustering effect; and (4) carrying out principal component analysis and dimension reduction processing on the input variable to remove redundant information.
The model parameter determining unit is used for taking a penalty factor C and a kernel function width parameter sigma of the LSSVM model as position coordinates of particles of a particle swarm algorithm in a search space, inputting a position vector [ C, sigma ] of each particle in each iteration process into the LSSVM model, and calculating errors of a predicted value and an actual value of a training set to serve as fitness; selecting particles with good fitness as parent particles in an iteration process, performing cross operation, and then performing variation on position variables of the particles; the particle coordinate vector value [ C, σ ] that minimizes the fitness value is selected as an output of the particle swarm algorithm.
And the load prediction unit is used for solving a load prediction regression equation and performing load prediction by using the regression equation by using the input variable data subjected to the optimization processing by the training sample processing unit as the input and the output of the LSSVM model, using the coordinate value C of [ C, sigma ] obtained by the model parameter determination unit as a penalty factor of the LSSVM model and the coordinate value sigma as a kernel function width parameter of the LSSVM model.
Optionally, the input variable data after being optimized by the training sample processing unit is used as the input and output of the training set in the model parameter determination unit.
Optionally, the training sample processing unit is configured to perform the following steps:
step 1-1, for the original n-day history load data, starting from the 2 nd time point of the D-day, calculating the average value of the load variation D (D, t) = L (D, t) -L (D, t-1) of the t time point relative to the t-1 time point and the load variation of the n-dayDetermining whether the daily load variation exceeds a threshold, i.e., determining D (D, t)&And if yes, jumping to the step 1-2, otherwise jumping to the step 1-4, wherein lambda is a correction factor constant, d is more than or equal to 1 and less than or equal to n, t is more than or equal to 2 and less than or equal to infinity, and L (d, t) is the load capacity at the t moment on the day d.
Step 1-2, calculating the transverse threshold value of the bidirectional comparison method If the load data value to be determined satisfiesAnd (4) judging that the load data to be judged is abnormal data, and jumping to the step 1-3, otherwise jumping to the step 1-4.
And 1-3, screening abnormal data and counting the number of the abnormal data.
And 1-4, counting the data judged to be normal, and calculating the average value of the normal data to be used as the correction value of the abnormal data.
And 1-5, correcting abnormal data.
And 1-6, judging whether the data correction of the day d is finished, if so, jumping to the step 1-6, otherwise, jumping to the step 1-1 by t + 1.
And 1-7, judging whether the data of n days are corrected, if so, ending the Correct part of the algorithm, otherwise, d +1, and jumping to the step 1-1.
And 1-8, forming similar day characteristic vectors by using the load data, the daily average temperature data and the date type data after the correction in the steps 1-7, and carrying out k-means clustering on the similar day characteristic vectors for n days to complete the Cluster part of the algorithm.
And 1-9, selecting data of the same date as the date to be predicted as a training sample according to the clustering result in the step 1-8, and performing dimensionality reduction on the input vector by utilizing a PCA (principal component analysis) algorithm, namely a principal component analysis method, so as to complete the PCA part of the algorithm.
Optionally, the model parameter determining unit is configured to perform the following steps:
step 2-1, initializing the particle swarm size of a particle swarm PSO algorithm, the search range of a parameter C and the search range of a parameter sigma, the maximum iteration number maxgen, an inertia factor parameter w, acceleration parameters C1 and C2, a speed upper limit parameter vmax, a cross probability pc and a variation probability pm.
Step 2-2, initializing the positions of the particles in the search ranges of C and sigma, and taking the current position of each particle as the optimal position P of the individual history id 。
Step 2-3, initializing the position vector [ C, sigma ] of the particles in the step 2-2]The coordinate values are respectively used as a penalty factor C and a kernel function width parameter sigma of the LSSVM model, data obtained in the step 1-9 and processed by the Correct-Cluster-PCA algorithm are used as input, historical load data are used as output, and the average absolute error between the predicted load value and the actual predicted value is calculated and used as a fitness value; finding out the position with the minimum fitness value from the individual historical optimal positions of each particle as a group optimal position P gd 。
Step 2-4, the speed updating formula of the particles in the (n + 1) th iteration is as follows: the location update formula is The positions of the particles are updated according to a velocity update formula and a position update formula, respectively, wherein,is the velocity of the particle at the nth iteration,is the position of the particle at the nth iteration, w is the inertia factor constant of the particle at the nth iteration,for the individual optimal position of the particle at the nth iteration,for the optimal position of the particle group in the nth iteration, c1 and c2 are acceleration constant factors, and r1 and r2 are value ranges of [0, 1%]The random number of (2).
