CN109146121A - The power predicating method stopped in the case of limited production based on PSO-BP model - Google Patents
The power predicating method stopped in the case of limited production based on PSO-BP model Download PDFInfo
- Publication number
- CN109146121A CN109146121A CN201810661629.9A CN201810661629A CN109146121A CN 109146121 A CN109146121 A CN 109146121A CN 201810661629 A CN201810661629 A CN 201810661629A CN 109146121 A CN109146121 A CN 109146121A
- Authority
- CN
- China
- Prior art keywords
- pso
- neural network
- model
- weight
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a kind of power predicating methods stopped under limited production policy based on PSO-BP model for belonging to power quantity predicting technical field.This method is first analyzed and processed input data;Then, using history electricity consumption influence factor as independent variable, sample training is carried out using history electricity consumption as dependent variable, uses the weight and threshold value of PSO algorithm optimization BP neural network, the precision of prediction for calculating different parameters, obtains the weight and threshold value of the high BP model of precision of prediction;Finally to BP neural network model prediction, parameter and forecast sample input prediction model after particle swarm algorithm is optimized obtain predicted value.The present invention utilizes PSO Optimized BP Neural Network algorithm, consider air quality index, meteorologic factor and mainly stops influence of the products under restricted quota yield factors to electricity consumption, learning training is carried out to the feature vector of electricity consumption, experiments verify that prediction effect is more satisfactory, a kind of new approaches are provided for the regional electricity demand forecasting stopped under limited production policy implication.
Description
Technical field
The invention belongs to power quantity predicting technical field more particularly to it is a kind of based on PSO-BP model stop limited production in the case of
Power predicating method.
Background technique
The development of power industry must meet socio-economic development and the people by the way that the operational efficiency of power grid is continuously improved
The demand of household electricity.Correctly judgement and prediction it is following electricity needs variation tendency it is accurate for electric power enterprise, it is scientific, close
That manages plan, the stability and economy for improving Operation of Electric Systems have vital meaning.In recent years, in order to have
Effect guarantees major event and the movable air quality for holding area, and government will implement temporary halt production, limited production measure.Great
Stop limited production measure during activity while being obviously improved air quality, terminal enterprise electricity consumption level is greatly lowered very much.Therefore,
There is most important theories value and realistic meaning to research is carried out based on the regional electricity demand forecasting method for stopping limiting the production under policy implication.
For electricity demand forecasting method, existing Predicting Technique can be divided mainly into two classes: one kind is with regression analysis, grey
Forecasting type, the traditional prediction method that time series predicting model is representative, one kind is using BP neural network, support vector machines as generation
The intelligent predicting technology of table.With the raising required model prediction accuracy, Classical forecast technology can not meet prediction gradually,
Just gradually substituted by intelligent predicting technology.At present in electricity demand forecasting research, researchers are primarily upon economic factor, season
Rule between section factor, meteorologic factor and electricity consumption.But rarely has from air quality index, meteorologic factor and stop limited production amount
Angle considers to stop the intelligent Forecasting of regional electricity consumption under limited production policy implication.
Summary of the invention
In view of the above-mentioned problems, what the invention proposes a kind of based on PSO-BP model stops the power quantity predicting side in the case of limited production
Method, which comprises the following steps:
Step 1: the influence factor of analysis regional electricity demand forecasting in the case where stopping limited production policy constructs PSO-BP model;
Step 2: using sample data as the input of PSO-BP model, and it being normalized, wherein sample number
According to including history electricity consumption and history electricity consumption influence factor;
Step 3: PSO-BP model training sample data is utilized, using history electricity consumption influence factor as PSO-BP model
Independent variable, utilize PSO in network error back-propagation process using history electricity consumption as the dependent variable of PSO-BP model
The weight and threshold value of algorithm optimization BP neural network, the BP nerve by calculating the precision of prediction of different parameters, after being optimized
The weight and threshold value of network;
Step 4: training PSO-BP model, by the weight of the BP neural network after PSO algorithm optimization and threshold value and
Forecast sample inputs in PSO-BP model, obtains the daily power consumption predicted value of forecast sample.
The method of the step 1 building PSO-BP model are as follows: define the particle position in PSO first and correspond to BP nerve net
One group of weight threshold to be optimized in network utilizes PSO algorithm optimization BP neural network in network error back-propagation process
The weight of BP neural network is by weight and threshold value by unified sequential arrangement that is, on the basis of determining neural network structure
The element of one vector obtains BP neural network forward-propagating process using the vector as a particle in population
Fitness function of the error as PSO algorithm is found optimal BP network by the loop iteration of BP neural network and PSO algorithm
Weight.
The calculation formula that data are normalized in the step 2 are as follows:
Wherein,It is the sample data through normalized, xiIt is sample data, xmaxIt is the maximum value in sample data,
xminIt is the minimum value in sample data.
The history electricity consumption influence factor includes AQI index, iron and steel output, cement output, the highest temperature, minimum gas
Temperature, wind-force, precipitation and day type.
