CN110807508B - Bus peak load prediction method considering complex weather influence - Google Patents
Bus peak load prediction method considering complex weather influence Download PDFInfo
- Publication number
- CN110807508B CN110807508B CN201911056853.6A CN201911056853A CN110807508B CN 110807508 B CN110807508 B CN 110807508B CN 201911056853 A CN201911056853 A CN 201911056853A CN 110807508 B CN110807508 B CN 110807508B
- Authority
- CN
- China
- Prior art keywords
- peak load
- bus
- load prediction
- value
- prediction
- 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 47
- 239000002245 particle Substances 0.000 claims abstract description 45
- 230000000694 effects Effects 0.000 claims description 16
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims description 11
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000005312 nonlinear dynamic Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 abstract description 11
- 230000006870 function Effects 0.000 description 13
- 238000012549 training Methods 0.000 description 11
- 230000004913 activation Effects 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Evolutionary Computation (AREA)
- Entrepreneurship & Innovation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A bus peak load prediction method considering complex weather influence belongs to the technical field of bus peak load prediction. According to the invention, the feature importance results of the condition mutual information on the features to be selected in the original feature set are used as a basis, the IPSO-ELM is combined as a predictor, forward feature selection is carried out, the optimal feature set of the bus peak load prediction is determined, the influence of feature redundancy on the prediction precision in the bus peak load prediction is reduced, the optimal prediction model is respectively constructed for different buses, the prediction precision of different buses is effectively improved, an improved particle swarm optimization extreme learning machine is introduced to be combined with a linear method, the peak load prediction under different scenes is carried out, and the prediction needs under small-sample or sample-free scenes are met.
Description
Technical Field
The invention belongs to the technical field of bus peak load prediction, and particularly relates to a bus peak load prediction method considering complex weather influences.
Background
The problems of limited bus peak load historical data, severe fluctuation, nonlinearity and randomness, low prediction precision and difficult prediction become the problems to be solved urgently, and how to improve the bus peak load prediction precision becomes the problem to be solved urgently. The bus peak load prediction is an important basis for guaranteeing reliable and stable operation of the power system, and the method for improving the bus peak load prediction accuracy by analysis and research has very important significance.
At present, for bus load prediction, many researches exist, bus load prediction is optimized according to bus load characteristics, but influences of various factors such as natural weather, society and the like on the bus load are not fully analyzed, when the factors are considered, characteristic selection is not performed, differences of different bus influence factors are not considered, targeted characteristic selection is not performed on the influence factors of different bus loads, and a targeted bus load prediction model is not established on the basis. The existing peak load prediction research is developed aiming at influence factors and prediction methods of peak load of urban power grid, and although the existing research improves the peak load prediction precision to a certain extent, no targeted analysis is developed aiming at small sample problems with limited historical data in peak load prediction.
There is a need in the art for a new solution to this problem.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the bus peak load prediction method considering complex weather influences is used for solving the technical problem that in the prior art, no targeted analysis is performed on small sample problems with severe peak load fluctuation and limited historical data in peak load prediction.
The bus peak load prediction method considering complex weather influences comprises the following steps, and the following steps are sequentially carried out,
firstly, constructing an original characteristic set for predicting the bus peak load according to natural weather and social factors based on the influence of the natural weather and the social factors on the bus peak load prediction precision, and analyzing the correlation between each characteristic in the original characteristic set and the bus peak load through conditional mutual information (Conditional Mutual Information, CMI) to obtain characteristic importance ranking;
secondly, obtaining an improved particle swarm optimization extreme learning machine (Improved Particle Swarm Optimization-Extreme Learning Machine, IPSO-ELM) by improving the input weight and threshold of the particle swarm optimization extreme learning machine, then carrying out targeted forward feature selection according to the feature importance ranking result obtained in the first step by taking the prediction precision of the IPSO-ELM as a decision variable to obtain an optimal feature subset for bus peak load prediction, and retraining the IPSO-ELM according to the obtained optimal feature subset to obtain an optimal bus peak load prediction model;
substituting the test set data in the set time period in the historical data into the optimal bus peak load prediction model obtained in the step two to obtain a predicted value of the bus peak load;
step four, respectively counting the bus peak load under the extreme high temperature condition and the bus peak load under the extreme low temperature condition through a linear model, and obtaining that the bus peak load and the extreme high temperature and the extreme low temperature have linear relations according to the counting result;
step five, carrying out linear fitting on the extreme high temperature and the extreme low temperature and the bus peak load by using a least square method, and respectively obtaining the linear models of the peak load prediction of the bus 1 and the bus 2 under the extreme temperature condition as follows:
under the extreme high temperature condition, the linear model of the bus 1 peak load prediction is as follows:
under the extremely low temperature condition, the linear model of the bus 1 peak load prediction is as follows:
under the extreme high temperature condition, the linear model of the bus 2 peak load prediction is as follows:
under the extremely low temperature condition, the linear model of the bus 2 peak load prediction is as follows:
in the formula:
peak load of the bus bar 1 under extremely high temperature conditions; />
Peak load of the bus bar 2 under extremely high temperature conditions; />
Peak load of bus bar 1 at extremely low temperature; />
Peak load of the bus bar 2 under extremely low temperature conditions; t is the extreme temperature.