And 2-5, sorting the particles according to the size of the fitness value.
And 2-6, selecting particles with good fitness to form a particle pool, selecting the particles in the particle pool as parent particles according to the probability pc, and performing cross operation, wherein the speed updating formula of the child particles is as follows:
the location update formula is:wherein p is a value in the range of [0,1 ]]Random number of (2), parent 1 (V i )、parent 1 (X i ) Parent is the velocity and position of parent particle 1 2 (V i )、parent 2 (X i ) Respectively the velocity and position of the parent particle 2, child 1 (V i )、child 1 (X i ) For the velocity and position of the daughter particle 1, child 2 (V i )、child 2 (X i ) For the velocity and position of the daughter particle 2, i is the number of particle iterations.
Step 2-7, performing mutation operation on the filial generation particles obtained in the step 2-6 according to the probability pm, wherein the mutation formula is X i ′＝X i + rand × η, where rand is a value in the range of [0,1]Eta is equal to X i Homodimensionally random vectors, X, following a (0, 1) standard normal distribution i ' position of particle in space after performing mutation operation, X i The position of the particle in space before performing the mutation operation.
2-8, calculating the fitness value of the selected, crossed and mutated particles, and updating the individual optimal position P id And the group optimal position P gd 。
And 2-9, judging whether the maximum iteration number is reached or whether the optimal position adaptability value searched by the group so far is smaller than a preset threshold value, if so, ending the algorithm, and if not, returning to the step 2-4 to continue executing.
Generally, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:
according to the invention, after collected historical load data is processed by a Correct-Cluster-PCA algorithm, abnormal data in the historical load data is corrected, data of a prediction day with high similarity to a day to be predicted is selected as a training sample, simultaneously, dimension reduction and simplification are carried out on input variables, the quality of input and output data of an LSSVM model is greatly Improved, and meanwhile, an optimum punishment factor and a kernel function width parameter are found by utilizing an Improved-PSO-LSSVM algorithm, so that the load prediction precision based on the LSSVM is Improved.
Drawings
FIG. 1 is a schematic flow chart of a load prediction method based on LSSVM optimization according to the present invention;
FIG. 2 is a flow chart of a Correct-Cluster-PCA algorithm of the data processing module provided by the invention;
FIG. 3 is a flow chart of an Improved-PSO-LSSVM parameter optimization algorithm of the parameter optimization module provided by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the respective embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Fig. 1 is a schematic flow chart of a load prediction method based on LSSVM optimization in the system provided in the present invention. The method comprises the following steps:
step 1, after acquiring original historical load data, processing the data by using a bidirectional comparison method, and screening and correcting abnormal data in the data; then constructing a characteristic vector by using the temperature, date type and load data, carrying out k-means clustering on the characteristic vector, and selecting data of a date with higher similarity with the date data to be predicted as an input variable of a training sample according to a clustering effect; and performing principal component analysis and dimension reduction processing on the input variable, extracting main information in the input variable, and removing redundant information.
Step 2, taking a penalty factor C and a kernel function width parameter sigma of the LSSVM model as position coordinates of particles in a search space, inputting a position vector [ C, sigma ] of each particle in each iteration process into the LSSVM model, and calculating errors of a predicted value and an actual value of a training set as fitness; selecting particles with better fitness as parent particles in an iteration process, performing cross operation, and then performing variation on position variables of the particles; selecting a particle coordinate vector value [ C, sigma ] which enables the fitness value to be minimum as the output of the particle swarm optimization;
and 3, using the data optimized in the step 1 as input and output of the LSSVM model, using the coordinate value C of [ C, sigma ] obtained in the step 2 as a penalty factor of the LSSVM model, using the coordinate value sigma as a kernel function width parameter of the model, solving a load prediction regression equation, and using the regression equation to predict the load.
FIG. 2 shows a flow chart of the Correct-Cluster-PCA algorithm for optimizing data quality in the implementation of the invention. The process comprises the following steps:
step 10: inputting historical load data of n days to be processed.
