CN105913151A - Photovoltaic power station power generation amount predication method based on adaptive mutation particle swarm and BP network - Google Patents

Abstract

The invention discloses a photovoltaic power station power generation amount predication method based on an adaptive mutation particle swarm and a BP network. The method comprises the steps of acquiring and preprocessing a data sample, then constructing a BP neural network structure, optimizing the network by means of an adaptive mutation particle swarm algorithm, and introducing an optimal individual which is obtained after optimization into the BP neural network for predication. According to the method of the invention, the adaptive mutation particle swarm algorithm and the BP neural network are combined for performing real-time power generation amount predication on the photovoltaic power generation system; namely power generation state at next time period is predicated through power generation amount information of a former time period. According to the method of the invention, a relatively high global convergence capability of the adaptive mutation particle swarm algorithm is used for optimizing the initial weight and threshold of the network, thereby improving convergence speed and precision in power generation amount predication.

Description

Based on TSP question population and BP network photovoltaic power station power generation amount Forecasting Methodology

Technical field

The present invention relates to based on TSP question population and BP network photovoltaic power station power generation amount Forecasting Methodology, Belong to grid-connected photovoltaic power generation technical field.

Background technology

Solar energy power generating is that the photovoltaic effect utilizing solar cell can be straight by the radiation of the sun Switching through a kind of forms of electricity generation being changed to electric energy, solar energy is nexhaustible inexhaustible clean energy resource, Along with the development of photovoltaic power generation technology, photovoltaic generating system proportion shared by whole electrical network is more Come the biggest, owing to running of photovoltaic generating system is affected relatively greatly by extraneous factor, such as irradiation, temperature Degree, wind speed etc., can cause the problem such as jump volatility or intermittence of electricity generation system, this at random Property the operation of power system and power scheduling can be produced certain impact, the security of system of electrical network and stable Property also will be substantially reduced, and the most domestic also do not have the pre-examining system of the most ripe generated energy, therefore to photovoltaic The research of electricity generation system generated energy prediction the most increasingly highlights its importance, pre-by generated energy Survey can the generation schedule of schedule system within the specific limits, it is ensured that whole network system safety and stability Run.

Being currently based on the research that generated energy is predicted by BP neutral net more, BP neutral net has The strongest non-linear mapping capability, is good at from input and output signal finding rule, it is not necessary to accurately Mathematical Modeling, and computing capability is strong, is fixing yet with learning rate, therefore network Convergence rate is slow, needs the longer training time, and is easily trapped into the shortcomings such as local optimum.And it is adaptive Answer Mutation Particle Swarm Optimizer to make to jump out local optimum to be possibly realized, it is simply that at traditional particle cluster algorithm The process of middle addition genetic algorithm variation, carries out particle optimizing to initial weight, the threshold value of neutral net, Thus obtain network optimum initial weight, threshold value, improve precision of prediction.

Summary of the invention

Purpose: for the problems referred to above, the present invention proposes based on TSP question population and BP network light Overhead utility generated energy Forecasting Methodology, combines TSP question particle cluster algorithm with BP neutral net Photovoltaic generating system is carried out generated energy by time prediction, i.e. by upper moment generated energy information prediction not Carry out the power generation situation of subsequent time.

Technical scheme is as follows:

Based on TSP question population and BP network photovoltaic power station power generation amount Forecasting Methodology, comprise following Step:

(1), data sample is gathered: in choosing certain 12 days moon, 6:00 18:00 period every day affects light The principal element of volt generated energy, including solar irradiance, instantaneous wind speed, backboard temperature, environment temperature And ambient humidity, and gather influence factor and the reality of corresponding generated energy power of these photovoltaic power generation quantities Time data, with 1 hour as time scale, by five influence factors and generated output according to time series One_to_one corresponding, determining data sample collection, data sample totally 156 groups of data, in the sample set of collection before Within 11 days, as BP train samples collection, last day is then as test sample collection；