The natural weather includes longitude, latitude, temperature, air pressure, humidity, wind direction and wind speed.
The social factors include date and holiday.
The prediction precision in the second step is the average absolute percentage error (Mean Absolute Percent Error, MAPE) between the predicted value and the actual value, and the calculation formula of MAPE is as follows:
in the formula, Y i The actual measurement value of the bus peak load;
predicting a bus peak load value; n is the number of predicted samples.
In the second step, the specific method for obtaining the IPSO-ELM by improving the input weight and the threshold value of the particle swarm optimization extreme learning machine comprises the following steps:
1) Constructing an ELM model for predicting bus peak load based on the original feature set, and randomly generating an input weight omega of the ELM and a hidden layer unit bias threshold b;
2) Determining data of an original feature set required by input bus peak load prediction, and setting iteration times;
3) Predicting original feature set data aiming at bus peak load, and carrying out normalization processing;
4) Obtaining a fitness value according to the average absolute percentage error value of the bus peak load prediction of the ELM, and obtaining a current individual fitness value and a group optimal fitness value;
5) Based on a particle swarm algorithm PSO, updating the particle speed and the position according to a calculation formula of a nonlinear dynamic inertia weight coefficient and a learning factor calculation formula;
6) Calculating and obtaining a current particle fitness value, comparing the current particle fitness value with a historical optimal individual particle fitness value, and updating a particle optimal solution if the current particle fitness value is better; otherwise, maintaining the optimal fitness value of the individual;
7) The fitness value of the current particle is better than the optimal solution of the group, and the optimal solution of the group is updated; otherwise, maintaining the optimal solution of the population unchanged;
8) And (3) returning to the step (3) until the set iteration times are not reached, substituting the optimal solution input weight omega and the hidden layer unit bias threshold b into ELM, and constructing a bus peak load prediction model.
The expression of the nonlinear dynamic inertia weight coefficient is as follows:
wherein: w is inertial weight, w min Is the minimum value of inertia weight, w max Is the maximum value of the inertial weight; f is the particle fitness; f (f) avg Is the average fitness; f (f) min Is the minimum fitness.
The expression of the learning factor is:
wherein: c 1 And c 2 C is a learning factor 1s C is 1 Initial value of c 2s C is 2 Initial value of c 1c C is 1 Final value of c 2c C is 2 A final value of (2); item is the number of current iterations; ter (iter) max Is the total number of iterations.
The number of iterations is 200.
Through the design scheme, the invention has the following beneficial effects:
according to the invention, the characteristic importance results of the condition mutual information on the to-be-selected characteristics in the original characteristic set are used as a basis, the IPSO-ELM is combined as a predictor, forward characteristic selection is carried out, the optimal characteristic set of the bus peak load prediction is determined, the influence of the characteristic redundancy on the prediction precision in the bus peak load prediction is reduced, the optimal prediction model is respectively constructed for different buses, the prediction precision of different buses is effectively improved, the IPSO-ELM is combined with a linear method, and the peak load prediction under different scenes is carried out, so that the prediction needs under small samples or no-sample scenes are met.
Drawings
The invention is further described with reference to the drawings and detailed description which follow:
FIG. 1 is a scatter plot and fitted plot of the peak load average for bus 1 under extreme high temperature conditions in an embodiment of the bus peak load prediction method of the present invention accounting for complex weather effects.
FIG. 2 is a scatter plot and fitted plot of peak load averages for bus 2 under extreme high temperature conditions in an embodiment of a bus peak load prediction method of the present invention that accounts for complex weather effects.