Step 11: calculating the average value of the load change quantity D (D, t) = L (D, t) -L (D, t-1) of the time point t relative to the time point t-1 and the load change quantity of n days from the time point 2 of the D day
Step 12: and judging whether the load variation per day exceeds a threshold value, namely judging whether D (D, t) > lambda alpha (t) is true, if so, jumping to the step 13, and otherwise, jumping to the step 15.
Wherein, steps 10 to 12 correspond to step 1-1.
Step 13: calculating lateral thresholds for two-way comparison If it satisfiesJudging the data to be abnormal data, jumping to a step 14, otherwise, jumping to a step 15.
Wherein step 13 corresponds to step 1-2.
Step 14: and screening abnormal data, and counting the number of the abnormal data.
Step 15: and counting the data judged to be normal, and calculating the average value of the normal data to be used as the correction value of the abnormal data.
Step 16: the abnormal data screened out in step 14 is corrected.
And step 17: and (4) judging whether the data correction of the day d is finished, if so, jumping to the step 18, otherwise, jumping to the step 10 at t + 1.
Wherein, the steps 14 to 17 correspond to the steps 1-3 to 1-6, respectively.
Step 18: and judging whether the data of n days are corrected, if so, jumping to the step 19, otherwise, jumping to the step 10 by the step d + 1.
Step 19: outputting the corrected data, and ending the Correct part of the algorithm;
wherein steps 18 to 19 correspond to steps 1 to 7.
Step 20: and (5) forming similar day characteristic vectors by using the load data, the daily average temperature data and the date type data after the correction in the step (19), and carrying out k-means clustering on the similar day characteristic vectors of n days to complete the Cluster part of the algorithm.
Step 21: and (4) selecting data of the same date as the date to be predicted as a training sample according to the clustering result in the step (20), and performing dimensionality reduction on the input vector by utilizing a PCA algorithm to complete the PCA part of the algorithm.
Wherein, the steps 20 to 21 correspond to the steps 1 to 8 to 1 to 9, respectively.
FIG. 3 shows an Improved-PSO-LSSVM algorithm flow of the parameter optimization module of the present invention, which comprises the following steps:
and step 30: initializing particle swarm scale sizepop of particle swarm PSO algorithm and searching range [ Cmin, cmax ] of parameter C]Parameter sigma search range [ sigma min, sigma max]Maximum iteration number maxgen, inertia factor parameter w, acceleration parameters c1 and c2, speed upper limit parameter vmax and cross probability p c And the probability of variation p m ；
Step 31: initializing the positions of the particles in the search ranges of C and sigma, and taking the current position of each particle as the optimal position P of the individual history id An initial value of (1);
step 32: initializing the position vector [ C, sigma ] of the particle in step 31]The coordinate values of the two-dimensional space vector machine (LSSVM) model are respectively used as a penalty factor C and a kernel function width parameter sigma of the LSSVM model, data obtained in the step 15 after being processed by the Correct-Cluster-PCA algorithm are used as input, historical load data are used as output, and the average absolute error between the predicted load value and the actual predicted value is calculated and used as a fitness value; finding adaptations in individual historical optimal locations for each particleThe position with the minimum value is taken as the optimal position P of the group gd 。
Step 33: the velocity update formula for the particle at n +1 iterations is: the location update formula is And updating the positions of the particles according to the speed updating formula and the position updating formula respectively.
Step 34: sorting the particles according to the size of the fitness value;
step 35: selecting particles with better fitness to form a particle pool;
step 36: according to a certain probability p c Selecting the particles in the particle pool as parent particles, and performing cross operation, wherein the speed updating formula of the child particles is as follows:
the location update formula is:
step 37: according to a certain probability p m Performing mutation operation on the filial generation particles obtained in the step 3-6, wherein the mutation formula is X i ′＝X i +rand×η。
Step 38: and judging whether the maximum iteration number is reached or whether the optimal position adaptability value searched by the group so far is smaller than a preset threshold value, if so, jumping to the step 39, and if not, returning to the step 33 to continue execution.
Step 39: and outputting an optimal solution [ C, sigma ], wherein the position coordinates are respectively an optimal penalty factor C and a kernel function width parameter sigma selected by the LSSVM model, and finishing the algorithm.