(2), data prediction: use the minimax method set of data samples to gathering in step (1) It is normalized pretreatment, obtains training sample set after treatment and test sample collection, described maximum Minimum method formula is as follows:

x_k=(x_k-x_min)/(x_max-x_min)

In formula, x_minFor the minimum number of data sequence, x_maxMaximum number for data sequence；

(3), network topology structure determines: BP neural network structure uses three layers of feed-forward type neutral net Structure, including input layer, hidden layer and output layer,

Input layer: include 6 nodes, each vector element of input vector group is to be generated electricity merit by history Rate and the solar irradiance of current time node, instantaneous wind speed, backboard temperature, environment temperature, environment Humidity is constituted, totally 132 groups of data；

Output layer: include 1 node, the current generated output in the most corresponding power station；

Hidden layer: nodes can determine according to below equation:

$l=round(\sqrt{n_i{m}_i}+0.5),$

In formula: round () expression rounds, n_iRepresent input layer number, m_iRepresent output layer node Number；

(4), network initial weight threshold value is determined, random in each weight threshold imparting (-1,1) Random value；

(5), network initial weight threshold value optimizing: use TSP question particle cluster algorithm neural to BP Network is optimized, and to obtain optimum initial weight threshold value, is embodied as step as follows:

(5a) network initial weight threshold length, is determined: input according to network described in step (3) Layer, hidden layer and output layer nodes determine network initial weight threshold length；

(5b), initialization of population and parameter and population: described parameter and population includes population scale, iteration time Number, the random position Pi of particle, speed Vi, velocity interval Vmin, Vmax, overall situation particle are best Fitness position P_(gbest), Studying factors c, described initialization of population refers to based on described parameter and population Randomly generate a population；

(5c) particle adaptive value, according to object function f () is calculated, the fitness letter of each particle Number is:

F ()=M | u_expk-u_actk|

In formula, u_actkFor kth step reality output, u_expkBeing that kth walks desired output, M is normal number, Calculate population particle fitness value, obtain P_(ibest)、P_(gbest), P_(ibest)Represent particle i position up till now Put the best fitness position self searched, P_(gbest)Represent the fitness that overall all particles are best Position；

(5d), particle rapidity, location updating: in population, the speed of i-th particle is: V_i=(v_i1,v_i2,…,v_iD), its individual extreme value is: P_i=(P_i1,P_i2,…,P_iD), population global extremum is: P_g=(P_g1,P_g2,…,P_gD), following formula represent particle pass through individual extreme value and global extremum update self speed, Position:

$V_i^{k+1}={\mathrm{wV}}_i^k+c_1{c}_2(P_i^k-X_i^k)+c_1{c}_2(P_g^k-X_i^k)$

$X_i^{k+1}=X_i^k+V_i^{k+1}$

In formula: k represents that the step number of iteration, Studying factors c1, c2 are the constant of non-negative, and w is inertia Weights, its value changes along with adaptive optimal control value rate of change K, and the expression formula of the two is as follows:

$K=\frac{|f\left(t\right)-f(t-5)|}{\left|f\left(t\right)\right|}$

When K >=0.05, w=0.6+r/2；

As K ＜ 0.05, w=0.2+r/2.

In formula: r for being uniformly distributed in the random number between [0,1], f (t) be population t generation Excellent adaptive value, f (t-5) is the adaptive optimal control value in population (t-5) generation；K represents that population is recently The relative change rate of adaptive optimal control value in 5 generations；

(5e), particle fitness value calculation: calculate current P according to step (5d)_(ibest)And P_(gbest), and make ratio with corresponding adaptive value, if the current adaptive value of particle is better than P_(ibest), then P is updated_(ibest)；If being better than P_(gbest), then P is updated_(gbest)；If P_(gbest)In a long time without significant change, Then need some particles in population according to certain Probability p_mCarrying out variation process, particle can reenter Other regional implementation search of the overall situation, constantly circulates, until finding new optimal solution；

