CN104376388A

CN104376388A - Ultra-short period wind power prediction method based on wind speed factor control model

Info

Publication number: CN104376388A
Application number: CN201410745990.1A
Authority: CN
Inventors: 于炳霞; 谭志萍; 陈梅; 陈志宝; 程序; 周海; 丁杰; 崔方; 王知嘉; 曹潇; 丁宇宇; 周强; 丁煌; 朱想
Original assignee: State Grid Corp of China SGCC; China Electric Power Research Institute Co Ltd CEPRI
Current assignee: State Grid Corp of China SGCC; China Electric Power Research Institute Co Ltd CEPRI
Priority date: 2014-12-08
Filing date: 2014-12-08
Publication date: 2015-02-25

Abstract

The invention provides an ultra-short period wind power prediction method based on a wind speed factor control model. The ultra-short period wind power prediction method includes performing an ultra-short period prediction according to an ARIMA (autoregressive integrated moving average) model; performing the ultra-short period power prediction according to a BP (back propagation) neutral network model; completing the ultra-short period wind power prediction based on the wind speed factor control model. In this way, at the premise that the ARIMA model and the BP neutral network model are evaluated, the ARIMA model and the BP neutral network model can be switched according to the running environment, so that accuracy of the ultra-short wind power prediction is improved.

Description

Wind power ultra-short-term power prediction method based on wind speed factor control model

Technical Field

The invention belongs to the technical field of wind power prediction, and particularly relates to a wind power ultra-short-term power prediction method based on a wind speed factor control model.

Background

The wind power ultra-short term prediction technology mainly aims at predicting wind power with time resolution of 15 minutes of 0-4 hours in the future, accurate ultra-short term prediction can provide guarantee for AGC control of a power grid, real-time adjustment of a power grid dispatching plan and arrangement of reserve capacity are facilitated, and safety, stability and economic operation of a system are improved.

At present, a single model for ultra-short-term prediction has relatively deep research, and research results of the single model are applied to provincial dispatching departments and stations of the large network. The invention patent with application number 201110388041.9 provides a wind power ultra-short-term prediction method, which comprises the steps of firstly, acquiring wind speed, wind direction and wind power data of a wind power plant to form a sample set; then, carrying out data preprocessing on the sample set; then, reducing the dimension of the preprocessed sample set by using a depth automatic encoder network; and finally, training a regression model of the relevance vector machine by using the reduced-dimension sample set, and predicting the ultra-short-term wind power by using the trained regression model of the relevance vector machine. The support vector machine method in the invention is suitable for solving the conditions of small samples and nonlinearity, the neural network model is favorable for solving the condition of a large sample training set, and the application condition determines that the accuracy of ultra-short-term power prediction can be reduced by a single model under certain conditions.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides a wind power ultra-short-term power prediction method based on a wind speed factor control model, which can switch an ARIMA model and a BP neural network model aiming at the operating environment on the basis of evaluating the two models, thereby improving the accuracy of wind power ultra-short-term power prediction.

In order to achieve the purpose of the invention, the invention adopts the following technical scheme:

the invention provides a wind power ultra-short-term power prediction method based on a wind speed factor control model, which comprises the following steps:

step 1: performing ultra-short-term power prediction by using an ARIMA model;

step 2: carrying out ultra-short term power prediction by using a BP neural network model;

and step 3: and wind power ultra-short-term power prediction is completed through a wind speed factor control model.

The step 1 specifically comprises the following steps:

step 1-1: differentiating the measured power by adopting 96 points in the previous 24 hours to obtain a stable time sequence;

step 1-2: and establishing an ARIMA model to obtain the ultra-short-term power prediction result of 16 points in the future 4 hours.

The step 1-1 specifically comprises the following steps:

step-1-1: let Δ x_tMeasured power x at time t_tMeasured power x at time t-1_t-1Is expressed as:

Δx_t＝x_t-x_t-1＝x_t-Lx_t＝(1-L)x_t

thus, there are:

wherein, L is a lag operator, and d is the difference times corresponding to the stable time sequence;

step 1-1-2: judging ARIMA order parameters according to the akage information quantity criterion, wherein the ARIMA order parameters comprise an autoregressive term number p, a moving average term number q and difference times d corresponding to a stationary time sequence;

step 1-1-3: the stationary time series is represented as:

w_t＝Δ^dx_t＝(1-L)^dx_t

wherein, w_tRepresenting a stationary time series.