The step 3 is determining neural network structure using the weight and threshold value of PSO algorithm optimization BP neural network
On the basis of, the error that BP neural network is obtained in forward-propagating process is as the fitness function of PSO algorithm, by BP nerve net
The loop iteration of network and PSO algorithm finds the weight of optimal BP neural network, uses in network error back-propagation process
The weight and threshold value of PSO algorithm optimization BP neural network, wherein selecting adaptation of the mean square deviation of BP neural network as PSO algorithm
Function is spent, by the weight and threshold value of PSO algorithm optimization BP neural network, so that the mean square deviation of BP neural network is minimum, it is described
Fitness function indicates are as follows:
In formula, N indicates the sample number of training;Indicate the desired output of i-th of sample, j-th of network output node;
yj,iIndicate the real output value of j-th of network output node of i-th of sample;M indicates neural network output layer number of nodes.
The step 3 utilizes PSO-BP model training sample data method particularly includes:
(1) input variable, output variable, input layer number n, node in hidden layer h of BP neural network model are determined
And output node layer m, establish network topology structure;
(2) weight and threshold value of PSO-BP model, the position including particle are initializedAnd speedPopulation sum
N, maximum number of iterations Tmax, inertia weight ω maximum value ωmaxAnd minimum value ωmin, Studying factors c1And c2;
(3) sample data of the input variable to BP neural network model and output variable is normalized;
(4) by the input variable and output variable data input BP neural network model after normalized, particle is calculated
Fitness function value, obtain the individual optimal value and global optimum of particle;
(5) by the individual optimal value pbest and global optimum of the fitness function value of each particle and current time
Gbest is compared, and records the position of current optimal particle;
(6) fitness value of each particle is evaluated, if the value is better than individual optimal value Pbd, then by individual optimal value PbdIf
It is set to current value, and updates the individual optimal value of the particle;If the individual optimal value in particle is better than current global optimum,
Global optimum then is set by the individual optimal value and updates global optimum;
(7) weight and threshold value for utilizing PSO algorithm optimization BP neural network, by the weight and threshold value substitution BP mind after optimization
Through being trained in network, and weight and threshold value are adjusted, is less than preset error when meeting BP neural network mean square error
Or when maximum number of iterations, then stop iteration, otherwise output is as a result, continue iteration until algorithmic statement.
The beneficial effects of the present invention are:
The present invention considers air quality using population (PSO) algorithm optimization model parameter by BP neural network algorithm
Index, meteorologic factor and mainly stop influence of the products under restricted quota yield factors to electricity consumption, to the feature vector of electricity consumption
Training is practised, prediction effect is more satisfactory, and the regional electricity demand forecasting under policy implication provides a kind of new approaches to stop limiting the production.
Detailed description of the invention
Attached drawing 1 is the short-term power predicating method flow diagram under the influence of the limited production of stopping based on PSO-BP model;
Attached drawing 2 is PSO-BP neural network algorithm flow chart;
Attached drawing 3 is that 1 period of large-scale activity key industry electricity consumption cuts down schematic diagram;
Attached drawing 4 is that large-scale activity 2 period electricity consumption cuts down schematic diagram;
Attached drawing 5 is 1 front and back of large-scale activity and period AQI index and yield data and electricity consumption tendency chart;
Attached drawing 6 is 2 front and back of large-scale activity and period AQI index and yield data and electricity consumption tendency chart;
Attached drawing 7 is 1 period of large-scale activity and 2 period of large-scale activity and its front and back area B daily power consumption match value and reality
Value;
Attached drawing 8 is 3 period of large-scale activity electricity demand forecasting result;
Specific embodiment
The present invention is described in detail with reference to the accompanying drawings and examples.
Attached drawing 1 is the short-term power predicating method flow diagram under the influence of the limited production of stopping based on PSO-BP model, such as Fig. 1 institute
Show, described method includes following steps:
Step 1: analyzing the influence factor that electricity consumption is cut down under the influence of stopping limited production, construct PSO-BP model;
Step 2: using sample data as the input of PSO-BP model, and it being normalized, wherein sample number
According to including history electricity consumption and history electricity consumption influence factor;
Step 3: PSO-BP model training sample data is utilized, using history electricity consumption influence factor as PSO-BP model
Independent variable, utilize PSO in network error back-propagation process using history electricity consumption as the dependent variable of PSO-BP model
The weight and threshold value of algorithm optimization BP neural network, the BP nerve by calculating the precision of prediction of different parameters, after being optimized
The weight and threshold value of network;
Step 4: training PSO-BP model, by the weight of the BP neural network after PSO algorithm optimization and threshold value and
Forecast sample inputs in PSO-BP model, obtains the daily power consumption predicted value of forecast sample.
Specifically, passing through the influence factor of analysis electricity consumption reduction in the step 1, it is believed that under the influence of stopping limited production
Area's electricity consumption variation with air quality level, stop limited production yield there are certain correlations.Therefore, the present invention to stop limit the production shadow
When ringing lower regional electricity consumption and being predicted, by AQI index, iron and steel output, cement output, the highest temperature, the lowest temperature, wind-force,
Precipitation, day type this eight factors are included in electricity demand forecasting important factor in order.And use 1 period of large-scale activity and large-scale activity
2 periods and its area the B history daily power consumption data in surrounding time section and air quality index, temperature are meteorological, stop limited production amount
Historical data constructs PSO-BP model.