FIG. 3 is a scatter plot and fitted plot of the peak load average for bus 1 under extreme low temperature conditions in an embodiment of the bus peak load prediction method of the present invention accounting for complex weather effects.
FIG. 4 is a plot and fit of a plot of average peak load scatter for bus 2 at extreme low temperatures in an embodiment of a bus peak load prediction method of the present invention that accounts for complex weather effects.
FIG. 5 is a graph of a characteristic importance analysis of the peak load of bus 1 in an embodiment of the bus peak load prediction method of the present invention that accounts for complex weather effects.
FIG. 6 is a graph of a peak load characteristic importance analysis of bus 2 in an embodiment of a bus peak load prediction method of the present invention that accounts for complex weather effects.
FIG. 7 is a diagram of optimal feature set selection for conditional mutual information in combination with different predictors in an embodiment of a bus peak load prediction method of the present invention that accounts for complex weather effects.
FIG. 8 is a graph comparing targeted and unified model bus peak load predictions in an embodiment of a bus peak load prediction method of the present invention that accounts for complex weather effects.
FIG. 9 is a graph comparing bus peak load predictions for three methods in an embodiment of a bus peak load prediction method of the present invention that accounts for complex weather effects.
Detailed Description
As shown in the figure, the bus peak load prediction method considering complex weather influences comprises the following steps:
1. bus peak load prediction original feature set construction
(1) Original feature set for predicting bus peak load
The peak load of the bus can be fluctuated due to various factors such as natural weather, society and the like. The raw feature set constructed by analysis of complex weather factors and related literature studies is shown in table 1.
TABLE 1 bus peak load prediction raw feature set
Note that:
1)F T indicating the peak value of the temperature on the day of the prediction, F Tave Mean value of temperature of the day of prediction; f (F) T(max,d-1) Representing the predicted day-to-day temperature peak, F T(ave,d-1) Representing the average temperature of the day before the predicted day; f (F) A Air pressure, F, representing the predicted time H Indicating humidity at the predicted time, F W Indicating the direction of wind at the predicted time, F W1 Wind speed F representing the predicted time A 、F H 、F W F (F) W1 Equal are meteorological features;
2)F G1 representing longitude, F of bus to be predicted G2 Representing the latitude of a bus to be predicted; f (F) T 、F Tave 、F T(max,d-1) 、F T(ave,d-1) 、F A 、F H 、F W 、F W1 、F G1 F (F) G2 The same is natural weather characteristics;
3)F D1 to F D7 Is the day of week; f (F) J1 Marking a working day; f (F) J2 Is a non-workday; f (F) H1 Representing holidays; f (F) H2 Indicating a normal day; f (F) D1 To F D7 、F J1 、F J2 、F H1 F (F) H2 Are all social factors;
4)F L(max,d-1) indicating the predicted day-to-day bus peak load, F L(max,d-2) Representing the predicted peak bus load two days before day, and so on; f (F) L(t-15) Represents the bus load 15 minutes before the peak load of the day before the day to be measured, F L(t-30) The bus load 30 minutes before the bus peak load on the day before the test day is indicated, and so on.
(2) Conditional mutual information
In the bus peak load prediction, setting D as an original feature set containing various factors such as natural weather, society and the like; q is the actual measured bus peak load value; the Z set is the selected feature. Mutual information between D and Q is:
in the formula (1), F (D; Q) represents mutual information between D and Q, P (D) is a marginal density function of D, P (Q) is a marginal density function of Q, and P (D, Q) is a joint probability density of D and Q.
Under the condition of knowing the Z set, the mutual information of the set D and the condition of the measured bus peak load value Q is as follows:
in the formula (2), F (D, q|Z) represents the condition mutual information between D and Q under the Z condition, and P (d|z) is a probability density function of D, Q under the Z condition; p (q|z) is a probability density function of Q under Z conditions; p (d, q|z) is a joint probability density function for D, Q under Z; p (d, q, z) is a joint probability density function of D, Q, Z.
The bus peak load has a plurality of influence factors, if all the influence factors are considered, information redundancy can be caused, and the bus peak load prediction precision is low. In order to improve the bus peak load prediction precision, the correlation between each influencing factor and different bus peak loads is analyzed through the condition mutual information, and the feature importance ranking in the original feature set is obtained.