(6), BP neural network forecast: comprise the following steps:

(6a), the optimum individual that step (5e) obtains is assigned to the initial weight threshold of BP neutral net Value；

(6b), the training sample after step (2) normalized is inputted BP network, input vector It is history generated output and the solar irradiation of corresponding moment future time node of the 1st day to the 11st day Degree, instantaneous wind speed, backboard temperature, environment temperature, ambient humidity；Object vector be the 1st day to The generated output of 11 days corresponding moment future time nodes；

(6c), e-learning: utilize in training sample input vector, optimum initial weight threshold calculations Interbed each unit exports, and then calculates output layer each unit by transferometer and responds, calculates accordingly Output layer each unit vague generalization error, then can obtain intermediate layer each unit vague generalization error, utilize defeated Go out a layer each unit vague generalization error, the output of intermediate layer each unit, intermediate layer vague generalization error and input Layer each unit input can be updated revising by layer weight threshold each to network respectively, to realize network by mistake The reverse propagation of difference, until precision or reach maximum iteration time needed for reaching network, so far, net Network learning process terminates；

(6d), by the test input vector in the test sample after normalized in step (2) and Test target vector inputs the BP neutral net trained respectively and is predicted drawing prediction output, will be pre- Survey output result carries out renormalization process and obtains predicted value.

Preferably, described in step (3), the nodes of hidden layer is 4.

Preferably, in described step (5a), network initial weight threshold length is 33, and weights are 28 Individual, threshold value is 5, then network weight threshold value can be considered as i-th particle vector, X_i=(x_i1,x_i2,…,x_iD), D is vector dimension, is weight threshold length 33.

Preferably, what variation described in step (5e) processed specifically comprises the following steps that

A (), size particle all to population according to adaptive value are ranked up, take out adaptive value best M particle, and except optimal particle；

B () produces corresponding m and is positioned at random number r before [0,1]_i(1≤i≤m), takes 0.1≤p_m≤ 0.3, if r_i＜ p_m, produce new position the most as the following formula:

$x_{id}^{i+1}=x_{id}^i(1+0.5\mathrm{\η})$

In formula:Be respectively variation before and after position, η be obey Gauss (0,1) be distributed with Machine variable；

C (), after iteration reaches predetermined step number, can obtain optimal solution P_(gbest), use following public affairs Formula is evolved, and search there may be more excellent solution, and constantly circulation is until upper limit iterative steps, so far Iteration terminates:

ΔP_(n+1)=m Δ P_n+(1-m)gP_n

P_(n+1)=P_n+ΔP_(n+1)

In formula: n represents the particle vector when the n-th generation, g is the random number between [0,0.1], m=0.5, Δ p represents current neighborhood scope, for random value, represents in current optimal solution P_(gbest)Neighbouring search is more excellent The process solved；

Particle P_(gbest)From the n-th generation to the (n+1)th generation, if current P_(n+1)Fitness value be better than P_(n), Then use P_(n+1)Replace P_(n)Otherwise, then keep P_(n)Being worth constant, being so circulated until reaching Till big iterations.

Preferably, in described step (6a), front 24 real numbers of optimum individual are followed successively by input layer to implicit The weights of layer, the 25 to 28th is followed successively by the input layer threshold value to hidden layer, and the 29 to 32nd depends on Secondary for hidden layer to output layer weights, the 33rd for hidden layer to output layer threshold value.

Preferably, the history generated output that test input vector is the 12nd day in described step (6d) The solar irradiance of future time node corresponding with the 12nd day, instantaneous wind speed, backboard temperature, environment temperature Degree, ambient humidity, test target vector be the 12nd day real-time generated output, prediction be output as the 12nd It prediction generated output.