The step 1-2 specifically comprises the following steps:

step 1-2-1: establishing an ARIMA model, which can be expressed as:

w_t＝φ₁w_t-1+φ₂w_t-2+...+φ_iw_t-i+...+φ_pw_t-p++u_t+θ₁u_t-1+θ₂u_t-2+...+θ_ju_t-j+...+θ_qu_t-q

wherein, is a constant term; u. of_tWhite noise is represented and follows a standard normal distribution; phi is a_iIs a coefficient of the auto-regressive term, i ═ 1, 2.., p; theta_jIs the coefficient of the moving average term, j ═ 1, 2.

Step 1-2-2: using maximum likelihood method to phi_i、θ_jAnd estimating respectively to make parameter set theta ═ phi_i,θ_jAnd constructing a likelihood function L (theta, sigma)²) The following were used:

wherein, sigma is a standard deviation, and T is the number of learning samples;

for L (theta, sigma)²) Taking natural logarithm to obtain l (theta, sigma)²) It is expressed as:

thus, the estimate of ΘExpressed as:

<math> <mrow> <mover> <mi>Θ</mi> <mo>^</mo> </mover> <mo>=</mo> <mi>arg</mi> <munder> <mi>max</mi> <mrow> <mi>Θ</mi> <mo>&Element;</mo> <msup> <mi>R</mi> <mrow> <mi>p</mi> <mo>+</mo> <mi>q</mi> <mo>+</mo> <mn>1</mn> </mrow> </msup> </mrow> </munder> <mi>l</mi> <mrow> <mo>(</mo> <mi>Θ</mi> <mo>,</mo> <msup> <mi>σ</mi> <mn>2</mn> </msup> <mo>)</mo> </mrow> </mrow> </math>

thereby obtaining phi_i、θ_jAnd respective estimated valuesAnd

thus, w_tIs estimated value ofCan be expressed as:

by passingThe expressions are sequentially deduced, so that the stable time sequence of 16 points in the future 4 hours can be calculated, and the power prediction result in the future 4 hours can be obtained.

The step 2 specifically comprises the following steps:

step 2-1: inputting historical wind speed and predicted power results as learning samples, repeatedly adjusting and training the weight and the deviation of the network by using a back propagation algorithm to enable the output vector to be as close to the expected vector as possible, finishing training when the sum of squares of errors of the output layer of the network is smaller than a specified value of the error, and storing the weight and the deviation of the network;

step 2-2: and inputting the future predicted wind speed value from the numerical weather forecast into a BP neural network model to obtain an ultra-short-term power prediction result within 4 hours in the future.

The step 3 specifically comprises the following steps:

step 3-1: respectively calculating the error e between the ultra-short-term power prediction result obtained by the ARIMA model and the actual measurement power value obtained by the BP neural network model₁(t)、e₂(t) having:

e_{1} (t) = P_{Ft} - P_{P_{1} t}

e_{2} (t) = P_{Ft} - P_{P_{2} t}

wherein,for ultra-short term power prediction results obtained by an ARIMA model,for obtaining ultra-short term power prediction results through BP neural network model, P_FtIs the measured power value;

step 3-2: e of 96 points in 24 hours in the previous day by utilizing BP neural network model₁(t)、e₂(t) training with a historical predicted wind speed value w to obtain a wind speed factor control model;

step 3-3: and switching the ARIMA model and the BP neural network model in real time by using the wind speed factor control model to obtain a final ultra-short-term power prediction result.

In the step 3-2, the historical predicted wind speed value w is taken as an input, and respective controllers of the ARIMA model and the BP neural network model are assumed to be C₁And C₂The corresponding control function is expressed as:

C₁(t)＝I(w,t)

C₂(t)＝1-C₁(t)

wherein I (w, t) is a function of t and w, and the change of the function with time and wind speed is 0 or 1, so C₁(t),C₂(t)∈{0,1}；

The final error e of the ultra-short term power prediction based on the wind speed factor control model is then expressed as:

e＝C₁(t)e₁(t)+C₂(t)e₂(t)

e of 96 points in 24 hours in the previous day by utilizing BP neural network model₁(t)、e₂(t) and w are trained, and the training target is set as the sum of squares of e is minimum, namely, the target function Σ e of the wind speed factor control model is obtained²And finally, obtaining the wind speed factor control model through training.

Compared with the prior art, the invention has the beneficial effects that:

(1) the problem of single model error uneven distribution is solved, and the unstable factor of the single model is overcome through the switching of the models.