2. Bus peak load prediction model construction
(2.1) extreme learning principle based on improved particle swarm optimization
Training predictors such as neural networks with limited bus peak load data makes it difficult to obtain an ideal bus peak load prediction model. Thus, an extreme learning mechanism suitable for small sample training is applied to construct a bus peak load predictor. In order to avoid the influence of improper parameter selection on the prediction effect, an improved particle swarm algorithm is applied to optimize the input weight and the threshold value of the extreme learning machine, and the bus peak load predictor is used for further improving the prediction precision.
(2.1.1) extreme learning machine
Is provided with N samples
Wherein the input data is x i ＝[x i1 ,x i2 ,…,x in ] T ∈R n The target output value is t i ＝[t i1 ,t i2 ,…,t im ] T ∈R m . The ELM extreme learning machine network model of the single hidden layer neural network with the number of hidden layer nodes being L can be expressed as
In the formula, o j Representing a network output value; g represents an activation function; omega i Is the input weight; beta i Is the output weight; b i Is the bias of the ith hidden layer unit; x is x j Represents x i Is a data set of the data set.
Without error, the activation function approaches arbitrarily near any N samples, i.e
Obtainable according to formula (4)
Wherein g represents an activation function; omega i Is the input weight; beta i Is the output weight; b i Is the bias of the ith hidden layer unit; t is t j Representing a network output value when the activation function is capable of approximating any N samples with zero error;
the N equation matrices in equation (5) are in the form of
Hβ＝T (6)
Wherein, the liquid crystal display device comprises a liquid crystal display device,
h represents hidden layer node output; t is the desired output; beta represents a weight matrix between the output layer and the hidden layer.
By obtaining
And->
Implementing ELM training of single hidden layer so that
In the method, in the process of the invention,
representing beta i An optimal solution; />
Represents ω i Is the optimal solution of (a); />
Representation b i Is the optimal solution of (a); h represents hidden layer node output; t is the desired output;
the minimum loss function equivalent to equation (9) is
Wherein E represents a minimum loss function; g represents an activation function; omega i Is the input weight; beta i Is the output weight; b i Is the bias of the ith hidden layer cell.
In ELM, the input weight ω and the bias threshold b of the hidden layer unit once randomly determined, a unique hidden layer node output H is obtained. From this, ELM structure is determined.
(2.2) improved particle swarm theory based on improved particle swarm extreme learning machine parameter optimization (2.2.1)
In order to optimize the problem that the particle swarm algorithm is easy to fluctuate around the global optimal solution, the conventional particle swarm is improved by inertia weight, and the expression is as follows:
in the formula (11): w represents inertial weight, w min Is the minimum value of inertia weight, w max Is the maximum value of the inertial weight; f is the particle fitness; f (f) avg Is the average fitness; f (f) min Is the minimum fitness.
In order to enable the particle swarm optimization algorithm to quickly determine the globally optimal solution, the conventional particle swarm of the formula (12) is dynamically adjusted, namely
In (12): c 1 And c 2 Are learning factors, c 1s C is 1 Initial value of c 2s C is 2 Initial value of c 1c C is 1 A final value of (2); c 2c C is 2 A final value of (2); item is the number of current iterations; ter (iter) max Is the total number of iterations.
(2.2.2) improved particle swarm-based ELM parameter optimization
Taking the optimal process of constructing a bus peak load prediction model by taking an original feature set as an example, the IPSO (Internet protocol security) optimization ELM (electronic component management) process is shown as follows:
1) Constructing a bus peak load prediction ELM based on the original feature set, and randomly generating an input weight omega of the ELM and a hidden layer unit bias threshold b;
2) Determining data of an original feature set required by input bus load peak prediction, and setting iteration times;
3) Aiming at the data of the original characteristic set of bus peak load prediction, carrying out normalization processing;
4) Obtaining a fitness value according to the average absolute percentage error of the bus peak load prediction of the ELM, and determining the optimal fitness value of the current individuals and the current groups;
5) Based on the traditional PSO, according to formulas (11) and (12), updating the particle speed and the position;
6) Firstly, calculating a current particle fitness value, then comparing the current particle fitness value with a historical optimal value, and if the current particle fitness value is more optimal, updating a particle optimal solution; otherwise, maintaining the optimal fitness value of the individual;
7) If the fitness value of the current particle is better than the optimal solution of the group, updating the optimal solution of the group; otherwise, maintaining the optimal solution of the population unchanged;
8) And if the iteration times are not reached, returning to 3), otherwise, substituting the optimal solutions omega and b into ELM, and constructing a bus peak load prediction model.