Specifically, the PSO-BP model that the step 1 constructs is by the global search feature of PSO algorithm and BP nerve net
The local fast search feature of network algorithm combines, and using PSO algorithm optimization BP neural network, falls into local pole to avoid network
It is small, and then improve the training speed of network;Define first the particle position in PSO correspond in BP neural network one group it is to be optimized
Weight threshold, pass through the optimal solution of random particles group in interative computation search space, in iterative process each time, particle energy
Enough find out the optimal solution p at current time itselfbestWith the optimal solution g of entire populationbest, the search of space optimal solution is completed, is led to
It crosses and finds optimum particle position and obtain optimal neural network structure, the optimal neural network structure is recycled to be predicted;
Concrete analysis process is as follows:
(1) BP neural network
BP neural network is become as a kind of artificial intelligence approach by the training of neural network and the available output of study
Measure the interneuronal connection weight of numerical function relationship and interlayer about input variable.Commonly used to the nerve net predicted
Network is a three-layer forward networks, and the transfer function of neuron is nonlinear function, and the most commonly used is logsig and tansig letters
Number, output are as follows:
A=logsig (Wp+b) (1)
If the output layer of network uses S-shaped transfer function (such as logsig), output valve will be limited in one smaller
In the range of (0,1);And arbitrary value can then be taken using linear transmission function.
Network is trained by outputting and inputting sample set, that is, threshold value and weight are learnt and corrected, so that
The given input-output mappings relationship of network implementations.The learning process of network is divided into two stages:
First stage is to input known learning sample, passes through the network structure of setting and the weight and threshold of preceding an iteration
Value calculates the output of a neuron from the first layer of network backward.
Second stage is modified to weight and threshold value, is calculated each weight and threshold value forward from the last layer and is missed to total
The influence (gradient) of difference, accordingly modifies to each weight and threshold value.Above two processes alternately and repeatedly, are converged to until reaching
Only.
(2) particle swarm algorithm (PSO)
The basic thought of particle swarm algorithm (PSO) is potentially optimal by random particles group in interative computation search space
Solution, in iteration each time, particle can find out the optimal solution p at current time itselfbestWith the optimal solution of entire population
gbest, in fact, PSO algorithm is also the search for completing space optimal solution by the competition and cooperation of individual.
Assuming that, by the molecular group of m grain, the position of i-th of particle is expressed as x in D dimension object spacei=
(xi1,xi2,...,xiD), i=1,2 ..., m, speed are expressed as vi=(vi1,vi2,...,viD), i=1,2 ..., m, itself
Optimal location is expressed as Pbest=(pi1,pi2,...,piD), the optimal location of i=1,2 ..., m, entire population are expressed as
gbest=(g1,g2,...,gD), then the speed of particle and location updating equation are as follows:
Speed renewal equation:
vid(t+1)=vid(t)+c1r1(pbest-xid(t))+c2r2(gbest-xid(t)) (2)
Location updating equation:
xid(t+1)=xid+vid(t+1) (3)
In formula, c1, c2For non-negative aceleration pulse, r1,r2For the uniform random number on obedience [0,1];
In order to preferably control the detection and development ability of PSO algorithm, introduced on the basis of PSO algorithm for its speed term
Inertia weight factor ω, it may be assumed that
vid(t+1)=ω * vid(t)+c1r1(pbest-xid(t))+c2r2(gbest-xid(t)) (4)
In fact, basic PSO is the special circumstances under inertia weight factor w=1, local convergence energy is lacked in iteration
Power is easily trapped into local minimum;In order to solve this problem, the balance of the global and local optimizing ability of PSO is realized, we
It joined inertia weight factor ω before speed term.In each iteration, ω is linear decrease, its calculation formula is:
In formula, ωmax, ωminThe respectively maximin of ω;T, TmaxRespectively current iteration number and greatest iteration number.
(3) PSO-BP neural network model
Currently, there are mainly two types of schemes for PSO Optimizing BP Network.The first is instructed by the weight and threshold value of optimization network
Practice neural network, i.e., on the basis of determining neural network structure, it is one that the weight of BP network, which is pressed unified sequential arrangement,
Then the element of vector obtains BP neural network forward-propagating process using the vector as a particle in population
Fitness function of the error as PSO algorithm is found optimal BP network by the loop iteration of BP neural network and PSO algorithm
Weight;For second the topological structure with PSO algorithm optimization BP neural network, weight, the implicit number of plies including network and its
Number of nodes, this method is more complicated, because the variation of network structure influences whether the variation of PSO algorithm solution space dimension, realize compared with
Difficulty also largely influences convergence speed of the algorithm.Therefore the present invention is with the first optimization method to BP nerve net
Network optimizes.
The key of PSO Optimized BP Neural Network weight is: how to establish the dimension of particle in the weight in BP network and PSO
Between mapping.The component of each of PSO particle is corresponding with a weight in BP network, i.e., BP network how many
How many a weight and threshold value, each of population particle are tieed up with regard to;The fitness function of PSO selects the mean square error of BP network
Difference, the PSO algorithm by powerful search property keep the mean square error of BP network minimum.PSO Optimizing BP Network is anti-in network error
The weight and threshold value of PSO Optimizing BP Network are used into communication process.The present invention selects mean square deviation (Mean Square
Error, MSE) fitness function as PSO-BP neural network, calculation formula is as follows:
In formula, N indicates the sample number of training;Indicate the desired output of i-th of sample, j-th of network output node;
yj,iIndicate the real output value of j-th of network output node of i-th of sample;M indicates neural network output layer number of nodes.