Beneficial effect: the one that the present invention proposes is neural with BP based on TSP question particle cluster algorithm The photovoltaic generating system generated energy of network by time Forecasting Methodology, utilize TSP question particle cluster algorithm to have There is preferable global convergence ability that initial weight, the threshold value of network are carried out optimizing, improve generated energy pre- The convergence rate surveyed and precision.

Accompanying drawing explanation

Fig. 1 is the generated energy prediction flow process frame diagram of the present invention；

Fig. 2 is the topology diagram of BP neutral net.

Detailed description of the invention

For the technical scheme making those skilled in the art be more fully understood that in the application, below will knot Close the accompanying drawing in the embodiment of the present application, the technical scheme in the embodiment of the present application is carried out clear, complete Ground describes, it is clear that described embodiment is only some embodiments of the present application rather than all Embodiment.Based on the embodiment in the application, those of ordinary skill in the art are not making creation Property work premise under the every other embodiment that obtained, all should belong to the scope of the application protection.

Based on TSP question population and BP network photovoltaic power station power generation amount Forecasting Methodology, comprise following Step:

(1), data sample is gathered: in choosing certain 12 days moon, 6:00 18:00 period every day affects light The principal element of volt generated energy, including solar irradiance, instantaneous wind speed, backboard temperature, environment temperature And ambient humidity, and gather influence factor and the reality of corresponding generated energy power of these photovoltaic power generation quantities Time data, with 1 hour as time scale, by five influence factors and generated output according to time series One_to_one corresponding, determining data sample collection, data sample totally 156 groups of data, in the sample set of collection before Within 11 days, as BP train samples collection, last day is then as test sample collection；

(2), data prediction: use the minimax method set of data samples to gathering in step (1) It is normalized pretreatment, obtains training sample set after treatment and test sample collection, described maximum Minimum method formula is as follows:

x_k=(x_k-x_min)/(x_max-x_min)

In formula, x_minFor the minimum number of data sequence, x_maxMaximum number for data sequence；

(3), network topology structure determines: BP neural network structure uses three layers of feed-forward type neutral net Structure, including input layer, hidden layer and output layer,

Input layer: input layer by solar irradiance, instantaneous wind speed, backboard temperature, environment temperature, Ambient humidity and six elements of generated output are constituted, therefore include 6 nodes, input vector group each Vector element was by history (a upper moment, 1 hour) generated output and current time (subsequent time) The solar irradiance of node, instantaneous wind speed, backboard temperature, environment temperature, ambient humidity are constituted, altogether 132 groups of data；

Output layer: include 1 node, the current generated output in the most corresponding power station；

Hidden layer: nodes can determine according to below equation:

$l=round(\sqrt{n_i{m}_i}+0.5),$

In formula: round () expression rounds, n_iRepresent input layer number, m_iRepresent output layer node Number；

(4), network initial weight threshold value is determined, random in each weight threshold imparting (-1,1) Random value；

(5), network initial weight threshold value optimizing: use TSP question particle cluster algorithm neural to BP Network is optimized, and to obtain optimum initial weight threshold value, so-called TSP question particle cluster algorithm is sought Excellent, will network initial weight threshold value as particle in population, in population, each particle is according to self The process that fitness updates position, speed is exactly optimizing, but in conventional particle group's algorithm, particle holds Easily be absorbed in precocity, search precision is relatively low, later stage iteration is inefficient, therefore introduces in genetic algorithm and make a variation Operation, i.e. reinitializes with certain probability some variable, expands and the most constantly reduces Population search volume, make particle jump out prior searches to local optimum position, improve final gained grain The possibility that son is optimum, is embodied as step as follows:

(5a) network initial weight threshold length, is determined: input according to network described in step (3) Layer, hidden layer and output layer nodes determine network initial weight threshold length；

(5b), initialization of population and parameter and population: described parameter and population includes population scale, iteration time Number, the random position Pi of particle, speed Vi, velocity interval Vmin, Vmax, overall situation particle are best Fitness position P_(gbest), Studying factors c, described initialization of population refers to based on described parameter and population Randomly generate a population；

(5c) particle adaptive value, according to object function f () is calculated, the fitness letter of each particle Number is:

F ()=M | u_expk-u_actk|

In formula, u_actkFor kth step reality output, u_expkBeing that kth walks desired output, M is normal number, Calculate population particle fitness value, obtain P_(ibest)、P_(gbest), P_(ibest)Represent particle i position up till now Put the best fitness position self searched, P_(gbest)Represent the fitness that overall all particles are best Position；

Particle fitness value is calculated by fitness function, and fitness value size represents the quality of particle, The least more excellent, calculated fitness value optimal location in the most individual the experienced position of individual extreme value, Can be understood as the Pi corresponding to ideal adaptation angle value minimum；And all particles of colony's extreme value i.e. population are searched Rope is to fitness optimal location it can be understood as individual optimum set；

(5d), particle rapidity, location updating: in population, the speed of i-th particle is: V_i=(v_i1,v_i2,…,v_iD), its individual extreme value is: P_i=(P_i1,P_i2,…,P_iD), population global extremum is: P_g=(P_g1,P_g2,…,P_gD), following formula represent particle pass through individual extreme value and global extremum update self speed, Position:

$V_i^{k+1}={\mathrm{wV}}_i^k+c_1{c}_2(P_i^k-X_i^k)+c_1{c}_2(P_g^k-X_i^k)$

$X_i^{k+1}=X_i^k+V_i^{k+1}$

In formula: k represents that the step number of iteration, Studying factors c1, c2 are the constant of non-negative, and w is inertia Weights, its value changes along with adaptive optimal control value rate of change K, and the expression formula of the two is as follows:

$K=\frac{|f\left(t\right)-f(t-5)|}{\left|f\left(t\right)\right|}$

When K >=0.05, w=0.6+r/2；

As K ＜ 0.05, w=0.2+r/2.

In formula: r for being uniformly distributed in the random number between [0,1], f (t) be population t generation Excellent adaptive value, f (t-5) is the adaptive optimal control value in population (t-5) generation；K represents that population is recently The relative change rate of adaptive optimal control value in 5 generations；

(5e), particle fitness value calculation: calculate current P according to step (5d)_(ibest)And P_(gbest), and make ratio with corresponding adaptive value, if the current adaptive value of particle is better than P_(ibest), then P is updated_(ibest)；If being better than P_(gbest), then P is updated_(gbest)；If P_(gbest)In a long time without significant change, Then need some particles in population according to certain Probability p_mCarrying out variation process, particle can reenter Other regional implementation search of the overall situation, constantly circulates, until finding new optimal solution；

(6), BP neural network forecast: comprise the following steps:

(6a), the optimum individual that step (5e) obtains is assigned to the initial weight threshold of BP neutral net Value；

(6b), the training sample after step (2) normalized being inputted BP network, the 1st day arrives The history generated output of the 11st day and the solar irradiance of corresponding moment future time node (1 hour), Instantaneous wind speed, backboard temperature, environment temperature, ambient humidity；Object vector is the 1st day to the 11st day The generated output of corresponding moment future time node；

(6c), e-learning: utilize in training sample input vector, optimum initial weight threshold calculations Interbed each unit exports, and then calculates output layer each unit by transferometer and responds, calculates accordingly Output layer each unit vague generalization error, then can obtain intermediate layer each unit vague generalization error, utilize defeated Go out a layer each unit vague generalization error, the output of intermediate layer each unit, intermediate layer vague generalization error and input Layer each unit input can be updated revising by layer weight threshold each to network respectively, to realize network by mistake The reverse propagation of difference, until precision or reach maximum iteration time needed for reaching network, so far, net Network learning process terminates；

(6d), by the test input vector in the test sample after normalized in step (2) and Test target vector inputs the BP neutral net trained respectively and is predicted drawing prediction output, will be pre- Survey output result carries out renormalization process and obtains predicted value.

Preferably, described in step (3), the nodes of hidden layer is 4.

Preferably, in described step (5a), network initial weight threshold length is 33, and weights are 28 Individual, threshold value is 5, then network weight threshold value can be considered as i-th particle vector, X_i=(x_i1,x_i2,…,x_iD), D is vector dimension, is weight threshold length 33.

Preferably, what variation described in step (5e) processed specifically comprises the following steps that

A (), size particle all to population according to adaptive value are ranked up, take out adaptive value best M particle, and except optimal particle；

B () produces corresponding m and is positioned at random number r before [0,1]_i(1≤i≤m), takes 0.1≤p_m≤ 0.3, if r_i＜ p_m, produce new position the most as the following formula:

$x_{id}^{i+1}=x_{id}^i(1+0.5\mathrm{\η})$

In formula:Be respectively variation before and after position, η be obey Gauss (0,1) be distributed with Machine variable；

C (), after iteration reaches predetermined step number, can obtain optimal solution P_(gbest), use following public affairs Formula is evolved, and search there may be more excellent solution, and constantly circulation is until upper limit iterative steps, so far Iteration terminates:

ΔP_(n+1)=m Δ P_n+(1-m)gP_n

P_(n+1)=P_n+ΔP_(n+1)

In formula: n represents the particle vector when the n-th generation, g is the random number between [0,0.1], m=0.5, Δ p represents current neighborhood scope, for random value, represents in current optimal solution P_(gbest)Neighbouring search is more excellent The process solved；

Particle P_(gbest)From the n-th generation to the (n+1)th generation, if current P_(n+1)Fitness value be better than P_(n), Then use P_(n+1)Replace P_(n)Otherwise, then keep P_(n)Being worth constant, being so circulated until reaching Till big iterations.

Preferably, in described step (6a), front 24 real numbers of optimum individual are followed successively by input layer to implicit The weights of layer, the 25 to 28th is followed successively by the input layer threshold value to hidden layer, and the 29 to 32nd depends on Secondary for hidden layer to output layer weights, the 33rd for hidden layer to output layer threshold value.

Preferably, the history generated output that test input vector is the 12nd day in described step (6d) The solar irradiance of future time node corresponding with the 12nd day, instantaneous wind speed, backboard temperature, environment temperature Degree, ambient humidity, test target vector be the 12nd day real-time generated output, prediction be output as the 12nd It prediction generated output.

Calculating process and function formula that the present invention does not elaborates are those skilled in the art and are grasped Routine techniques means.

The above is only the preferred embodiment of the present invention, it is noted that for the art For those of ordinary skill, on the premise of without departing from the technology of the present invention principle, it is also possible to make some Improving and deformation, these improve and deformation also should be regarded as protection scope of the present invention.

Claims (6)

1., based on TSP question population and BP network photovoltaic power station power generation amount Forecasting Methodology, it is special Levy and be, comprise the steps of

(1), data sample is gathered: in choosing certain 12 days moon, 6:00 18:00 period every day affects light The principal element of volt generated energy, including solar irradiance, instantaneous wind speed, backboard temperature, environment temperature And ambient humidity, and gather influence factor and the reality of corresponding generated energy power of these photovoltaic power generation quantities Time data, with 1 hour as time scale, by five influence factors and generated output according to time series One_to_one corresponding, determining data sample collection, data sample totally 156 groups of data, in the sample set of collection before Within 11 days, as BP train samples collection, last day is then as test sample collection；

(2), data prediction: use the minimax method set of data samples to gathering in step (1) It is normalized pretreatment, obtains training sample set after treatment and test sample collection, described maximum Minimum method formula is as follows:

x_k=(x_k-x_min)/(x_max-x_min)

In formula, x_minFor the minimum number of data sequence, x_maxMaximum number for data sequence；

(3), network topology structure determines: BP neural network structure uses three layers of feed-forward type neutral net Structure, including input layer, hidden layer and output layer,

Input layer: include 6 nodes, each vector element of input vector group is to be generated electricity merit by history Rate and the solar irradiance of current time node, instantaneous wind speed, backboard temperature, environment temperature, environment Humidity is constituted, totally 132 groups of data；

Output layer: include 1 node, the current generated output in the most corresponding power station；

Hidden layer: nodes can determine according to below equation:

$l=round(\sqrt{n_i{m}_i}+0.5),$

In formula: round () expression rounds, n_iRepresent input layer number, m_iRepresent output layer node Number；

(4), network initial weight threshold value is determined, random in each weight threshold imparting (-1,1) Random value；

(5), network initial weight threshold value optimizing: use TSP question particle cluster algorithm neural to BP Network is optimized, and to obtain optimum initial weight threshold value, is embodied as step as follows:

(5a) network initial weight threshold length, is determined: input according to network described in step (3) Layer, hidden layer and output layer nodes determine network initial weight threshold length；

(5b), initialization of population and parameter and population: described parameter and population includes population scale, iteration time Number, the random position Pi of particle, speed Vi, velocity interval V_min、V_max, best suitable of overall situation particle Response position P_(gbest), Studying factors c, described initialization of population refer to based on described parameter and population with Machine produces a population；

(5c) particle adaptive value, according to object function f () is calculated, the fitness letter of each particle Number is:

F ()=M | u_expk-u_actk|

In formula, u_actkFor kth step reality output, u_expkBeing that kth walks desired output, M is normal number, Calculate population particle fitness value, obtain P_(ibest)、P_(gbest), P_(ibest)Represent particle i position up till now Put the best fitness position self searched, P_(gbest)Represent the fitness that overall all particles are best Position；

(5d), particle rapidity, location updating: in population, the speed of i-th particle is: V_i=(v_i1,v_i2,…,v_iD), its individual extreme value is: P_i=(P_i1,P_i2,…,P_iD), population global extremum is: P_g=(P_g1,P_g2,…,P_gD), following formula represent particle pass through individual extreme value and global extremum update self speed, Position:

$V_i^{k+1}={\mathrm{wV}}_i^k+c_1{c}_2(P_i^k-X_i^k)+c_1{c}_2(P_g^k-X_i^k)$

$X_i^{k+1}=X_i^k+V_i^{k+1}$

In formula: k represents that the step number of iteration, Studying factors c1, c2 are the constant of non-negative, and w is inertia Weights, its value changes along with adaptive optimal control value rate of change K, and the expression formula of the two is as follows:

$K=\frac{|f\left(t\right)-f(t-5)|}{\left|f\left(t\right)\right|}$

When K >=0.05, w=0.6+r/2；

As K ＜ 0.05, w=0.2+r/2.

In formula: r for being uniformly distributed in the random number between [0,1], f (t) be population t generation Excellent adaptive value, f (t-5) is the adaptive optimal control value in population (t-5) generation；K represents that population is recently The relative change rate of adaptive optimal control value in 5 generations；

(5e), particle fitness value calculation: calculate current P according to step (5d)_(ibest)And P_(gbest), and make ratio with corresponding adaptive value, if the current adaptive value of particle is better than P_(ibest), then P is updated_(ibest)；If being better than P_(gbest), then P is updated_(gbest)；If P_(gbest)In a long time without significant change, Then need some particles in population according to certain Probability p_mCarrying out variation process, particle can reenter Other regional implementation search of the overall situation, constantly circulates, until finding new optimal solution；

(6), BP neural network forecast: comprise the following steps:

(6a), the optimum individual that step (5e) obtains is assigned to the initial weight threshold of BP neutral net Value；

(6b), the training sample after step (2) normalized is inputted BP network, input vector It is history generated output and the solar irradiation of corresponding moment future time node of the 1st day to the 11st day Degree, instantaneous wind speed, backboard temperature, environment temperature, ambient humidity；Object vector be the 1st day to The generated output of 11 days corresponding moment future time nodes；

(6c), e-learning: utilize in training sample input vector, optimum initial weight threshold calculations Interbed each unit exports, and then calculates output layer each unit by transferometer and responds, calculates accordingly Output layer each unit vague generalization error, then can obtain intermediate layer each unit vague generalization error, utilize defeated Go out a layer each unit vague generalization error, the output of intermediate layer each unit, intermediate layer vague generalization error and input Layer each unit input, can be updated revising, to realize network by layer weight threshold each to network respectively The reverse propagation of error, until precision or reach maximum iteration time needed for reaching network, so far, Network learning procedure terminates；

(6d), by the test input vector in the test sample after normalized in step (2) and Test target vector inputs the BP neutral net trained respectively and is predicted drawing prediction output, will be pre- Survey output result carries out renormalization process and obtains predicted value.

A kind of BP neutral net photovoltaic plant based on genetic algorithm the most according to claim 1 Generated energy Forecasting Methodology, it is characterised in that described in step (3), the nodes of hidden layer is 4.

A kind of BP neutral net photovoltaic plant based on genetic algorithm the most according to claim 2 Generated energy Forecasting Methodology, it is characterised in that network initial weight threshold length in described step (5a) Being 33, and weights are 28, threshold value is 5, then network weight threshold value can be considered as i-th particle Vector, X_i=(x_i1,x_i2,…,x_iD), D is vector dimension, is weight threshold length 33.

A kind of BP neutral net photovoltaic plant based on genetic algorithm the most according to claim 1 Generated energy Forecasting Methodology, it is characterised in that the concrete steps that variation described in step (5e) processes are such as Under:

A (), size particle all to population according to adaptive value are ranked up, take out adaptive value best M particle, and except optimal particle；

B () produces corresponding m and is positioned at random number r before [0,1]_i(1≤i≤m), takes 0.1≤p_m≤ 0., if 3 r_i＜ p_m, produce new position the most as the following formula:

$x_{id}^{i+1}=x_{id}^i(1+0.5\mathrm{\η})$

In formula:Be respectively variation before and after position, η be obey Gauss (0,1) be distributed with Machine variable；

C (), after iteration reaches predetermined step number, can obtain optimal solution P_(gbest), use following public affairs Formula is evolved, and search there may be more excellent solution, and constantly circulation is until upper limit iterative steps, so far Iteration terminates:

ΔP_(n+1)=m Δ P_n+(1-m)gP_n

P_(n+1)=P_n+ΔP_(n+1)

In formula: n represents the particle vector when the n-th generation, g is the random number between [0,0.1], m=0.5, Δ p represents current neighborhood scope, for random value, represents in current optimal solution P_(gbest)Neighbouring search is more excellent The process solved；

Particle P_(gbest)From the n-th generation to the (n+1)th generation, if current P_(n+1)Fitness value be better than P_(n), Then use P_(n+1)Replace P_(n)Otherwise, then keep P_(n)Being worth constant, being so circulated until reaching Till big iterations.

A kind of BP neutral net photovoltaic plant based on genetic algorithm the most according to claim 1 Generated energy Forecasting Methodology, it is characterised in that in described step (6a), front 24 real numbers of optimum individual depend on The secondary weights for input layer to hidden layer, the 25 to 28th is followed successively by the input layer threshold value to hidden layer, 29 to 32nd is followed successively by hidden layer to output layer weights, the 33rd for hidden layer to output layer threshold value.

A kind of BP neutral net photovoltaic plant based on genetic algorithm the most according to claim 1 Generated energy Forecasting Methodology, it is characterised in that the test input vector in described step (6d) is the 12nd It history generated output and the solar irradiance of the 12nd day corresponding future time node, instantaneous wind speed, Backboard temperature, environment temperature, ambient humidity, test target vector be the 12nd day real-time generated output, Prediction is output as the prediction generated output of the 12nd day.