And in the later feature selection link, constructing a targeted optimal ELM predictor for different dimension feature sets, and adopting the method.
3. No training sample generating line peak load predictive model
The ELM et al predictor, while suitable for small sample prediction, still relies on historical samples for its construction. If the extreme temperatures (extreme high and extreme low temperatures) which have not occurred in the history of occurrence of the predicted day, the prediction effect cannot be ensured. In order to improve the prediction precision of the bus peak load without training samples, a linear model is introduced to independently predict the bus peak load of the extremely high and low temperature scene without training samples.
Examples:
statistical analysis was performed on the peak loads of multiple bus bars under extreme weather conditions of extreme high and low temperatures in 2018 in certain city in northeast. In this example, the extremely high temperature is set to be higher than 30 degrees in the day maximum temperature, and the extremely low temperature is set to be lower than 20 degrees below zero in the day minimum temperature. The statistical result shows that the peak load of the rhizoma anemarrhenae line has obvious linear relation with the highest temperature, so that the least square method is utilized to carry out linear fitting on the peak load of the pole end and the bus.
Fig. 1 to 4 show a scatter plot of the average value of the peak loads of the bus bars and corresponding fitted curves under extreme weather conditions of extreme high and low temperatures of bus bar 1 and bus bar 2 in 2018 in a city in northeast, respectively. From fig. 1 to 4 and the linear model of the peak load of the bus under extreme weather conditions of the two buses, it is known that the peak load of the different buses is affected differently by the extreme weather. Therefore, in extreme weather conditions, specific modeling analysis is required for different buses, and the prediction accuracy of the bus peak load is improved.
The linear model of peak load prediction for bus 1 and bus 2 under extreme weather conditions by linear fitting of extreme temperature to bus peak load using least squares is as follows:
under the extreme high temperature condition, the linear model of the bus 1 peak load prediction is as follows:
under the extremely low temperature condition, the linear model of the bus 1 peak load prediction is as follows:
and obtaining the linear prediction model of the bus 2 by adopting the same method.
Under the extreme high temperature condition, the linear model of the bus 2 peak load prediction is as follows:
under the extremely low temperature condition, the linear model of the bus 2 peak load prediction is as follows:
in the formula:
and->
Peak loads of the bus 1 and the bus 2 under the extremely high temperature condition respectively; />
And->
Peak loads of the bus 1 and the bus 2 under the extremely low temperature condition respectively; t is the extreme temperature.
In the research, the peak load and meteorological information data of buses in the northeast of 2018 are applied, wherein 1, 4, 8 and 10 months of data are used as verification sets, 7 months of data are used as test sets, the rest 7 months are used as training sets, and targeted feature selection is carried out on different buses. To demonstrate the advancement of the new method, experiments were performed with IPSO-ELM in comparison with ELM and BP neural networks (Back Propagation Neural Model, BPNN). The model prediction effect was evaluated with mean absolute percentage error (Mean Absolute Percent Error, MAPE), root mean square error (Root Mean Square Error, RMSE), and the index calculation method was as follows:
in (17) and (18) Y i The actual measurement value of the bus peak load;
predicting a bus peak load value; n is the number of predicted samples.
(1) Conditional mutual information feature selection analysis
Fig. 5 and 6 are graphs showing characteristic importance of peak loads of different buses, and it is clear from the graphs that there is a difference in correlation between peak loads of different buses and characteristics. FIG. 7 shows the number of features contained in the optimal feature set and the corresponding MAPE value after CMI is combined with IPSO-ELM, BPNN, ELM to select the optimal feature set of the busbar 1 respectively; table 2 gives that the optimal feature set contains features. As can be seen from fig. 7, the characteristic set to be selected is different and the prediction error is different in predicting the bus peak load. Errors in bus peak load prediction are minimal when IPSO-ELM, BPNN, ELM is used as the predictor and feature subset feature dimensions are 25, 28, 34, respectively. As can be seen from fig. 7, when BPNN, ELM, IPSO-ELM is used as the predictors according to the optimal feature set, the peak load prediction MAPE values of the bus 1 are respectively 4.54%, 3.75% and 3.04%, and among the three predictors, the peak load prediction MAPE value of the IPSO-ELM bus is the smallest, which indicates the advantage of high prediction accuracy of the IPSO-ELM. Likewise, an optimal feature subset for bus 2 may be determined.