Specifically, choosing AQI index, iron and steel output, cement output, the highest temperature, the lowest temperature, wind in the step 2
Speed, input variable of 8 factors of precipitation and day type as model, the output for choosing daily power consumption data as model become
Amount, is normalized according to formula (7):
Wherein,It is the sample data through normalized, xiIt is sample data, xmaxIt is the maximum value in sample data,
xminIt is the minimum value in sample data.
Specifically, the step 3 using PSO-BP model training sample data specific method as shown in Fig. 2, include with
Lower step:
(1) it initializes.The input variable and output variable for determining BP neural network model, establish network topology structure, i.e.,
It determines input layer number n, node in hidden layer h, exports node layer m;Initialize the weight and threshold value of PSO-BP model, packet
Include the position of particleAnd speedPopulation sum N, maximum number of iterations Tmax, inertia weight ω maximum value ωmaxAnd
Minimum value ωmin, Studying factors c1And c2;
(2) data prediction.The sample data of input variable and output variable to BP neural network model carries out normalizing
Change processing, eliminates the difference as caused by different dimensions between data, data is made more to adapt to BP neural network prediction model, data normalizing
The method of change includes standardization method, threshold method, extreme value facture and normalized method etc., can be selected according to actual needs
Take appropriate method.
(3) training sample data.First by the input variable and output variable data input BP nerve after normalized
Network model calculates the fitness function value of particle, obtains the individual optimal value and global optimum of particle;Then by each grain
The fitness function value of son is compared with the individual optimal value pbest and global optimum gbest at current time, and record is current
The position of optimal particle;The fitness value of each particle is evaluated, if the value is better than individual optimal value Pbd, then by individual optimal value
PbdIt is set as current value, and the individual for updating the particle is optimal;If the individual optimal value in particle is better than current global optimum
Value then by the optimal global optimum for being set to the particle of individual, and updates global extremum;Finally, utilizing PSO algorithm optimization BP nerve
The weight and threshold value of network, by after optimization weight and threshold value substitute into BP neural network and be trained, and adjust weight and threshold
Value then stops iteration when meeting BP neural network mean square error less than preset error or maximum number of iterations, exports
As a result, otherwise continuing iteration until algorithmic statement.
BP neural network passes through the training and the available output of study of BP neural network as a kind of artificial intelligence approach
Numerical function relationship and interlayer interneuronal connection weight of the variable about input variable.Network is by outputting and inputting sample
This collection is trained, i.e., threshold value and weight is learnt and corrected, to realize given input-output mappings relationship.Population
Algorithm (PSO) is a kind of Swarm Intelligence Algorithm that the search of space optimal solution is completed by individual competition and cooperation, base
This thought is by the potential optimal solution of random particles group in interative computation search space, in iteration each time, particle meeting
Find out the optimal solution p at current time itselfbestWith the optimal solution g of entire populationbest.Currently, utilizing PSO algorithm optimization BP mind
Through network, there are mainly two types of methods, the first is to train neural network by the weight and threshold value of optimization network, i.e., in determination
On the basis of neural network structure, the weight of BP network is pressed into the element that unified sequential arrangement is a vector, by the vector
As a particle in population, the error for then obtaining BP neural network forward-propagating process is as the suitable of PSO algorithm
Response function is found the weight of optimal BP network by the loop iteration of BP neural network and PSO algorithm;Second is to use
The topological structure of PSO algorithm optimization BP neural network, weight, the implicit number of plies and its number of nodes including network, however this method
More complicated, because the variation of network structure influences whether the variation of PSO algorithm solution space dimension, therefore realization is more difficult, also exists
Largely influence convergence speed of the algorithm.Therefore, the present invention carries out BP neural network using the first optimization method
Optimization.Using PSO algorithm optimization BP neural network weight and threshold value it is crucial that how to establish the weight and PSO in BP network
Mapping between the dimension of middle particle.The component of each of PSO particle is corresponding with a weight in BP network, false
If BP network has n weight and threshold value, then corresponding to each of population particle has n dimension.Determining neural network structure
On the basis of, the error that BP neural network is obtained in forward-propagating process is as the fitness function of PSO algorithm, by BP nerve net
The loop iteration of network and PSO algorithm finds the weight of optimal BP neural network, uses in network error back-propagation process
The weight and threshold value of PSO algorithm optimization BP neural network, the present invention select mean square deviation MSE as the suitable of PSO-BP neural network
Response function keeps the mean square deviation MSE of BP neural network minimum by the weight and threshold value of PSO algorithm optimization BP neural network.Its
In, fitness function indicates are as follows:
In formula, N indicates the sample number of training;Indicate the desired output of i-th of sample, j-th of network output node;
yj,iIndicate the real output value of j-th of network output node of i-th of sample;M indicates neural network output layer number of nodes.
Embodiment 1
Stop limited production measure to the influence factor of electricity consumption reduction to analyze, the present embodiment is living with large-scale activity 1 and large size
For dynamic 2, practical decomposition is carried out to the electricity consumption during large-scale activity 1 and during large-scale activity 2, then cut down electricity consumption is decomposed
Analysis comparison, obtains stopping limited production measure to the influence factor of electricity consumption reduction.