(2) The method is flexible to apply, can realize online automatic model selection, does not need manual intervention, and can switch the models through online model error analysis.

Drawings

FIG. 1 is a flow chart of a wind power ultra-short-term power prediction method based on a wind speed factor control model.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings.

As shown in fig. 1, the invention provides a wind power ultra-short-term power prediction method based on a wind speed factor control model, which includes the following steps:

step 1: performing ultra-short-term power prediction by using an ARIMA model;

The step 1 specifically comprises the following steps:

The step 1-1 specifically comprises the following steps:

step-1-1: let Δ x_tMeasured power x at time t_tAnd time t-1Measured power x_t-1Is expressed as:

Δx_t＝x_t-x_t-1＝x_t-Lx_t＝(1-L)x_t

thus, there are:

step 1-1-3: the stationary time series is represented as:

w_t＝Δ^dx_t＝(1-L)^dx_t

wherein, w_tRepresenting a stationary time series.

The step 1-2 specifically comprises the following steps:

step 1-2-1: establishing an ARIMA model, which can be expressed as:

thus, the estimate of ΘExpressed as:

thereby obtaining phi_i、θ_jAnd respective estimated valuesAnd

thus, w_tIs estimated value ofCan be expressed as:

The step 2 specifically comprises the following steps:

The step 3 specifically comprises the following steps:

e_{1} (t) = P_{Ft} - P_{P_{1} t}

e_{2} (t) = P_{Ft} - P_{P_{2} t}

wherein,for ultra-short term power prediction results obtained by an ARIMA model,to pass through BP nerveThe network model obtains the prediction result of the ultra-short term power, P_FtIs the measured power value;

C₁(t)＝I(w,t)

C₂(t)＝1-C₁(t)

e＝C₁(t)e₁(t)+C₂(t)e₂(t)

Finally, it should be noted that: the above embodiments are only intended to illustrate the technical solution of the present invention and not to limit the same, and a person of ordinary skill in the art can make modifications or equivalents to the specific embodiments of the present invention with reference to the above embodiments, and such modifications or equivalents without departing from the spirit and scope of the present invention are within the scope of the claims of the present invention as set forth in the claims.

Claims

1. A wind power ultra-short-term power prediction method based on a wind speed factor control model is characterized by comprising the following steps: the method comprises the following steps:

step 1: performing ultra-short-term power prediction by using an ARIMA model;

2. The wind power ultra-short-term power prediction method based on the wind speed factor control model according to claim 1, characterized in that: the step 1 specifically comprises the following steps:

3. The wind power ultra-short-term power prediction method based on the wind speed factor control model according to claim 2, characterized in that: the step 1-1 specifically comprises the following steps:

Δx_t＝x_t-x_t-1＝x_t-Lx_t＝(1-L)x_t

thus, there are:

step 1-1-3: the stationary time series is represented as:

w_t＝Δ^dx_t＝(1-L)^dx_t

wherein, w_tRepresenting a stationary time series.

4. The wind power ultra-short-term power prediction method based on the wind speed factor control model according to claim 1, characterized in that: the step 1-2 specifically comprises the following steps:

step 1-2-1: establishing an ARIMA model, which can be expressed as:

thus, the estimate of ΘExpressed as:

thereby obtaining phi_i、θ_jAnd respective estimated valuesAnd

thus, w_tIs estimated value ofCan be expressed as:

by passingThe expression of (A) is sequentially deduced, namely the calculation can be carried out within 4 hours in the futureAnd a 16-point stationary time sequence is obtained, so that a power prediction result in 4 hours in the future is obtained.

5. The wind power ultra-short-term power prediction method based on the wind speed factor control model according to claim 1, characterized in that: the step 2 specifically comprises the following steps:

6. The wind power ultra-short-term power prediction method based on the wind speed factor control model according to claim 1, characterized in that: the step 3 specifically comprises the following steps:

e_{1} (t) = P_{Ft} - P_{P_{1} t}

e_{2} (t) = P_{Ft} - P_{P_{2} t}

7. The wind power ultra-short-term power prediction method based on the wind speed factor control model according to claim 8, characterized in that: in the step 3-2, the historical predicted wind speed value w is taken as an input, and respective controllers of the ARIMA model and the BP neural network model are assumed to be C₁And C₂The corresponding control function is expressed as:

C₁(t)＝I(w,t)

C₂(t)＝1-C₁(t)

e＝C₁(t)e₁(t)+C₂(t)e₂(t)