TABLE 2 optimal feature set
(2) Bus peak load prediction result analysis
In order to prove the advantage of high precision of the bus peak load prediction method, the peak load prediction results of the bus 1 and the bus 2 in the northeast of 7 months in 2018 are listed. FIG. 8 shows the bus peak load prediction results of two buses for targeted modeling and constructing a prediction model by adopting the original feature set; table 3 corresponds to the bus peak load prediction error in fig. 8. Under the optimal model, the MAPE predicted by the peak load of the two buses is 3.04% and 2.98% respectively; when modeling the original feature set, the peak load of the two buses was 3.89% and 4.01% respectively. By comparison, the method can develop targeted feature selection aiming at different buses, and a bus peak load prediction model is established in a targeted manner according to feature selection results, so that the prediction accuracy is high.
Fig. 9 shows the prediction results of 7 months bus peak load for bus 1 in northeast of 2018 when IPSO-ELM, and BPNN were used as bus peak load predictors, respectively, and table 4 corresponds to the prediction errors of the three bus peak load prediction methods in fig. 9. As can be seen from Table 4, the MAPE for predicting bus peak loads using BPNN, ELM and IPSO-ELM were 4.23%,3.93%, 3.04%, respectively, with the minimum MAPE for predicting peak loads of the IPSO-ELM bus. Therefore, for the problem of severe fluctuation of bus peak load caused by complex meteorological factors, the method provided by the invention has high prediction precision.
TABLE 3 bus peak load prediction error
Table 4 peak load prediction error for three methods
In order to further verify the prediction effectiveness of the novel method under the extremely high-temperature and extremely low-temperature scenes without training samples, the day with the highest and lowest air temperature in 2018 in the northeast city is removed from the training samples, the day with the highest and lowest air temperature is used as the day to be predicted, and the historical data before the day with the highest and lowest air temperature is used as training data. The daily peak loads of different buses were predicted by the improved particle swarm optimization extreme learning machine and the linear model, and the results are shown in Table 5. As can be seen from Table 5, the linear model is used for predicting the peak load of the bus without the historical temperature sample, which has higher precision, and the novel method is proved to be capable of effectively avoiding the peak load prediction error caused by the historical data deletion at the extreme temperature without the training sample. The new method has better applicability.
Table 5 extreme weather day bus peak load prediction
In summary, based on the condition mutual information value, the characteristic selection of bus peak load prediction is carried out, the influence of characteristic redundancy on prediction precision in bus peak load prediction is reduced, the optimal prediction model is respectively constructed for different buses, the prediction precision of different buses is effectively improved, an improved particle swarm optimization limit learning machine is introduced to be combined with a linear method, the peak load prediction in different scenes is carried out, and the prediction needs in small-sample or sample-free scenes are met.
Claims (5)
1. A bus peak load prediction method considering complex weather influence is characterized in that: comprising the following steps, and the following steps are carried out in sequence,
firstly, constructing an original characteristic set for predicting the bus peak load according to natural weather and social factors based on the influence of the natural weather and the social factors on the bus peak load prediction precision, and analyzing the correlation between each characteristic in the original characteristic set and the bus peak load through conditional mutual information (Conditional Mutual Information, CMI) to obtain characteristic importance ranking;
step two, optimizing the input weight and the threshold value of the extreme learning machine through an improved particle swarm algorithm to obtain an improved particle swarm optimized extreme learning machine (Improved Particle Swarm Optimization-Extreme Learning Machine, IPSO-ELM), then carrying out targeted forward feature selection according to the feature importance ranking result obtained in the step one by taking the IPSO-ELM prediction precision as a decision variable to obtain an optimal feature subset for bus peak load prediction, and retraining the IPSO-ELM according to the obtained optimal feature subset to obtain an optimal bus peak load prediction model;
the specific method for obtaining the IPSO-ELM by optimizing the input weight and the threshold value of the extreme learning machine through the improved particle swarm algorithm comprises the following steps:
1) Constructing an ELM model for predicting bus peak load based on the original feature set, and randomly generating an input weight omega of the ELM and a hidden layer unit bias threshold b;
2) Determining data of an original feature set required by input bus peak load prediction, and setting iteration times;
3) Predicting original feature set data aiming at bus peak load, and carrying out normalization processing;
4) Obtaining a fitness value according to the average absolute percentage error value of the bus peak load prediction of the ELM, and obtaining a current individual fitness value and a group optimal fitness value;
5) Based on a particle swarm algorithm PSO, updating the particle speed and the position according to a calculation formula of a nonlinear dynamic inertia weight coefficient and a learning factor calculation formula;
6) Calculating and obtaining a current particle fitness value, comparing the current particle fitness value with a historical optimal individual particle fitness value, and updating a particle optimal solution if the current particle fitness value is better; otherwise, maintaining the optimal fitness value of the individual;
7) The fitness value of the current particle is better than the optimal solution of the group, and the optimal solution of the group is updated; otherwise, maintaining the optimal solution of the population unchanged;
8) Returning to the step 3) until the set iteration times are not reached, substituting the optimal solution input weight omega and the hidden layer unit bias threshold b into ELM, and constructing a bus peak load prediction model;
the expression of the nonlinear dynamic inertia weight coefficient is as follows:
wherein: w is the inertial weight, w min Is the minimum value of inertia weight, w max Is the maximum value of the inertial weight; f is the particle fitness; f (f) avg Is the average fitness; f (f) min Is the minimum fitness;
the expression of the learning factor is:
wherein: c 1 And c 2 C is a learning factor 1s C is 1 Initial value of c 2s C is 2 Initial value of c 1c C is 1 Final value of c 2c C is 2 A final value of (2); item is the number of current iterations; ter (iter) max Is the total number of iterations;
substituting the test set data in the set time period in the historical data into the optimal bus peak load prediction model obtained in the step two to obtain a predicted value of the bus peak load;
step four, respectively counting the bus peak load under the extreme high temperature condition and the bus peak load under the extreme low temperature condition through a linear model, and obtaining that the bus peak load and the extreme high temperature and the extreme low temperature have linear relations according to the counting result;
step five, carrying out linear fitting on the extreme high temperature and the extreme low temperature and the bus peak load by using a least square method, and respectively obtaining the linear models of the peak load prediction of the bus 1 and the bus 2 under the extreme temperature condition as follows:
under the extreme high temperature condition, the linear model of the bus 1 peak load prediction is as follows:
under the extremely low temperature condition, the linear model of the bus 1 peak load prediction is as follows:
under the extreme high temperature condition, the linear model of the bus 2 peak load prediction is as follows:
under the extremely low temperature condition, the linear model of the bus 2 peak load prediction is as follows:
in the formula:
peak load of the bus bar 1 under extremely high temperature conditions; />
Peak load of the bus bar 2 under extremely high temperature conditions; />
Peak load of bus bar 1 at extremely low temperature; />
Peak load of the bus bar 2 under extremely low temperature conditions; t is the extreme temperature.
2. The bus peak load prediction method considering complex weather effects according to claim 1, wherein: the natural weather includes longitude, latitude, temperature, air pressure, humidity, wind direction and wind speed.
3. The bus peak load prediction method considering complex weather effects according to claim 1, wherein: the social factors include date and holiday.
4. The bus peak load prediction method considering complex weather effects according to claim 1, wherein: the prediction precision in the second step is the average absolute percentage error (Mean Absolute Percent Error, MAPE) between the predicted value and the actual value, and the calculation formula of MAPE is as follows:
in the formula, Y i The actual measurement value of the bus peak load;
predicting a bus peak load value; n is the number of predicted samples.
5. The bus peak load prediction method considering complex weather effects according to claim 1, wherein: and the iteration times in the second step are 200 times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911056853.6A CN110807508B (en) | 2019-10-31 | 2019-10-31 | Bus peak load prediction method considering complex weather influence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911056853.6A CN110807508B (en) | 2019-10-31 | 2019-10-31 | Bus peak load prediction method considering complex weather influence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110807508A CN110807508A (en) | 2020-02-18 |
| CN110807508B true CN110807508B (en) | 2023-06-09 |
Family
ID=69489951
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911056853.6A Active CN110807508B (en) | 2019-10-31 | 2019-10-31 | Bus peak load prediction method considering complex weather influence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110807508B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112348235A (en) * | 2020-10-21 | 2021-02-09 | 国网安徽省电力有限公司阜阳供电公司 | Wind-solar bus load self-adaptive prediction method and device and computer equipment |
| CN112308169B (en) * | 2020-11-10 | 2022-05-03 | 浙江大学 | Effluent quality prediction method based on improved online sequence extreme learning machine |
| CN116432821B (en) * | 2023-03-08 | 2023-09-26 | 河北省气象服务中心(河北省气象影视中心) | Meteorological condition-based next-day power load peak prediction method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102799953A (en) * | 2012-07-16 | 2012-11-28 | 安徽省电力公司池州供电公司 | Bus load prediction method based on stacked generalization training strategy |
| CN106485262A (en) * | 2016-09-09 | 2017-03-08 | 国网山西省电力公司晋城供电公司 | A kind of bus load Forecasting Methodology |
| CN106855957A (en) * | 2015-12-09 | 2017-06-16 | 四川大学 | Factory's bus load prediction based on similar day and least square method supporting vector machine |
| CN107403239A (en) * | 2017-07-25 | 2017-11-28 | 南京工程学院 | A kind of parameters analysis method for being used for control device in power system |
| CN107480815A (en) * | 2017-08-03 | 2017-12-15 | 国网河北省电力公司保定供电分公司 | A kind of power system taiwan area load forecasting method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8024076B2 (en) * | 2003-06-27 | 2011-09-20 | Intelilcon, Inc. | Voltage collapse diagnostic and ATC system |
-
2019
- 2019-10-31 CN CN201911056853.6A patent/CN110807508B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102799953A (en) * | 2012-07-16 | 2012-11-28 | 安徽省电力公司池州供电公司 | Bus load prediction method based on stacked generalization training strategy |
| CN106855957A (en) * | 2015-12-09 | 2017-06-16 | 四川大学 | Factory's bus load prediction based on similar day and least square method supporting vector machine |
| CN106485262A (en) * | 2016-09-09 | 2017-03-08 | 国网山西省电力公司晋城供电公司 | A kind of bus load Forecasting Methodology |
| WO2018045642A1 (en) * | 2016-09-09 | 2018-03-15 | 国网山西省电力公司晋城供电公司 | A bus bar load forecasting method |
| CN107403239A (en) * | 2017-07-25 | 2017-11-28 | 南京工程学院 | A kind of parameters analysis method for being used for control device in power system |
| CN107480815A (en) * | 2017-08-03 | 2017-12-15 | 国网河北省电力公司保定供电分公司 | A kind of power system taiwan area load forecasting method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110807508A (en) | 2020-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109615226B (en) | Operation index abnormity monitoring method | |
| CN110807508B (en) | Bus peak load prediction method considering complex weather influence | |
| CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
| Lin et al. | High-precision forecast using grey models | |
| CN107944594B (en) | Short-term load prediction method based on spearman grade and RKELM microgrid | |
| CN111861013A (en) | Power load prediction method and device | |
| CN109902859B (en) | Queuing peak period estimation method based on big data and machine learning algorithm | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN112435054B (en) | Nuclear extreme learning machine electricity sales prediction method based on generalized maximum correlation entropy criterion | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN111882157A (en) | Demand prediction method and system based on deep space-time neural network and computer readable storage medium | |
| CN115983485A (en) | Prediction method and system for carbon emission of power plant | |
| CN112381282A (en) | Photovoltaic power generation power prediction method based on width learning system | |
| Liu et al. | Short‐term load forecasting based on LSTNet in power system | |
| Girimurugan et al. | Application of deep learning to the prediction of solar irradiance through missing data | |
| Ibrahim et al. | LSTM neural network model for ultra-short-term distribution zone substation peak demand prediction | |
| CN117674119A (en) | Power grid operation risk assessment method, device, computer equipment and storage medium | |
| CN112288187A (en) | Big data-based electricity sales amount prediction method | |
| Peng et al. | Meteorological satellite operation prediction using a BiLSTM deep learning model | |
| Li et al. | Hydropower generation forecasting via deep neural network | |
| CN115310355A (en) | Multi-energy coupling-considered multi-load prediction method and system for comprehensive energy system | |
| Tong et al. | A prediction model for complex equipment remaining useful life using gated recurrent unit complex networks | |
| CN114943328A (en) | SARIMA-GRU time sequence prediction model based on BP neural network nonlinear combination | |
| CN114861555A (en) | Regional comprehensive energy system short-term load prediction method based on Copula theory | |
| CN116544931B (en) | Power load distribution prediction method based on integrated fragment transformation and time convolution network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |