CN111144649B - Urban gas daily load combined prediction method based on information fusion - Google Patents
Urban gas daily load combined prediction method based on information fusion Download PDFInfo
- Publication number
- CN111144649B CN111144649B CN201911360951.9A CN201911360951A CN111144649B CN 111144649 B CN111144649 B CN 111144649B CN 201911360951 A CN201911360951 A CN 201911360951A CN 111144649 B CN111144649 B CN 111144649B
- Authority
- CN
- China
- Prior art keywords
- prediction
- grnn
- data
- sheep
- predictor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 230000004927 fusion Effects 0.000 title claims abstract description 14
- 241001494479 Pecora Species 0.000 claims abstract description 49
- 230000035772 mutation Effects 0.000 claims abstract description 46
- 238000013528 artificial neural network Methods 0.000 claims abstract description 44
- 230000002159 abnormal effect Effects 0.000 claims abstract description 32
- 244000144992 flock Species 0.000 claims abstract description 14
- 238000009304 pastoral farming Methods 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 27
- 238000007781 pre-processing Methods 0.000 claims description 17
- 238000010606 normalization Methods 0.000 claims description 16
- 244000144980 herd Species 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000003993 interaction Effects 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 5
- 238000003825 pressing Methods 0.000 claims description 5
- 238000009827 uniform distribution Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 abstract description 7
- 238000001514 detection method Methods 0.000 abstract description 7
- 239000007789 gas Substances 0.000 description 96
- 230000008859 change Effects 0.000 description 17
- 238000005516 engineering process Methods 0.000 description 14
- 230000008569 process Effects 0.000 description 9
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 8
- 238000009499 grossing Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 239000003345 natural gas Substances 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 3
- 241000282414 Homo sapiens Species 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000001427 coherent effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 235000013311 vegetables Nutrition 0.000 description 2
- 235000019687 Lamb Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004380 ashing Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000002087 whitening effect 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Public Health (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the technical field of data prediction, and particularly relates to an urban gas daily load combined prediction method based on information fusion. The invention adopts a neural network predictor GRNN, a gray-GRNN and a gradient-GRNN for prediction, then designs a random set combined predictor for combined prediction, and optimizes parameters of the random set combined predictor based on a parameter optimizer of an improved sheep flock intelligent algorithm. The neural network predictor of the invention can not only remove the possible bad data, but also judge the single model fault. The invention can effectively eliminate the randomness and uncertainty of random variables, improve the prediction precision, and particularly can carry out mutation detection based on a mutation theory through an abnormal data detector when some prediction models have faults.
Description
Technical Field
The invention belongs to the technical field of data prediction, and particularly relates to an urban gas daily load combined prediction method based on information fusion.
Background
The urban gas daily load prediction is mainly used for predicting future use amount by collecting past historical data and data related to the influence on daily load, key factors such as economy and environment need to be comprehensively considered, and if the prediction method is mature and can reach certain precision, the prediction method plays a vital role in decision of a relevant department of an urban area. Because the daily load of the gas is related to various aspects such as climate, population, economy, living habits, local customs and the like, the daily load of the gas has random and multivariate complexity and even mutability.
At present, the combined prediction model is widely applied at home and abroad. The main research methods in the gas load prediction field of China include a multi-resolution wavelet network, a support vector machine, an exponential smoothing method and the like, but the methods have respective defects in the practical process. Therefore, the combined prediction method has high expectations in China, and can make up for deficiencies and effectively improve the gas load prediction accuracy.
Disclosure of Invention
The invention aims to provide an urban gas daily load combination prediction method based on information fusion.
The purpose of the invention is realized by the following technical scheme: the method comprises the following steps:
step 1: acquiring historical data including date, population, temperature and daily load of urban gas to form a training set, and preprocessing the data in the training set;
step 2: inputting the training set into a neural network predictor, wherein the input of the neural network predictor is the population and temperature corresponding to a certain date, and the output is a predicted value of the daily load of the urban gas; the neural network predictor comprises GRNN, gray-GRNN and gradient-GRNN, wherein the prediction result of GRNN is y 1 The prediction of gray-GRNN is y 2 The predicted result of gradient-GRNN is y 3 ；
And 3, step 3: inputting the prediction result of the neural network predictor into the random set combined predictor, and outputting the random set combined predictor
Comprises the following steps:
wherein, a j Calculating by a parameter optimizer, and adopting an improved intelligent algorithm of the sheep flock to the parameter a by the parameter optimizer j Optimization, performance index function extraction
y is the true value, e is the prediction error; optimization parameter x = (a) 1 ,a 2 ,a 3 ) Min (J) is obtained by minimizing J 1 ,J 2 ,…J M ) The sheep is set as a head sheep, and the method comprises the following specific steps:
step 3.1: initializing parameters;
setting population M, maximum iteration time T, reordered iteration time, threshold epsilon and maximum grazing probability omega max Minimum grazing probability ω min Solving an upper limit ub of a space and a lower limit lb of the space; initial population x 0 ＝(a 1 0 ,a 2 0 ,a 3 0 ) Pressing [0,1]Uniform distribution generation, performance index function extraction
Wherein y is the true value and e is the prediction error; />
Step 3.2: adjusting the position of each sheep, if moving, the performance
If not, abandoning the update; the formula for the adjustment is:
in the formula: g k Indicating the position of the first sheep of the kth generation; rand is a random number from 0 to 1;
a position after moving for the i (i =1,2, … M-1) th individual;
step 3.3: the ith sheep
Randomly selecting another sheep>
Interact if >>
In a position which is better than +>
Then->
To/>
Close to, or are present>
Searching around the user in small steps; />
The moving formula of (2) is:
after the interaction between two sheep, comparing with the value before the interaction, if moving, the performance
If not, abandoning the update; in the formula: />
For the perturbation operator, the constant c is [0,2]Adjustable parameters over an interval;
step 3.4: judging whether the local optimum is trapped or not; if the difference value between the first generation sheep and the previous generation sheep is smaller than the threshold epsilon, judging that the first generation sheep is trapped into local optimum; if the local optimum is trapped, executing the step 3.5; if the trapping is locally optimal, executing the step 3.6;
step 3.5: executing a shepherd dog supervision mechanism, and judging whether the current individual meets a grazing condition, wherein the grazing condition is that q is less than p, q is a random number of [0,1], and p is a resetting probability;
in the formula, omega max And ω min Respectively the maximum grazing probability and the minimum grazing probability;
if the grazing condition is met, grazing the current sheep, and updating the position according to the following formula;
if the herd condition is not met, randomly selecting a herd sheep
The position is adjusted according to the following formula, and the performance is greater or less after the movement>
If not, abandoning the update;
step 3.6: judging whether the maximum iteration number is reached; if the maximum iteration times are reached, outputting the current optimal solution; if the maximum iteration times are not reached, executing the step 3.7;
step 3.7: judging whether the number of reordered iterations is reached; if the number of the iterations of the reordering is reached, reordering, determining the position of the flocks, and returning to the step 3.2; if the reordering iteration times are not reached, returning to the step 3.2;
and 4, step 4: inputting the output results of the neural network predictor and the random set combined predictor into an abnormal data detector; if the abnormal data detector detects abnormal data, judging the data is abnormal, immediately judging whether the system is a system with mutation based on a mutation theory, selecting a control variable and a mutation model, and returning to the step 2;
and 5: and inputting the population and the temperature corresponding to the date to be predicted into the trained random set combined predictor to obtain the daily load prediction result of the urban gas on the day.
The present invention may further comprise:
the method for preprocessing the data in the training set in the step 1 specifically comprises the following steps:
the data preprocessing has the main function of preventing the increase of training time caused by the fluctuation of abnormal data, which may cause the convergence failure of a prediction algorithm in serious cases and needs to perform normalization technical processing on the original data; normalizing the parameters to [0,1] in the training set by adopting the following formula;
wherein X max Is the maximum value, X, in the training set min Is the minimum value in the training set, X i Represents the normalized data; the formula for denormalization is:
X＝X min +(X max -X min )X i
and when the data are output, the numerical value obtained by conversion of the inverse normalization formula is the predicted value.
The invention has the beneficial effects that:
the invention adopts a neural network predictor GRNN, a gray-GRNN and a gradient-GRNN to predict, then designs a random set combined predictor to carry out combined prediction, and optimizes the parameters of the random set combined predictor based on a parameter optimizer of an improved sheep flock intelligent algorithm. The neural network predictor of the invention can not only remove the possible bad data, but also judge the single model fault. The invention can effectively eliminate the randomness and uncertainty of random variables, improve the prediction precision, and particularly can carry out mutation detection based on a mutation theory through an abnormal data detector when some prediction models have faults.
Drawings
FIG. 1 is a diagram of a random set combinational predictor design of the present invention.
FIG. 2 is a table of relative error of predicted data for three neural network predictors in accordance with an embodiment of the present invention.
FIG. 3 is a table of the maximum relative error and average accuracy of the annual predictions for three neural network predictors in accordance with an embodiment of the present invention.
FIG. 4 is a table of the predicted relative error of the combination of the real value of the daily load of gas and the random set (GRNN failed in 27 days 1 month).
FIG. 5 is a table of the combined predicted relative error of the actual daily gas load and the random set (after the fault is eliminated) in the embodiment of the present invention.
FIG. 6 is a table showing the results of the cusp type mutation model in the examples of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
A combined prediction method based on information fusion comprises the following steps:
step 1: inquiring a historical time sequence according to the type of data to be predicted to form a training set, and preprocessing the data in the training set;
step 2: inputting the training set into a neural network predictor to obtain a prediction result of the neural network predictor; the neural network predictor comprises GRNN, gray-GRNN and gradient-GRNN, wherein the prediction result of GRNN is y 1 The prediction of gray-GRNN is y 2 The prediction of gradient-GRNN is y 3 ；
And step 3: inputting the prediction result of the neural network predictor into the random set combination predictor, and outputting the prediction result
Comprises the following steps: />
Wherein, a j Calculated by a parameter optimizer, the parameter optimizer adopts an improved intelligent algorithm for the parameter a j Optimization, performance index function extraction
y is the true value, e is the prediction error; optimization parameter x = (a) 1 ,a 2 ,a 3 ) Min (J) is obtained by minimizing J 1 ,J 2 ,…J M ) The sheep is set as a first sheep, and the method comprises the following specific steps:
step 3.1: initializing parameters;
setting population M, maximum iteration time T, reordered iteration time, threshold epsilon and maximum grazing probability omega max Minimum grazing probability ω min Solving an upper limit ub of the space and a lower limit lb of the space; initial population x 0 ＝(a 1 0 ,a 2 0 ,a 3 0 ) Pressing [0,1]Uniform distribution generation, performance index function extraction
Wherein y is the true value and e is the prediction error;
step 3.2: adjusting the position of each sheep, if moving, the performance
If not, abandoning the update; the formula for the adjustment is:
in the formula: g k Indicating the position of the kth generation of the sheep; rand is a random number from 0 to 1;
a position after moving for the i (i =1,2, … M-1) th individual;
step 3.3: the ith sheep
Randomly selecting another sheep>
Interact if->
Is located at a position better than->
Then->
To/>
Approach in>
Searching around the user in small steps; />
The movement formula of (c) is:
after and before the interaction between two sheepThe values are compared and the performance is determined if the movement is in progress
If not, abandoning the update; in the formula: />
For the perturbation operator, the constant c is [0,2 ]]Adjustable parameters over an interval;
step 3.4: judging whether the local optimum is trapped or not; if the difference value between the first generation sheep and the previous generation sheep is smaller than the threshold epsilon, judging that the first generation sheep is trapped into local optimum; if the local optimum is trapped, executing the step 3.5; if the trapping is locally optimal, executing the step 3.6;
step 3.5: executing a shepherd dog supervision mechanism, and judging whether the current individual meets a grazing condition, wherein the grazing condition is that q is less than p, q is a random number of [0,1], and p is a resetting probability;
in the formula, omega max And ω min The maximum grazing probability and the minimum grazing probability are respectively;
if the grazing condition is met, grazing the current sheep, and updating the position according to the following formula;
if the herd condition is not met, randomly selecting a herd sheep
The position is adjusted according to the following formula, and the performance is greater or less after the movement>
If not, abandoning the update; />
Step 3.6: judging whether the maximum iteration times is reached; if the maximum iteration times are reached, outputting the current optimal solution; if the maximum iteration times are not reached, executing the step 3.7;
step 3.7: judging whether the number of reordered iterations is reached; if the number of the iterations of the reordering is reached, reordering, determining the position of the flocks, and returning to the step 3.2; if the reordering iteration times are not reached, returning to the step 3.2;
and 4, step 4: inputting the output results of the neural network predictor and the random set combined predictor into an abnormal data detector; if the abnormal data detector detects abnormal data, judging the data is abnormal, immediately judging whether the system is a system with mutation based on a mutation theory, selecting a control variable and a mutation model, and returning to the step 2;
and 5: and inputting the date to be predicted into the trained random set combined predictor to obtain a prediction result.
Example 1:
the embodiment provides an intelligent combined prediction experimental technique for daily urban gas load based on a combined prediction method based on information fusion.
Abnormal data is detected through data preprocessing (including processing based on experience, experimental observation or mapping method and data normalization and inverse normalization technology) and a neural network (intelligent) predictor and a random set combined predictor, and mutation detection is carried out on daily gas load values based on a mutation theory. Firstly, a generalized regression neural network predictor (GRNN), a gray-GRNN and a gradient-GRNN are respectively adopted to predict daily gas load, then a random set combined predictor is designed to carry out combined prediction, and a parameter optimizer based on an improved intelligent lamb swarm algorithm optimizes parameters of the random set combined predictor. The neural network predictor of the embodiment can remove possible bad data and judge single model faults. The random set is used as a combined prediction method based on information fusion, randomness and uncertainty of random variables can be effectively eliminated, and prediction accuracy of daily gas load is improved. Particularly, when some prediction models are in failure, the random set technology shows the advantages. The method of the invention can also be applied to the technical field of combined prediction such as power load prediction.
Natural gas is one of the most green and efficient high-quality energy sources in the world at present and is widely used by people of all countries in the world. With the rapid development of global economy and the continuous deterioration of human living environment, the demand of human beings for natural gas is rapidly increased. Therefore, it is particularly important to predict the urban gas load.
The urban gas daily load prediction is mainly used for predicting future use amount by collecting past historical data and data related to the influence on daily load, key factors such as economy and environment need to be comprehensively considered, and if the prediction method is mature and can reach certain precision, the prediction method plays a vital role in decision of a relevant department of an urban area. Because the daily load of the gas is related to various aspects such as climate, population, economy, living habits, local customs and the like, the daily load of the gas has random and multivariate complexity and even mutability.
At present, the combined prediction model is widely applied at home and abroad. The main research methods in the gas load prediction field of China include a multi-resolution wavelet network, a support vector machine, an exponential smoothing method and the like, but the methods have respective defects in the practical process. Therefore, the combined prediction method has high expectations in China, and can make up for deficiencies and effectively improve the gas load prediction accuracy.
The invention aims to provide an effective urban gas daily load combination prediction experiment technical method aiming at randomness, mutation and the like of the urban gas daily load.
The method mainly carries out short-term prediction on the daily load of the urban gas. Because the factors such as temperature and population change in every 2 months in a small range all the year round. Therefore, the historical data of the daily gas load of a certain year in a certain city are divided into 6 groups, the daily gas load is subjected to combined prediction by utilizing a random set, and the combined prediction parameters are optimized by adopting an improved intelligent sheep flock algorithm. The method comprises the following specific steps:
(1) And (3) predicting the daily gas load by respectively adopting a generalized regression neural network predictor (GRNN), a gray-GRNN and a gradient-GRNN to obtain a relative error and prediction accuracy, and then performing combined prediction by taking the 3 prediction data as a data basis to improve the daily gas load prediction accuracy.
(2) The random set is adopted to carry out combined prediction on the daily load of the gas, and the consumption of the urban gas shows randomness to a certain extent. The prediction model data for the day GRNN, gray GRNN, and gradient GRNN were used for combined prediction using the properties of the random set.
(3) Optimizing the combined prediction parameters by adopting an improved flocks intelligent algorithm, finally processing mutation phenomena in the prediction process by adopting a mutation theory, establishing a cusp type mutation model aiming at the influence factors of population and temperature to obtain a bifurcation point set equation, and predicting the time point at which mutation is likely to occur in the future by using a mutation flow equation and a bifurcation set.
In order to achieve the purpose, the invention designs an intelligent combined predictor for daily load of urban gas, which comprises: abnormal data detector, generalized regression neural network predictor (GRNN), gray-GRNN, gradient-GRNN, random set composite predictor.
A parameter optimizer, etc. Specifically, the method comprises the following steps:
1. the abnormal data detector is used for detecting the following conditions: the abnormal data is detected by data preprocessing (technology), a neural network predictor (including GRNN, gray-GRNN and gradient-GRNN) and a random set combined predictor, and then whether the system is a system with mutation or not can be judged based on a mutation theory, namely, the daily load value of the gas is subjected to mutation detection based on the mutation theory.
GRNN, gray-GRNN, and gradient-GRNN are collectively referred to as neural network (intelligent) predictors. And carrying out experimental research on the neural network predictor, and judging and predicting the model fault through analysis.
And (3) the outputs of the GRNN, the gray-GRNN and the gradient-GRNN are sent to a random set combined predictor, the output of the random set combined predictor is used as a gas daily load predicted value, a prediction error is used as the input of a parameter optimizer (by utilizing an improved intelligent algorithm of the sheep flock), and the output (optimization parameter) of the parameter optimizer is sent to the random set combined predictor.
The invention has the following beneficial effects:
1. the method comprises the steps of preprocessing the historical measurement data of the daily load of the urban gas, namely removing possible bad data to ensure that the data are correct, and processing the data based on experience and experimental observation (or mapping method) and the normalization and inverse normalization technology of the original data.
2. The neural network predictors used (including GRNN, grey-GRNN and gradient-GRNN) can both remove potentially bad data and discriminate faults in the single model (GRNN or grey-GRNN or gradient-GRNN).
3. Factors influencing daily gas load such as climate, population, economy, living habits, local customs and the like are considered. The historical data of the gas daily load of a certain year in a certain city are divided into 6 groups, a prediction model is established by taking every 2 months as a group, the factors such as climate, population, economy and the like do not change greatly in a short period (2 months), the input node of the neural network predictor obtains the gas daily load data of three days before the predicted day and the daily load data of one day before the week, the output node obtains the predicted daily gas load, and the day factors (living habits, local customs) influencing the gas daily load can be comprehensively reflected
4. The random set is used as a combined prediction method based on information fusion, randomness and uncertainty of random variables can be effectively eliminated, and prediction accuracy of daily gas load is improved. Especially, when some single prediction models fail, the random set technology shows the advantages.
5. Parameters in the random set combined predictor are optimized by using an improved intelligent algorithm of the flocks of sheep, so that better prediction time and prediction precision can be obtained.
6. A system with mutation can be judged based on a mutation theory, and the time point of possible mutation of the daily load of the gas is predicted by establishing a cusp type mutation model.
The technical solution of the present invention is further explained with reference to fig. 1. The invention relates to an intelligent combined prediction experiment technical method for daily load of urban gas, which comprises the following steps:
the method comprises the following steps: arranging and collecting relevant historical data, which mainly comprises population, temperature, total daily load (value) of urban gas and the like; and data preprocessing is carried out, wherein the data preprocessing comprises parameter initialization (parameter initialization of a neural network predictor, parameter initialization of a parameter optimizer, selection of a mutation model control variable and a potential function and the like) and removal of possibly existing bad data, so that the data are correct and coherent, and the change rule of the gas load (value) in the data can be correctly reflected. And removing possible bad data, namely based on experience and through experimental observation (or a mapping method), and carrying out normalization technical processing on the original data according to an equation (3) and carrying out inverse normalization according to an equation (4) to obtain a predicted value of the daily load of the gas.
Step two: abnormal data and model faults are judged by the neural network predictor through experiments
The neural network predictors include generalized regression neural network predictors (GRNN), gray-GRNN, and gradient-GRNN. The technology is derived from documents Chen Fang, high-rise, vegetable market daily price prediction based on a generalized recurrent neural network [ J ]. Zhejiang agriculture academic newspaper 2015,25 (7): 1253-1258 ], chen Hongli, liu Wenyan, niu Yi Ning, application of gradient RBF in urban gas load short-term prediction [ J ]. Instrument and meter newspaper 2009,30 (6 supple supplement): 443-444 ].
Step three: random set combinatorial predictor design
The prediction results for GRNN, gray-GRNN, and gradient-GRNN are input as the random set combination predictor. Random set combined predictor outputs
Comprises the following steps:
wherein the prediction results of GRNN, gray-GRNN and gradient-GRNN are y 1 ，y 2 ，y D 。
Step four: design of parameter optimizer
The parameter optimizer uses an improved intelligent algorithm of the sheep flock to correct the parameter a in the formula (6) j And (6) optimizing.
Step five: anomaly data detector design
The abnormal data detector of the embodiment can detect abnormal data through data preprocessing (technology), a neural network predictor (including GRNN, gray-GRNN and gradient-GRNN) and a random set combination predictor, for example, the prediction precision of the GRNN, gray-GRNN, gradient-GRNN and random set combination prediction is obviously abnormal (very large), so that the abnormal data can be judged, whether the system is a system with mutation or not can be judged on the basis of a mutation theory, reasonable control variables in the system can be selected, and a mutation model can be selected at the same time. Considering the specific situation of the embodiment, the control variables select temperature and population, and a cusp type mutation model is adopted to carry out mutation detection on the daily load value of the gas.
The steps involved in the process of the present invention are described in detail below with reference to FIG. 1.
The main experimental work of urban gas daily load (value) prediction is to arrange historical real measurement data, mainly including population, temperature and other data, remove error data therein, and establish a reasonable prediction model, so that an accurate gas (daily) load prediction value can be obtained. Therefore, two conditions are required to predict the daily gas load. First, the authenticity of the historical data. Second, the feasibility of the model is predicted. Specifically, the method comprises the following steps:
(1) Relevant historical data are measured, sorted and collected, and the data mainly comprise population, temperature, daily (total) load capacity (value) of urban gas and the like.
(2) The data preprocessing is to remove the possible bad data, so that the data are correctly coherent and the change rule of the gas load (value) can be correctly reflected.
(3) And selecting a proper technology to establish a corresponding gas daily load prediction model.
(4) And (3) using the built gas prediction model to predict the daily load of the gas, and analyzing and correcting the prediction result.
The first step is as follows: and (4) sorting and collecting related historical data. Because the factors such as temperature and population change in every 2 months in a small range all the year round. Therefore, the historical data of the daily gas load of a certain city in a certain year are divided into 6 groups, a prediction model is established by taking every 2 months as one group, and data preprocessing is carried out.
The data preprocessing has the main function of preventing the increase of training time caused by the fluctuation of abnormal data, which may cause the convergence failure of the prediction algorithm in serious cases and needs to perform normalization technical processing on the original data. The parameters were normalized to [0,1] in the training sample using equation (1).
In the output layer, the value obtained by conversion using the formula (2) is the predicted value of the daily gas load.
X＝X min +(X max -X min )X i (2)
Wherein, X max Is the maximum value of the training sample, X min Is the minimum value of the training sample, X i And (4) expressing the normalized data, wherein X represents a predicted gas daily load value obtained after reverse normalization.
In order to avoid the interference of unstable data on the network training, a margin can be set before normalization, the normalized value falls into [0.1,0.9], and the normalization formula is deduced again, because ln0.1= -2.3026, ln0.9= -0.1054, so that the normalization formula is enabled to be carried out again
A new normalization formula can be derived:
the denormalization formula is:
where exp represents an exponential function based on a natural constant e and ln represents a logarithm based on e, the data can be normalized to [0.1,0.9] according to equations (3) and (4).
The second step is that: and (4) utilizing a generalized regression neural network predictor (GRNN), a gray-GRNN and a gradient-GRNN to predict the daily load of the urban gas.
The generalized regression neural network predictor (GRNN) technology is derived from the document Chen Fang, with high context, vegetable market daily price prediction [ J ]. Zhejiang agriculture academic newspaper 2015,25 (7): 1253-1258 ], and the working processes of gray-GRNN and gradient-GRNN are similar to GRNN, except that the technical processes of ashing, gradient and whitening are carried out on data input and output, and the specific method is derived from the documents Chen Hongli, liu Wenyan and Niu Yi Ning, and the application of gradient RBF in urban gas load short-term prediction [ J ]. Instrument academic newspaper 2009,30 (6 suppl) 443-444.
The specific technical process comprises the following steps: training and modeling are carried out on the gas daily load historical data, and 4 input nodes of the neural network are selected (gas daily load data of three days before the forecast day and daily load data of one day before the week); and 1 output node is taken, namely the daily gas load is predicted. The listed gas daily load prediction (relative) errors from 1 month to 25 days to 31 days (the same applies hereinafter) of a certain city of a certain year are shown in fig. 2. The predicted relative error formula is as follows:
in GRNN, the smoothing coefficient σ =0.12. The GRNN gas daily load prediction data has very large errors at 27, 29 and 30 months, possibly due to sudden changes in gas daily load data or failure of the prediction model. The prediction accuracy can be improved by performing verification through other prediction models or performing combined prediction. Fig. 3 shows GRNN prediction data (accuracy = 1-relative error) with 1 month, 27 days, 29 days, and 30 days culled throughout the year.
The smoothing coefficient of gray-GRNN was σ =0.35, and the prediction accuracy of gray-GRNN on day 27 was 8.133%, indicating that the GRNN model failed in the prediction on day 27. The smoothing coefficient σ =0.27 for the gradient-GRNN, and the prediction accuracy of the gradient-GRNN at 27 days has no larger error like that of the GRNN, which also indicates that the use of the GRNN fails at 27 days, so that the GRNN has a very large error in the prediction at 27 days.
Prediction data for 29 and 30 months of 1 year were eliminated and prediction data for gray-GRNN and gradient-GRNN were shown in FIG. 3 throughout the year.
The third step: random set combined predictor
The random set is a random element whose value is set, is the popularization of the concept of random variables in probability theory, and is actually the set of elements and the number of the elements which are random variables. The random set has good application in the fields of target tracking and the like.
The embodiment tries to provide a new solution for the daily gas load prediction problem by taking the random set as a combined prediction technical method based on information fusion. In the urban gas daily load prediction process, the gas daily load and the prediction (value) thereof have strong randomness and are random variables, and in the combined prediction based on information fusion, the failure of some prediction models is considered (a single prediction model may fail, and the failed prediction model is eliminated, for example, GRNN fails in 27 days of 1 month), that is, the number of the prediction models is a random variable, and the random variables have randomness and uncertainty. And a random set is adopted to effectively eliminate the randomness and uncertainty of random variables and improve the prediction precision of daily gas load.
Generalized regression neural network predictor (GRNN), gray-GRNNAnd the prediction results of the gradient-GRNN are input as a random set combination predictor. Let the generalized regression neural network predictor (GRNN), the grey-GRNN and gradient-GRNN prediction results be y 1 ，y 2 ，y 3 . Then the random set combined predictor output
A combination of its inputs, namely:
wherein the parameter a j Obtained by utilizing a parameter optimizer based on an improved herd of sheep algorithm,
the predictors are combined for the random set.
The combined prediction of the daily gas load is directly carried out through a random set, for example, very large errors occur in 27 days in 1 month, 29 days and 30 days in a certain year in a certain city, and the method is shown in a figure 4. Compared with prediction results of GRNN, gray-GRNN and gradient-GRNN, the method can judge that the daily gas load data are possible to generate mutation in two days of 1 month, 29 days and 30 days, and uses an abnormal data detector to judge based on a mutation theory so as to predict the time point at which mutation is possible. But the prediction error is larger at 27 days 1, mainly because the relative error of GRNN prediction data in the training samples is larger at 27 days (GRNN fails at 27 days 1).
The nature of the random set is utilized to remove the prediction data of GRNN in the training samples of 27 days in 1 month, the random set technology is utilized again to carry out the combined prediction of the daily gas load, the relative error of the daily gas load prediction is obviously improved, and after the faults are removed, the true value and the relative prediction error of the daily gas load are shown in figure 5.
As can be seen from the prediction results, the random set technology shows outstanding advantages when a single prediction model fails in the prediction process of a certain day.
The random set is used for predicting the daily gas load by using a combined prediction technology. And after the fault model of 1 month and 27 days is removed, information fusion is carried out, data which are possibly mutated in 1 month and 29 days and 30 days are removed, the maximum prediction (relative) error of random set combined prediction is 7.236% and the average (prediction) precision is 94.565% in the whole year range. The predicted average usage time was 0.432 seconds and the maximum predicted time was 0.897 seconds.
The prediction accuracy of the random set combined prediction (technology) is higher than that of GRNN, gray-GRNN and gradient-GRNN, and the advantages of the random set combined prediction are more obvious when the GRNN, gray-GRNN and gradient-GRNN fail.
The fourth step: parameter optimizing device
Parameter optimizer using improved intelligent algorithm for sheep flock to parameter a j Optimizing, comprising the following specific steps:
1. and initializing parameters. The population number M =5, the maximum iteration number T =20, the dimension D =3-n (n is the number of fault models, in this embodiment, n =1 when a GRNN model fails in 1 month and 27 days), and the threshold value epsilon =10 -4 Constant c is [0,2]Adjustable parameter, omega, over a range max ＝0.5，ω min =0.01, upper limit ub =1 and lower limit lb =0 of the solution space, initial population x 0 ＝(a 1 0 ,a 2 0 ,…a D 0 ) Pressing [0,1]]Uniform distribution generation, performance index function extraction
y is the true value of the predicted daily gas load, e is the prediction error, and the optimization parameter x = (a) 1 ,a 2 ,…a D ) To minimize J, i.e. </or>
Substituting the population into J to obtain min (J) 1 ,J 2 ,…J M ) The sheep is the first sheep.
2. Adjusting the position of each sheep according to formula (7), and if the sheep move, the performance
If not better, then putAbandoning the update.
In the formula: g k Indicating the position of the kth generation of the sheep; rand is a random number from 0 to 1;
the position after the movement is the i-th (i =1,2, … M-1) individual.
3. The ith sheep
Will randomly select another sheep>
Interact if >>
At a position better than
Then->
To->
Approach in>
Moved according to the formula (8)>
Search around itself in small steps according to equation (9). After the two sheep had interacted, the comparison was made with the values before the interaction, and if the performance did not get better, the update was discarded.
In the formula:
is a perturbation operator.
4. Judging whether the shepherd dog is trapped into local optimum or not (when the difference value between the first generation sheep and the previous generation sheep is smaller than a certain threshold epsilon, the algorithm is likely to be trapped into a certain local optimum solution), if so, turning to 5, and executing a shepherd dog supervision mechanism; otherwise go to 6.
5. Executing a shepherd dog supervision mechanism, if the current individual meets the grazing condition (q < p), taking a random number of [0,1] from q, wherein p is a reset probability
In the formula, ω max And ω min The maximum grazing probability and the minimum grazing probability are respectively. The current sheep is grazed and the position is updated according to the formula (11).
If the herd condition is not met, randomly selecting a herd sheep
The position is adjusted according to the formula (12), and if the moving effect (performance) is not good, the position is not updated.
6. If the maximum iteration times are reached, turning to the 8 th step; otherwise go to 7.
7. And reordering every certain iteration times, determining the position of the flocks, and turning to 2.
8. And outputting the current optimal solution, and finishing the algorithm.
The fifth step: abnormal data detector
The abnormal data detector of the embodiment can detect abnormal data through data preprocessing (technology), a neural network predictor (including GRNN, gray-GRNN and gradient-GRNN) and a random set combined predictor, for example, if the prediction precision of the GRNN, gray-GRNN and gradient-GRNN and random set combined prediction is obviously abnormal (very large), the abnormal data can be judged, and then judgment can be carried out based on a mutation theory. If the daily gas load is suddenly changed, actually, the system is supposed to be a system with the sudden change, and then the reasonable control variable is selected, and the sudden change model is selected.
And (4) discovering when the urban gas daily load combination prediction is carried out by using a random set. For example, in the 1 month 29 day and the 1 month 30 day two days, the daily gas load may have sudden changes, which results in inaccurate prediction of the prediction model established in the non-sudden situation. The mutation is caused by many reasons, such as population, temperature factors, spring festival and other holidays, and the daily gas load value at the time point when the mutation is likely to occur cannot be predicted by normal means.
In actual life, the factors influencing daily load of gas are many, and mainly include factors such as temperature (climate), population, day of week (living habits, holidays, local customs), natural gas price (economy) and the like. In the short term, the price of natural gas and the like do not change greatly, the factors of days of the week (the input node takes the daily load data of gas three days before the predicted day and the daily load data of one day before the week, and the output node takes the predicted daily gas load) are considered in the neural network predictor, the temperature and the population are selected as the control variables of the embodiment, and two control (variable) quantities are provided, so that the research is more suitable by adopting a cusp type mutation model. And predicting the time point of possible sudden change of the daily gas load by establishing a sharp point type sudden change model.
According to the short-term change rule of daily load of urban gas, setting the potential function of cusp mutation model as
Wherein x is a real value of daily gas load, d is population control quantity, and c is temperature control quantity.
1. The parameters a, x need to be identified 0 、b、c 0 、d 0 Value of (A)
Derivation of V (x) can result in an abrupt flow pattern:
V′(x)＝a(x-x 0 ) 3 +b(c+c 0 )(x-x 0 )+d+d 0 ＝0 (14)
then, the second derivative is obtained by solving V (x):
V″(x)＝3a(x-x 0 ) 2 +b(c+c 0 )＝0 (15)
from the expressions (14) and (15), the least square method is used to identify the potential function parameter of the mutation model by taking the previous data (the data of the related population and the temperature in 1 month in a certain city) as an example, and x is calculated 0 ＝114.2、c 0 ＝24.5。a＝0.324、d 0 ＝-303.4，b＝1.131，
The joint type (14) and (15) are used for solving a bifurcation (point) set equation:
8.74(d-303.4) 2 +5.79(c+24.5) 3 ＝0 (16)
then, according to the mutation theory, when the population control variable d and the temperature control variable c satisfy the formula (17), the daily gas load value is mutated.
Δ＝8.74(d-303.4) 2 +5.79(c+24.5) 3 ＜0 (17)
The daily gas load was subjected to sudden change detection according to equation (17), and the detection results are shown in fig. 6.
As can be seen from fig. 6, Δ < 0 on two days, i.e., 29 th and 30 th, of 1 month indicates that the daily gas load value has a sudden change in the two days, and this result is compared with the historical data of the daily gas load, and it is found that the relative error between the predicted gas load value and the actual gas load value on 29 th and 30 th of 1 month is very large, and a sudden change occurs. The reason for the sudden change is that the population does not change much and the temperature becomes sub-zero lower. From the formula (17), it is found that the daily gas load may change abruptly only when the temperature is below zero.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (2)
1. An urban gas daily load combination prediction method based on information fusion is characterized by comprising the following steps:
step 1: acquiring historical data including date, population, temperature and daily load of urban gas to form a training set, and preprocessing the data in the training set;
step 2: inputting the training set into a neural network predictor, wherein the input of the neural network predictor is the population and temperature corresponding to a certain date, and the output is a predicted value of the daily load of the urban gas; the neural network predictor comprises GRNN, gray-GRNN and gradient-GRNN, wherein the prediction result of GRNN is y 1 The prediction of gray-GRNN is y 2 The predicted result of gradient-GRNN is y 3 ；
And step 3: inputting the prediction result of the neural network predictor into the random set combination predictor, and outputting the prediction result
Comprises the following steps:
wherein, a j Calculating by a parameter optimizer, and adopting an improved intelligent algorithm of the sheep flock to the parameter a by the parameter optimizer j Optimization, performance index function extraction
y is the true value, e is the prediction error; optimization parameter x = (a) 1 ,a 2 ,a 3 ) Min (J) is obtained by minimizing J 1 ,J 2 ,…J M ) The sheep is set as a first sheep, and the method comprises the following specific steps:
step 3.1: initializing parameters;
setting the population number M, the maximum iteration number T, the reordered iteration number, the threshold value epsilon and the maximum grazing probability omega max Minimum grazing probability ω min Solving an upper limit ub of a space and a lower limit lb of the space; initial population x 0 ＝(a 1 0 ,a 2 0 ,a 3 0 ) Pressing [0,1]]Uniform distribution generation, performance index function extraction
Wherein y is the true value and e is the prediction error;
step 3.2: adjusting the position of each sheep, if moving, the performance
If not, abandoning the update; the formula for the adjustment is:
in the formula: g k Indicating the position of the kth generation of the sheep; rand is a random number from 0 to 1;
a position after moving for the i (i =1,2, … M-1) th individual;
step 3.3: the ith sheep
Randomly selecting another sheep>
Interact if->
At a position better than
Then->
To/>
Close to, or are present>
Searching around the user in small steps; />
The moving formula of (2) is:
after the interaction between two sheep, comparing with the value before the interaction, if moving, the performance
If not, abandoning the update; in the formula: />
For the perturbation operator, the constant c is [0,2 ]]Adjustable parameters over intervals;
step 3.4: judging whether the local optimum is trapped or not; if the difference value between the first generation sheep and the previous generation sheep is smaller than the threshold epsilon, judging that the first generation sheep is trapped into local optimum; if the local optimum is trapped, executing the step 3.5; if the trapping is locally optimal, executing the step 3.6;
step 3.5: executing a shepherd dog supervision mechanism, and judging whether the current individual meets a grazing condition, wherein the grazing condition is that q is less than p, q is a random number of [0,1], and p is a resetting probability;
in the formula, ω max And ω min Respectively the maximum grazing probability and the minimum grazing probability;
if the grazing condition is met, grazing the current sheep, and updating the position according to the following formula;
if the herd condition is not met, randomly selecting a herd sheep
Position adjustment by pressing, if moved, performance
If not, abandoning the update;
step 3.6: judging whether the maximum iteration times is reached; if the maximum iteration times are reached, outputting the current optimal solution; if the maximum iteration times are not reached, executing the step 3.7;
step 3.7: judging whether the number of reordered iterations is reached; if the number of the iterations of the reordering is reached, reordering, determining the position of the flocks, and returning to the step 3.2; if the reordering iteration times are not reached, returning to the step 3.2;
and 4, step 4: inputting the output results of the neural network predictor and the random set combined predictor into an abnormal data detector; if the abnormal data detector detects abnormal data, judging the data is abnormal, immediately judging whether the system is a system with mutation based on a mutation theory, selecting a control variable and a mutation model, and returning to the step 2;
and 5: and inputting the population and the temperature corresponding to the date to be predicted into the trained random set combined predictor to obtain the daily load prediction result of the urban gas on the day.
2. The urban gas daily load combination prediction method based on information fusion according to claim 1, characterized in that: the method for preprocessing the data in the training set in the step 1 specifically comprises the following steps:
the data preprocessing has the main functions of preventing the increase of training time caused by the fluctuation of abnormal data, possibly causing the convergence failure of a prediction algorithm in serious conditions and needing to carry out normalization technical processing on the original data; normalizing the parameters to [0,1] in the training set by adopting the following formula;
wherein X max Is the maximum value in the training set, X min Is the minimum value, X, in the training set i Represents the normalized data; the formula for denormalization is:
X＝X min +(X max -X min )X i
and when the data is output, the numerical value obtained by conversion by using the inverse normalization formula is the predicted value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911360951.9A CN111144649B (en) | 2019-12-26 | 2019-12-26 | Urban gas daily load combined prediction method based on information fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911360951.9A CN111144649B (en) | 2019-12-26 | 2019-12-26 | Urban gas daily load combined prediction method based on information fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111144649A CN111144649A (en) | 2020-05-12 |
| CN111144649B true CN111144649B (en) | 2023-04-07 |
Family
ID=70520223
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911360951.9A Active CN111144649B (en) | 2019-12-26 | 2019-12-26 | Urban gas daily load combined prediction method based on information fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111144649B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102073785A (en) * | 2010-11-26 | 2011-05-25 | 哈尔滨工程大学 | Daily gas load combination prediction method based on generalized dynamic fuzzy neural network |
| CN107730054A (en) * | 2017-11-15 | 2018-02-23 | 西南石油大学 | A kind of Gas Load combination forecasting method based on support vector regression |
| CN109767029A (en) * | 2018-12-14 | 2019-05-17 | 华北电力大学 | Cold, heat and power triple supply system capacity collocation method and system in local energy internet |
| CN110110930A (en) * | 2019-05-08 | 2019-08-09 | 西南交通大学 | A kind of Recognition with Recurrent Neural Network Short-Term Load Forecasting Method improving whale algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180284745A1 (en) * | 2016-05-09 | 2018-10-04 | StrongForce IoT Portfolio 2016, LLC | Methods and systems for self-organization of collected data using 3rd party data from a data marketplace in an industrial internet of things environment |
-
2019
- 2019-12-26 CN CN201911360951.9A patent/CN111144649B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102073785A (en) * | 2010-11-26 | 2011-05-25 | 哈尔滨工程大学 | Daily gas load combination prediction method based on generalized dynamic fuzzy neural network |
| CN107730054A (en) * | 2017-11-15 | 2018-02-23 | 西南石油大学 | A kind of Gas Load combination forecasting method based on support vector regression |
| CN109767029A (en) * | 2018-12-14 | 2019-05-17 | 华北电力大学 | Cold, heat and power triple supply system capacity collocation method and system in local energy internet |
| CN110110930A (en) * | 2019-05-08 | 2019-08-09 | 西南交通大学 | A kind of Recognition with Recurrent Neural Network Short-Term Load Forecasting Method improving whale algorithm |
Non-Patent Citations (3)
| Title |
|---|
| "源-网-荷-储"协调控制***研究及应用;张楠 等;《2017智能电网新技术发展与应用研讨会》;310-315 * |
| WANG Han 等."Application Research of Combined Forecasting Based on Induced Ordered Weighted Averaging Operator".《Management Science and Engineering》.2014,第8卷(第3期), * |
| 城市燃气日负荷组合预测方法研究;***;《中国优秀硕士学位论文全文数据库（工程科技Ⅱ辑）》(第3期);C038-1075 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111144649A (en) | 2020-05-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109902801B (en) | Flood collective forecasting method based on variational reasoning Bayesian neural network | |
| Heydari et al. | A novel composite neural network based method for wind and solar power forecasting in microgrids | |
| CN111783953B (en) | 24-point power load value 7-day prediction method based on optimized LSTM network | |
| CN110163410B (en) | Line loss electric quantity prediction method based on neural network-time sequence | |
| CN103577694B (en) | Aquaculture water quality short-time combination forecast method on basis of multi-scale analysis | |
| CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
| CN107092582A (en) | One kind is based on the posterior exceptional value on-line checking of residual error and method for evaluating confidence | |
| CN110163429B (en) | Short-term load prediction method based on similarity day optimization screening | |
| CN110335168B (en) | Method and system for optimizing power utilization information acquisition terminal fault prediction model based on GRU | |
| CN110571792A (en) | Analysis and evaluation method and system for operation state of power grid regulation and control system | |
| CN112990500B (en) | Transformer area line loss analysis method and system based on improved weighted gray correlation analysis | |
| CN108171379A (en) | A kind of electro-load forecast method | |
| CN112734128A (en) | 7-day power load peak value prediction method based on optimized RBF | |
| Ji et al. | Urban water demand forecasting by LS-SVM with tuning based on elitist teaching-learning-based optimization | |
| CN103676649A (en) | Local self-adaptive WNN (Wavelet Neural Network) training system, device and method | |
| CN106296434B (en) | Grain yield prediction method based on PSO-LSSVM algorithm | |
| Dang et al. | A novel multivariate grey model for forecasting periodic oscillation time series | |
| CN112949207A (en) | Short-term load prediction method based on improved least square support vector machine | |
| WO2017071369A1 (en) | Method and device for predicting user unsubscription | |
| CN114169639A (en) | Enterprise power consumption prediction method | |
| CN116578551A (en) | GRU-GAN-based power grid data restoration method | |
| CN112036598A (en) | Charging pile use information prediction method based on multi-information coupling | |
| CN111311026A (en) | Runoff nonlinear prediction method considering data characteristics, model and correction | |
| CN113762591A (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy | |
| CN111144649B (en) | Urban gas daily load combined prediction method based on information fusion |
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 |