(1) large-scale activity 1
The present embodiment chooses electricity consumption and accounts for key industry of the industry of 3% or more the area B Analyzing Total Electricity Consumption as research,
I.e. ferrous metal ore picks up industry, nonmetallic grounded module, chemical raw material and chemical product manufacturing, ferrous metal smelting and pressure
Prolong processing industry, metal product industry and electric power, heating power production and supply.This 6 trade power consumption amounts are total to account for the annual total use in the area B
The reduction of the 68.37% of electricity, these trade power consumption amounts can produce bigger effect the whole electricity consumptions in the area B, and electricity consumption is cut
It is as shown in Figure 3 to subtract situation.From the figure 3, it may be seen that the practical electricity consumption of every profession and trade large-scale activity 1 limit stop production during compared to event before and after
There is larger decline in period, wherein the biggish industry of power consumption is cut down amplitude and also become apparent from more greatly.With ferrous metal smelting
And for calendering processing industry, the sector electricity consumption in 2014 accounts for the 35.57% of the area B total electricity consumption, and electricity consumption reduction is with reaching B
The 39.12% of the total reduction of Qu, this is because highly energy-consuming trade is the emphasis for stopping limiting halt production in limited production policy implication, therefore can be right
The area B total electricity consumption produces a very large impact.Other key industrys also have during large-scale activity 1 cuts down by a relatively large margin, wherein gold
The downslide that metal products industry, ferrous metal ore pick up the trade power consumptions amounts such as industry also produces very big influence to the area B electricity consumption.
(2) large-scale activity 2
During large-scale activity 2, ferrous metal ore picks up industry, nonmetallic grounded module, chemical raw material and chemicals
Manufacturing industry, ferrous metal smelting and rolling processing industry, metal product industry and electric power, heating power production and supply this 6 key industrys
Electricity consumption is total to account for the 65.05% of the area B total electricity consumption in 2015, and the area B is all used in the reduction of these trade power consumption amounts
Electricity produces bigger effect, and electricity consumption cuts down situation as shown in Fig. 4.As seen from Figure 4, each row during large-scale activity 2
The reduction of industry electricity consumption is obvious, has certain gap compared with the electricity consumption of front and back period, compared with history same period electricity consumption
Difference is also larger, is equally that the reduction of highly energy-consuming trade is bigger.Wherein, ferrous metal smelting and rolling processing industry is stopping to limit twice
Reduction rate in production policy event is 35% or more, more than the 1/3 of the full area B whole society electricity consumption reduction, in every profession and trade
Reduction is maximum, and the degree for illustrating that relevant enterprise limit stops production is maximum.The reduction rate twice of the tertiary industry is 11% or so, in B
Also occupy larger specific gravity in area's electricity consumption reduction.Ferrous metal ore picks up industry, nonmetallic grounded module and electric power, heating power life
It produces and supply industry also all occupies larger specific gravity, larger impact is all produced to the reduction of the area B electricity consumption.
Analysis is cut down it is found that the 6 big important trade power consumption amounts in the area B account for about regional total electricity consumption by above-mentioned electricity needs
66.7%.The 6 big industry is the emphasis for stopping limited production simultaneously, therefore stopping limited production measure can be to the very big shadow of regional total electricity consumption generation
It rings.In addition, six big industries are also the main source of the area B air pollution, local pollution control discharge will be significantly reduced by stopping limited production measure,
Improve air quality.The present embodiment chooses air quality index, iron and steel output and cement output representative and stops limited production measure influence, and
Using it as electricity demand forecasting model important factor in order.Attached drawing 5 is 1 front and back of large-scale activity and period AQI index and yield number
According to electricity consumption tendency chart, attached drawing 6 is the front and back of large-scale activity 2 and period AQI index and yield data and electricity consumption tendency chart, by
Fig. 5 and Fig. 6 can be seen that halt production, the limited production measure implemented during large-scale activity 1 and large-scale activity 2 make the weight such as steel, cement
The output of industrial product is substantially cut down, and a large amount of pollutant effulent is reduced, and significantly improves the air quality in the area B.Stop limiting the production
The electricity consumption level in the area B is also significantly cut down in the implementation of measure.Since the meteorologic factors such as temperature, precipitation also can be to the use of electricity consumption
Situation has an impact.Meanwhile day type (working day, weekend) can also have an impact electricity consumption level.Therefore, the present embodiment is right
When the regional electricity consumption under policy implication that stops limiting the production is predicted, by AQI index, iron and steel output, cement output, the highest temperature, most
Low temperature, wind speed, precipitation and day type are included in electricity consumption important factor in order.
Embodiment 2
In order to verify effectiveness of the invention and practicability, the present embodiment chooses large-scale activity 1, large-scale activity 2 and large size
Input of 8 influence factors of 3 period of activity and its front and back area stage B air quality indexes as PSO-BP prediction model
Variable carries out predictive simulation, the sky to the electricity consumption during activity using the daily power consumption data in the area B as output variable
8 influence factors of makings figureofmerit include AQI index, iron and steel output, cement output, the highest temperature, the lowest temperature, wind speed,
Precipitation and day type, the network input layer neuron number of the PSO-BP prediction model are 8, hidden layer neuron number
It is 1 for 16, output layer neuron number, population sum N=60, maximum number of iterations Tmax=100, inertia weight ω is most
Big value and minimum value ωmax=0.9 and ωmin=0.3, Studying factors c1=c2=2, picture display rate 100;Anticipation error is most
Small value is 0.00065;Daily power consumption is learnt and emulates that details are provided below:
(1) data input and initialization
Input data is normalized according to formula (9), the input data include AQI index, iron and steel output,
Cement output, the highest temperature, the lowest temperature, wind speed, precipitation and day type;
Wherein, xiIt is sample data, xmaxIt is the maximum value in sample data, xminIt is the minimum value in sample data.
(2) training PSO-BP prediction model
By the history number of large-scale activity 1, large-scale activity 2 related electricity consumption and influence factor in above-mentioned normalized data
Learn according to bringing into PSO-BP model, and according to the training of model, obtains the daily power consumption prediction under the influence of limited production is stopped in the area B
As a result.
Using the obtained fitting output valve of PSO-BP neural network and actual value as shown in fig. 7, being solved according to formula (10)
Average relative error, calculation formula are as follows:
Wherein, Y indicates actual value,Indicate match value.
(3) prediction result and analysis
The area large-scale activity 3 period B daily power consumption is predicted using trained PSO-BP prediction model, and prediction result is such as
Shown in Fig. 8.Table 1 is that large-scale activity 3 period electricity demand forecasting error compares under different models, by PSO-BP prediction result in table 1
It compares and finds with SVM, BP prediction result, SVM model relatine error for prediction is that 4.69%, BP relatine error for prediction is
2.35%, PSO-BP relatine error for prediction are 1.37%, it can thus be seen that improved PSO-BP fitting degree is better than
SVM and BP, and there is smaller relative prediction residual, illustrate that particle swarm algorithm can improve BP model prediction to a certain extent
Precision.Experiments verify that prediction effect of the invention is more satisfactory, to consider that the Accurate Prediction for the daily power consumption for stopping limited production policy mentions
A kind of new approaches are supplied.
Large-scale activity 3 period electricity demand forecasting error compares under the different models of table 1
This embodiment is merely preferred embodiments of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
Subject to.
Claims (6)
1. a kind of power predicating method stopped under limited production policy based on PSO-BP model, which comprises the following steps:
Step 1: the influence factor of analysis regional electricity demand forecasting in the case where stopping limited production policy constructs PSO-BP model;
Step 2: using sample data as the input of PSO-BP model, and it being normalized, wherein sample data packet
Include history electricity consumption and history electricity consumption influence factor;
Step 3: PSO-BP model training sample data is utilized, using history electricity consumption influence factor becoming certainly as PSO-BP model
Amount, it is excellent using PSO algorithm in network error back-propagation process using history electricity consumption as the dependent variable of PSO-BP model
The weight and threshold value for changing BP neural network, by calculating the precision of prediction of different parameters, BP neural network after being optimized
Weight and threshold value;
Step 4: training PSO-BP model, by the weight of the BP neural network after PSO algorithm optimization and threshold value and prediction
Sample inputs in PSO-BP model, obtains the daily power consumption predicted value of forecast sample.
2. a kind of power predicating method stopped under limited production policy based on PSO-BP model according to claim 1, special
Sign is, the method for the step 1 building PSO-BP model are as follows: define the particle position in PSO first and correspond to BP nerve net
One group of weight threshold to be optimized in network utilizes PSO algorithm optimization BP neural network in network error back-propagation process
The weight of BP neural network is by weight and threshold value by unified sequential arrangement that is, on the basis of determining neural network structure
The element of one vector obtains BP neural network forward-propagating process using the vector as a particle in population
Fitness function of the error as PSO algorithm is found optimal BP network by the loop iteration of BP neural network and PSO algorithm
Weight.
3. a kind of power predicating method stopped under limited production policy based on PSO-BP model according to claim 1, special
Sign is, the calculation formula that data are normalized in the step 2 are as follows:
Wherein,It is the sample data through normalized, xiIt is sample data, xmaxIt is the maximum value in sample data, xminIt is
Minimum value in sample data.
4. a kind of power predicating method stopped under limited production policy based on PSO-BP model according to claim 1, special
Sign is, the history electricity consumption influence factor include AQI index, iron and steel output, cement output, the highest temperature, the lowest temperature,
Wind-force, precipitation and day type.
5. a kind of power predicating method stopped under limited production policy based on PSO-BP model according to claim 1, special
Sign is the step 3 is determining neural network structure using the weight and threshold value of PSO algorithm optimization BP neural network
On the basis of, the error that BP neural network is obtained in forward-propagating process is as the fitness function of PSO algorithm, by BP nerve net
The loop iteration of network and PSO algorithm finds the weight of optimal BP neural network, uses in network error back-propagation process
The weight and threshold value of PSO algorithm optimization BP neural network, wherein selecting adaptation of the mean square deviation of BP neural network as PSO algorithm
Function is spent, by the weight and threshold value of PSO algorithm optimization BP neural network, so that the mean square deviation of BP neural network is minimum, it is described
Fitness function indicates are as follows:
In formula, N indicates the sample number of training;Indicate the desired output of i-th of sample, j-th of network output node;yj,i
Indicate the real output value of j-th of network output node of i-th of sample;M indicates neural network output layer number of nodes.
6. a kind of power predicating method stopped under limited production policy based on PSO-BP model according to claim 1, special
Sign is that the step 3 utilizes PSO-BP model training sample data method particularly includes:
(1) determine the input variable of BP neural network model, output variable, input layer number n, node in hidden layer h and
Node layer m is exported, network topology structure is established;
(2) weight and threshold value of PSO-BP model, the position including particle are initializedAnd speedPopulation sum N, maximum
The number of iterations Tmax, inertia weight ω maximum value ωmaxAnd minimum value ωmin, Studying factors c1And c2;
(3) sample data of the input variable to BP neural network model and output variable is normalized;
(4) by the input variable and output variable data input BP neural network model after normalized, the suitable of particle is calculated
Response functional value obtains the individual optimal value and global optimum of particle;
(5) by the individual optimal value pbest and global optimum gbest at the fitness function value of each particle and current time into
Row compares, and records the position of current optimal particle;
(6) fitness value of each particle is evaluated, if the value is better than individual optimal value Pbd, then by individual optimal value PbdIt is set as working as
Preceding value, and update the individual optimal value of the particle;If the individual optimal value in particle is better than current global optimum, should
Individual optimal value is set as global optimum and updates global optimum;
(7) weight and threshold value for utilizing PSO algorithm optimization BP neural network, by the weight and threshold value substitution BP nerve net after optimization
It is trained in network, and adjusts weight and threshold value, when meeting BP neural network mean square error less than preset error or most
When big the number of iterations, then stop iteration, otherwise output is as a result, continue iteration until algorithmic statement.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810661629.9A CN109146121A (en) | 2018-06-25 | 2018-06-25 | The power predicating method stopped in the case of limited production based on PSO-BP model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810661629.9A CN109146121A (en) | 2018-06-25 | 2018-06-25 | The power predicating method stopped in the case of limited production based on PSO-BP model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109146121A true CN109146121A (en) | 2019-01-04 |
Family
ID=64802297
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810661629.9A Pending CN109146121A (en) | 2018-06-25 | 2018-06-25 | The power predicating method stopped in the case of limited production based on PSO-BP model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109146121A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110309971A (en) * | 2019-06-28 | 2019-10-08 | 广州市准诚市场咨询有限公司 | A kind of monthly electricity demand forecasting method and system with automatic super rope of engaging in an inspection |
CN110533484A (en) * | 2019-09-05 | 2019-12-03 | 四川长虹电器股份有限公司 | A kind of product Method for Sales Forecast method based on PCA and improved BP |
CN110895772A (en) * | 2019-11-11 | 2020-03-20 | 国网浙江省电力有限公司宁波供电公司 | Electricity sales amount prediction method based on combination of grey correlation analysis and SA-PSO-Elman algorithm |
CN111339716A (en) * | 2020-02-19 | 2020-06-26 | 浙江大学 | Boiler high-temperature flue gas flow field online proxy model construction method |
CN111339157A (en) * | 2020-02-20 | 2020-06-26 | 南方电网科学研究院有限责任公司 | Method, system and equipment for calculating and predicting daily operation efficiency of power distribution network |
CN111695800A (en) * | 2020-06-03 | 2020-09-22 | 南京中电科能技术有限公司 | Method for positioning abnormal electricity consumption behavior user through cloud edge collaborative step-by-step analysis |
CN112215391A (en) * | 2020-08-07 | 2021-01-12 | 沈阳工业大学 | Gear remanufacturing quality prediction model method based on PSO-BP |
CN112819189A (en) * | 2019-11-15 | 2021-05-18 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Wind power output prediction method based on historical predicted value |
CN112966867A (en) * | 2021-03-11 | 2021-06-15 | 山东德佑电气股份有限公司 | Transformer early warning method based on PSO-BP neural network and quartile method |
CN113052385A (en) * | 2021-03-29 | 2021-06-29 | 国网河北省电力有限公司经济技术研究院 | Method, device, equipment and storage medium for predicting power consumption in steel industry |
CN113191574A (en) * | 2021-05-28 | 2021-07-30 | 上海申瑞继保电气有限公司 | Daily electricity prediction method for single product production line |
CN113326960A (en) * | 2020-02-28 | 2021-08-31 | 南京理工大学 | Subway traction energy consumption prediction method based on particle swarm optimization LSTM |
CN113392100A (en) * | 2021-08-17 | 2021-09-14 | 北京好欣晴移动医疗科技有限公司 | System intelligent verification method, device and system based on particle swarm optimization neural network |
CN114239972A (en) * | 2021-12-20 | 2022-03-25 | 广州汇锦能效科技有限公司 | Campus energy efficiency and electrical safety management method and system based on artificial intelligence technology |
CN114970774A (en) * | 2022-07-29 | 2022-08-30 | 国网经济技术研究院有限公司 | Intelligent transformer fault prediction method and device |
CN115656446A (en) * | 2022-12-26 | 2023-01-31 | 沃客森信息科技(常州)有限公司 | Air quality detection system and method based on Internet of things |
-
2018
- 2018-06-25 CN CN201810661629.9A patent/CN109146121A/en active Pending
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110309971A (en) * | 2019-06-28 | 2019-10-08 | 广州市准诚市场咨询有限公司 | A kind of monthly electricity demand forecasting method and system with automatic super rope of engaging in an inspection |
CN110533484A (en) * | 2019-09-05 | 2019-12-03 | 四川长虹电器股份有限公司 | A kind of product Method for Sales Forecast method based on PCA and improved BP |
CN110895772A (en) * | 2019-11-11 | 2020-03-20 | 国网浙江省电力有限公司宁波供电公司 | Electricity sales amount prediction method based on combination of grey correlation analysis and SA-PSO-Elman algorithm |
CN112819189A (en) * | 2019-11-15 | 2021-05-18 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Wind power output prediction method based on historical predicted value |
CN111339716A (en) * | 2020-02-19 | 2020-06-26 | 浙江大学 | Boiler high-temperature flue gas flow field online proxy model construction method |
CN111339716B (en) * | 2020-02-19 | 2022-07-22 | 浙江大学 | Boiler high-temperature flue gas flow field online agent model construction method |
CN111339157A (en) * | 2020-02-20 | 2020-06-26 | 南方电网科学研究院有限责任公司 | Method, system and equipment for calculating and predicting daily operation efficiency of power distribution network |
CN111339157B (en) * | 2020-02-20 | 2023-05-16 | 南方电网科学研究院有限责任公司 | Method, system and equipment for calculating and predicting daily operation efficiency of power distribution network |
CN113326960A (en) * | 2020-02-28 | 2021-08-31 | 南京理工大学 | Subway traction energy consumption prediction method based on particle swarm optimization LSTM |
CN113326960B (en) * | 2020-02-28 | 2022-08-23 | 南京理工大学 | Subway traction energy consumption prediction method based on particle swarm optimization LSTM |
CN111695800A (en) * | 2020-06-03 | 2020-09-22 | 南京中电科能技术有限公司 | Method for positioning abnormal electricity consumption behavior user through cloud edge collaborative step-by-step analysis |
CN111695800B (en) * | 2020-06-03 | 2023-10-31 | 南京中电科能技术有限公司 | Method for locating abnormal electricity behavior user through cloud edge collaborative progressive analysis |
CN112215391A (en) * | 2020-08-07 | 2021-01-12 | 沈阳工业大学 | Gear remanufacturing quality prediction model method based on PSO-BP |
CN112966867A (en) * | 2021-03-11 | 2021-06-15 | 山东德佑电气股份有限公司 | Transformer early warning method based on PSO-BP neural network and quartile method |
CN113052385A (en) * | 2021-03-29 | 2021-06-29 | 国网河北省电力有限公司经济技术研究院 | Method, device, equipment and storage medium for predicting power consumption in steel industry |
CN113191574A (en) * | 2021-05-28 | 2021-07-30 | 上海申瑞继保电气有限公司 | Daily electricity prediction method for single product production line |
CN113392100A (en) * | 2021-08-17 | 2021-09-14 | 北京好欣晴移动医疗科技有限公司 | System intelligent verification method, device and system based on particle swarm optimization neural network |
CN114239972A (en) * | 2021-12-20 | 2022-03-25 | 广州汇锦能效科技有限公司 | Campus energy efficiency and electrical safety management method and system based on artificial intelligence technology |
CN114970774A (en) * | 2022-07-29 | 2022-08-30 | 国网经济技术研究院有限公司 | Intelligent transformer fault prediction method and device |
CN115656446A (en) * | 2022-12-26 | 2023-01-31 | 沃客森信息科技(常州)有限公司 | Air quality detection system and method based on Internet of things |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109146121A (en) | The power predicating method stopped in the case of limited production based on PSO-BP model | |
Shi et al. | Energy consumption prediction of office buildings based on echo state networks | |
CN107909220A (en) | Electric heating load prediction method | |
CN112415924A (en) | Energy-saving optimization method and system for air conditioning system | |
CN108846526A (en) | A kind of CO2 emissions prediction technique | |
CN105447509A (en) | Short-term power prediction method for photovoltaic power generation system | |
CN109214449A (en) | A kind of electric grid investment needing forecasting method | |
CN109376950A (en) | A kind of polynary Load Forecasting based on BP neural network | |
CN110222883A (en) | Load Prediction In Power Systems method based on wind Drive Optimization BP neural network | |
CN106920014A (en) | A kind of short-term load forecasting method and device | |
CN110443418A (en) | Urban water consumption prediction technique based on GA-BP neural network | |
CN106447133A (en) | Short-term electric load prediction method based on deep self-encoding network | |
CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
CN108038568A (en) | A kind of changeable weight combination Short-Term Load Forecasting of Electric Power System based on particle cluster algorithm | |
CN110276472A (en) | A kind of offshore wind farm power ultra-short term prediction method based on LSTM deep learning network | |
Song et al. | An indoor temperature prediction framework based on hierarchical attention gated recurrent unit model for energy efficient buildings | |
CN110262431A (en) | A kind of sewage disposal process multiobjective optimization control method based on operating mode's switch | |
CN115186803A (en) | Data center computing power load demand combination prediction method and system considering PUE | |
CN112396301A (en) | Power consumer demand response characteristic control method based on energy big data driving | |
CN111428766B (en) | Power consumption mode classification method for high-dimensional mass measurement data | |
CN116562583A (en) | Multidimensional water resource supply and demand prediction method and system | |
CN105302976A (en) | Aluminum electrolysis production optimization method based on SPEA2 algorithm | |
CN109214610A (en) | A kind of saturation Methods of electric load forecasting based on shot and long term Memory Neural Networks | |
Teng et al. | Prediction of particulate matter concentration in Chengdu based on improved differential evolution algorithm and BP neural network model | |
CN111724064A (en) | Energy-storage-containing power distribution network planning method based on improved immune algorithm |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190104 |
|
RJ01 | Rejection of invention patent application after publication |