The correspondence between steps 30 to 39 and steps 2-1 to 2-9 will not be described in detail.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. A load prediction method based on LSSVM optimization is characterized by comprising the following steps:
(1) After the original historical load data are obtained, processing the data by using a bidirectional comparison method, and screening and correcting abnormal data in the data; then constructing a characteristic vector by using the temperature, the date type and the load data, carrying out k-means clustering on the characteristic vector, and selecting data of a date with higher characteristic similarity with the date data to be predicted as an input variable of a training sample according to a clustering effect; performing principal component analysis and dimension reduction processing on the input variable to remove redundant information;
(2) Taking a penalty factor C and a kernel function width parameter sigma of the LSSVM model as position coordinates of particles of a particle swarm algorithm in a search space, inputting a position vector [ C, sigma ] of each particle in each iteration process into the LSSVM model, and calculating errors of a predicted value and an actual value of a training set to serve as fitness; selecting particles with good fitness as parent particles in an iteration process, performing cross operation, and then performing variation on position variables of the particles; selecting the particle coordinate vector value [ C, sigma ] which enables the fitness value to be minimum as the output of the particle swarm optimization;
(3) And (2) using the input variable data after the optimization processing in the step (1) as the input and the output of the LSSVM model, using the coordinate value C of [ C, sigma ] obtained in the step (2) as a penalty factor of the LSSVM model, using the coordinate value sigma as a kernel function width parameter of the LSSVM model, solving a load prediction regression equation, and using the regression equation to predict the load.
2. The load prediction method according to claim 1, wherein in step (2), the input variable data after the optimization processing in step (1) is used as the input and output of the training set in step (2).
3. The load prediction method according to claim 1 or 2, characterized in that the LSSVM load prediction model performs the step (1) by using a Correct-Cluster-PCA algorithm optimization model, wherein the step (1) specifically comprises the following steps:
step 1-1, for the original n-day history load data, starting from the 2 nd time point of the D-day, calculating the average value of the load variation D (D, t) = L (D, t) -L (D, t-1) of the t time point relative to the t-1 time point and the load variation of the n-dayJudging whether the daily load variation exceeds a threshold value, namely D (D, t)&If so, jumping to the step 1-2, otherwise jumping to the step 1-4, wherein lambda is a correction factor constant, d is more than or equal to 1 and less than or equal to n, t is more than or equal to 2 and less than or equal to infinity, and L (d, t) is the load capacity at the t moment on the d day;
step 1-2, calculating the transverse threshold value of the bidirectional comparison method i =1, 2.. N, if the load data value to be determined satisfiesJudging the load data to be judged as abnormal data, jumping to the step 1-3, otherwise jumping to the step 1-4;
step 1-3, screening abnormal data, and counting the number of the abnormal data;
step 1-4, counting the data judged to be normal, and calculating the average value of the normal data to be used as the correction value of the abnormal data;
step 1-5, correcting abnormal data;
step 1-6, judging whether data correction on the day d is finished, if so, jumping to the step 1-6, and if not, jumping to the step 1-1 by t + 1;
step 1-7, judging whether the data of n days are corrected, if so, ending the Correct part of the algorithm, otherwise, d +1, and jumping to the step 1-1;
step 1-8, forming similar day feature vectors by using the load data, the daily average temperature data and the date type data after the correction in the step 1-7, and carrying out k-means clustering on the similar day feature vectors of n days to complete the Cluster part of the algorithm;
and 1-9, selecting data of the same date as the date to be predicted as a training sample according to the clustering result in the step 1-8, and performing dimensionality reduction on the input vector by utilizing a PCA (principal component analysis) algorithm, namely a principal component analysis method, so as to complete the PCA part of the algorithm.
4. The load prediction method according to claim 3, wherein the LSSVM model uses an Improved-PSO-LSSVM parameter optimization algorithm to find the optimum penalty factor parameter C and kernel function width parameter σ of the model, and step (2) comprises the following steps:
step 2-1, initializing the particle swarm size of a particle swarm PSO algorithm, the search range of a parameter C and the search range of a parameter sigma, the maximum iteration number maxgen, an inertia factor parameter w, acceleration parameters C1 and C2, a speed upper limit parameter vmax, a cross probability pc and a variation probability pm;
step 2-2, initializing the positions of the particles in the search ranges of C and sigma, and taking the current position of each particle as the optimal position P of the individual history id ；
Step 2-3, initializing the position vector [ C, sigma ] of the particles in the step 2-2]The coordinate values of the two-dimensional space-time domain parameter are respectively used as a penalty factor C and a kernel function width parameter sigma of the LSSVM model, and the correction-Taking data processed by the Cluster-PCA algorithm as input, taking historical load data as output, and calculating an average absolute error between a predicted load value and an actual predicted value as a fitness value; finding out the position with the minimum fitness value from the individual historical optimal positions of each particle as the group optimal position P gd ；
Step 2-4, the speed updating formula of the particles in the (n + 1) th iteration is as follows: the location update formula is The positions of the particles are updated according to a velocity update formula and a position update formula, respectively, wherein,is the velocity of the particle at the nth iteration,is the position of the particle at the nth iteration, w is the inertia factor constant of the particle at the nth iteration,for the individual optimal position of the particle at the nth iteration,the optimal position of the particle population in the nth iteration is represented by c1 and c2 which are acceleration constant factors, and r1And r2 is a value range of [0,1 ]]The random number of (2);
2-5, sorting the particles according to the size of the fitness value;
and 2-6, selecting particles with good fitness to form a particle pool, selecting the particles in the particle pool as parent particles according to the probability pc, and performing cross operation, wherein the speed updating formula of the child particles is as follows:
the location update formula is:wherein p is a value in the range of [0,1 ]]Random number of (2), parent 1 (V i )、parent 1 (X i ) Parent is the velocity and position of parent particle 1 2 (V i )、parent 2 (X i ) Respectively the velocity and position of the parent particle 2, child 1 (V i )、child 1 (X i ) For the velocity and position of the daughter particle 1, child 2 (V i )、child 2 (X i ) The speed and the position of the filial generation particle 2 are shown in the formula, wherein i is the iteration number of the particle;
step 2-7, performing mutation operation on the filial generation particles obtained in the step 2-6 according to the probability pm, wherein the mutation formula is X i ′＝X i + rand × η, where rand is a value in the range of [0,1]Eta is equal to X i Homologized random vectors, X, following a (0, 1) standard normal distribution i ' position of particle in space after performing mutation operation, X i The position of the particle in space before performing the mutation operation;
step 2-8, calculating the fitness value of the selected, crossed and mutated particles, and updating the individual optimal position P id And the group optimal position P gd ；
And 2-9, judging whether the maximum iteration number is reached or whether the optimal position adaptability value searched by the group so far is smaller than a preset threshold value, if so, ending the algorithm, and if not, returning to the step 2-4 to continue executing.
5. A load prediction system based on LSSVM optimization is characterized by comprising:
the training sample processing unit is used for processing data by using a bidirectional comparison method after acquiring original historical load data, and screening and correcting abnormal data in the data; then constructing a characteristic vector by using the temperature, date type and load data, carrying out k-means clustering on the characteristic vector, and selecting data of a date with higher characteristic similarity with the date data to be predicted as an input variable of a training sample according to a clustering effect; performing principal component analysis and dimension reduction processing on the input variable to remove redundant information;
the model parameter determining unit is used for taking a penalty factor C and a kernel function width parameter sigma of the LSSVM model as position coordinates of particles of a particle swarm algorithm in a search space, inputting a position vector [ C, sigma ] of each particle in each iteration process into the LSSVM model, and calculating errors of a predicted value and an actual value of a training set to serve as fitness; selecting particles with good fitness as parent particles in an iteration process, performing cross operation, and then performing variation on position variables of the particles; selecting a particle coordinate vector value [ C, sigma ] which enables the fitness value to be minimum as an output of the particle swarm optimization;
and the load prediction unit is used for solving a load prediction regression equation and performing load prediction by using the regression equation by using the input variable data subjected to the optimization processing by the training sample processing unit as the input and the output of the LSSVM model, using the coordinate value C of [ C, sigma ] obtained by the model parameter determination unit as a penalty factor of the LSSVM model and the coordinate value sigma as a kernel function width parameter of the LSSVM model.
6. The load prediction system of claim 5, wherein the input variable data optimized by the training sample processing unit is used as input and output of the training set in the model parameter determination unit.
7. The load prediction system according to claim 5 or 6, wherein the training sample processing unit is configured to perform the following steps:
step 1-1, for the original n-day history load data, starting from the 2 nd time point of the D-day, calculating the average value of the load variation D (D, t) = L (D, t) -L (D, t-1) of the t time point relative to the t-1 time point and the load variation of the n-dayJudging whether the daily load variation exceeds a threshold value, namely D (D, t)&If yes, jumping to the step 1-2, otherwise jumping to the step 1-4, wherein lambda is a correction factor constant, d is more than or equal to 1 and less than or equal to n, t is more than or equal to 2 and less than or equal to infinity, and L (d, t) is the load capacity at the t moment on the day d;
step 1-2, calculating transverse threshold value of bidirectional comparison method i =1,2, \ 8230;, n, if the load data value to be determined satisfiesJudging the load data to be judged as abnormal data, jumping to the step 1-3, otherwise jumping to the step 1-4;
step 1-3, screening abnormal data, and counting the number of the abnormal data;
step 1-4, counting the data judged to be normal, and calculating the average value of the normal data to be used as the correction value of the abnormal data;
step 1-5, correcting abnormal data;
step 1-6, judging whether data correction on the day d is finished, if so, jumping to step 1-6, otherwise, jumping to step 1-1 by t + 1;
step 1-7, judging whether the data of n days are corrected, if so, ending the Correct part of the algorithm, otherwise, d +1, and jumping to the step 1-1;
step 1-8, forming similar day characteristic vectors by using the load data, the daily average temperature data and the date type data after the correction in the step 1-7, and carrying out k-means clustering on the similar day characteristic vectors for n days to complete a Cluster part of an algorithm;
and 1-9, selecting data of the same date as the date to be predicted as a training sample according to the clustering result in the step 1-8, and performing dimensionality reduction on the input vector by utilizing a PCA (principal component analysis) algorithm, namely a principal component analysis method, so as to complete the PCA part of the algorithm.
8. The load prediction system of claim 7, wherein the model parameter determination unit is configured to perform the steps of:
step 2-1, initializing a particle swarm scale of a particle swarm PSO algorithm, a parameter C search range, a parameter sigma search range, a maximum iteration number maxgen, an inertia factor parameter w, acceleration parameters C1 and C2, a speed upper limit parameter vmax, a cross probability pc and a variation probability pm;
step 2-2, initializing the positions of the particles in the search ranges of C and sigma, and taking the current position of each particle as the optimal position P of the individual history id ；
Step 2-3, initializing the position vector [ C, sigma ] of the particles in the step 2-2]The coordinate values are respectively used as a penalty factor C and a kernel function width parameter sigma of the LSSVM model, data obtained in the step 1-9 and processed by the Correct-Cluster-PCA algorithm are used as input, historical load data are used as output, and the average absolute error between the predicted load value and the actual predicted value is calculated and used as a fitness value; finding out the position with the minimum fitness value from the individual historical optimal positions of each particle as a group optimal position P gd ；
Step 2-4, the updating formula of the particle speed in the (n + 1) th iteration is as follows: the location update formula is The positions of the particles are updated according to a velocity update formula and a position update formula, respectively, wherein,is the velocity of the particle at the nth iteration,is the position of the particle at the nth iteration, w is the inertia factor constant of the particle at the nth iteration,for the individual optimal position of the particle at the nth iteration,for the optimal position of the particle population in the nth iteration, c1 and c2 are acceleration constant factors, and r1 and r2 are in the value range of [0, 1%]The random number of (2);
2-5, sorting the particles according to the size of the fitness value;
and 2-6, selecting particles with good fitness to form a particle pool, selecting the particles in the particle pool as parent particles according to the probability pc, and performing cross operation, wherein the speed updating formula of the child particles is as follows:
the location update formula is:wherein p is a value in the range of [0,1 ]]Random number, parent 1 (V i )、parent 1 (X i ) Parent is the velocity and position of parent particle 1 2 (V i )、parent 2 (X i ) Respectively, the velocity and position of the parent particle 2, child 1 (V i )、child 1 (X i ) For the velocity and position of the daughter particle 1, child 2 (V i )、child 2 (X i ) The speed and position of the particle 2 are shown, wherein i is the number of particle iterations;
step 2-7, performing variation operation on the child particles obtained in the step 2-6 according to the probability pm, wherein the variation formula is X i ′＝X i + rand × η, where rand is a value in the range of [0,1]Eta is equal to X i Homologized random vectors, X, following a (0, 1) standard normal distribution i ' position of particle in space after performing mutation operation, X i The position of the particle in space before performing the mutation operation;
step 2-8, calculating the fitness value of the selected, crossed and mutated particles, and updating the individual optimal position P id And the group optimal position P gd ；
And 2-9, judging whether the maximum iteration number is reached or whether the optimal position adaptability value searched by the group so far is smaller than a preset threshold value, if so, ending the algorithm, and if not, returning to the step 2-4 to continue executing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710767106.8A CN107506865B (en) | 2017-08-30 | 2017-08-30 | Load prediction method and system based on LSSVM optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710767106.8A CN107506865B (en) | 2017-08-30 | 2017-08-30 | Load prediction method and system based on LSSVM optimization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107506865A true CN107506865A (en) | 2017-12-22 |
| CN107506865B CN107506865B (en) | 2020-09-08 |
Family
ID=60694376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710767106.8A Expired - Fee Related CN107506865B (en) | 2017-08-30 | 2017-08-30 | Load prediction method and system based on LSSVM optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107506865B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034476A (en) * | 2018-07-27 | 2018-12-18 | 中南大学 | A kind of line of high-speed railway extreme wind speeds big data Forecast method |
| CN110674947A (en) * | 2019-09-02 | 2020-01-10 | 三峡大学 | Spectral feature variable selection and optimization method based on Stacking integrated framework |
| CN110837857A (en) * | 2019-10-31 | 2020-02-25 | 杭州中恒云能源互联网技术有限公司 | Industrial electricity load prediction method, system and storage medium thereof |
| WO2020140560A1 (en) * | 2019-05-31 | 2020-07-09 | 南京航空航天大学 | Malfunction early-warning method for production logistics delivery equipment |
| CN111598255A (en) * | 2020-04-27 | 2020-08-28 | 同济大学 | Method for compensating cogging effect of suspension gap sensor of maglev train |
| CN112417363A (en) * | 2020-11-11 | 2021-02-26 | 深圳供电局有限公司 | Load analysis method and system for transformer substation |
| CN112541296A (en) * | 2020-07-22 | 2021-03-23 | 华北电力大学（保定） | SO2 prediction method based on PSO-LSSVM |
| CN112686571A (en) * | 2021-01-12 | 2021-04-20 | 山东电力工程咨询院有限公司 | Comprehensive intelligent energy optimization scheduling method and system based on dynamic adaptive modeling |
| CN113177597A (en) * | 2021-04-30 | 2021-07-27 | 平安国际融资租赁有限公司 | Model training data determination method, detection model training method, device and equipment |
| CN113516291A (en) * | 2021-05-24 | 2021-10-19 | 国网河北省电力有限公司经济技术研究院 | Power load prediction method, device and equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160203419A1 (en) * | 2013-03-09 | 2016-07-14 | Bigwood Technology, Inc. | Metaheuristic-guided trust-tech methods for global unconstrained optimization |
| CN106503788A (en) * | 2016-10-28 | 2017-03-15 | 中国矿业大学 | Least square method supporting vector machine Forecasting Methodology based on self adaptation population |
| CN106529124A (en) * | 2016-10-14 | 2017-03-22 | 云南电网有限责任公司昆明供电局 | A transformer insulation state assessment method based on principal component analysis and support vector machines |
-
2017
- 2017-08-30 CN CN201710767106.8A patent/CN107506865B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160203419A1 (en) * | 2013-03-09 | 2016-07-14 | Bigwood Technology, Inc. | Metaheuristic-guided trust-tech methods for global unconstrained optimization |
| CN106529124A (en) * | 2016-10-14 | 2017-03-22 | 云南电网有限责任公司昆明供电局 | A transformer insulation state assessment method based on principal component analysis and support vector machines |
| CN106503788A (en) * | 2016-10-28 | 2017-03-15 | 中国矿业大学 | Least square method supporting vector machine Forecasting Methodology based on self adaptation population |
Non-Patent Citations (3)
| Title |
|---|
| 王惠中 等: "电力系统短期负荷预测建模仿真研究", 《计算机仿真》 * |
| 陈璐璐 等: "改进的遗传粒子群混合优化算法", 《计算机工程与设计》 * |
| 龚文龙: "基于最小二乘支持向量机的短期负荷预测", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034476B (en) * | 2018-07-27 | 2020-11-17 | 中南大学 | High-speed railway line extreme wind speed big data clustering prediction method |
| CN109034476A (en) * | 2018-07-27 | 2018-12-18 | 中南大学 | A kind of line of high-speed railway extreme wind speeds big data Forecast method |
| US11740619B2 (en) | 2019-05-31 | 2023-08-29 | Nanjing University Of Aeronautics And Astronautics | Malfunction early-warning method for production logistics delivery equipment |
| WO2020140560A1 (en) * | 2019-05-31 | 2020-07-09 | 南京航空航天大学 | Malfunction early-warning method for production logistics delivery equipment |
| CN110674947B (en) * | 2019-09-02 | 2021-02-19 | 三峡大学 | Spectral feature variable selection and optimization method based on Stacking integrated framework |
| CN110674947A (en) * | 2019-09-02 | 2020-01-10 | 三峡大学 | Spectral feature variable selection and optimization method based on Stacking integrated framework |
| CN110837857A (en) * | 2019-10-31 | 2020-02-25 | 杭州中恒云能源互联网技术有限公司 | Industrial electricity load prediction method, system and storage medium thereof |
| CN111598255A (en) * | 2020-04-27 | 2020-08-28 | 同济大学 | Method for compensating cogging effect of suspension gap sensor of maglev train |
| CN112541296A (en) * | 2020-07-22 | 2021-03-23 | 华北电力大学（保定） | SO2 prediction method based on PSO-LSSVM |
| CN112417363A (en) * | 2020-11-11 | 2021-02-26 | 深圳供电局有限公司 | Load analysis method and system for transformer substation |
| CN112417363B (en) * | 2020-11-11 | 2022-06-24 | 深圳供电局有限公司 | Load analysis method and system for transformer substation |
| CN112686571A (en) * | 2021-01-12 | 2021-04-20 | 山东电力工程咨询院有限公司 | Comprehensive intelligent energy optimization scheduling method and system based on dynamic adaptive modeling |
| CN112686571B (en) * | 2021-01-12 | 2023-01-20 | 山东电力工程咨询院有限公司 | Comprehensive intelligent energy optimization scheduling method and system based on dynamic adaptive modeling |
| CN113177597A (en) * | 2021-04-30 | 2021-07-27 | 平安国际融资租赁有限公司 | Model training data determination method, detection model training method, device and equipment |
| CN113516291A (en) * | 2021-05-24 | 2021-10-19 | 国网河北省电力有限公司经济技术研究院 | Power load prediction method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107506865B (en) | 2020-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107506865B (en) | Load prediction method and system based on LSSVM optimization | |
| CN111738512B (en) | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model | |
| CN109388565B (en) | Software system performance optimization method based on generating type countermeasure network | |
| CN114373101A (en) | Image classification method for neural network architecture search based on evolution strategy | |
| CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
| CN112540849B (en) | Parameter configuration optimization method and system for distributed computing operation | |
| CN115578248B (en) | Generalized enhanced image classification algorithm based on style guidance | |
| CN112733997A (en) | Hydrological time series prediction optimization method based on WOA-LSTM-MC | |
| CN108984830A (en) | A kind of building efficiency evaluation method and device based on FUZZY NETWORK analysis | |
| CN112491891B (en) | Network attack detection method based on hybrid deep learning in Internet of things environment | |
| CN111832839B (en) | Energy consumption prediction method based on sufficient incremental learning | |
| CN113743538A (en) | Intelligent building energy consumption prediction method, equipment and medium based on IPSO-BP neural network | |
| CN110987436A (en) | Bearing fault diagnosis method based on excitation mechanism | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN113935557A (en) | Same-mode energy consumption big data prediction method based on deep learning | |
| CN116542701A (en) | Carbon price prediction method and system based on CNN-LSTM combination model | |
| CN113762591A (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy | |
| CN109492816B (en) | Coal and gas outburst dynamic prediction method based on hybrid intelligence | |
| Bi et al. | Self-adaptive Teaching-learning-based Optimizer with Improved RBF and Sparse Autoencoder for Complex Optimization Problems | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN111984514A (en) | Prophet-bLSTM-DTW-based log anomaly detection method | |
| CN115619028A (en) | Clustering algorithm fusion-based power load accurate prediction method | |
| CN106960260B (en) | Wind power prediction system convenient for power dispatching | |
| CN112749211B (en) | Novel tea yield prediction method based on electric power big data | |
| CN113188243B (en) | Comprehensive prediction method and system for air conditioner energy consumption |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200908 Termination date: 20210830 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |