CN104298859A - Multi-interval-prediction-result weighting integration method - Google Patents

Abstract

The invention discloses a multi-interval-prediction-result weighting integration method. The method specifically comprises the steps of (1) acquiring multiple interval prediction results of one prediction object, wherein each interval prediction result contains a prediction expectation and the variance of the prediction expectation; (2) setting different weights for different interval prediction results; (3) calculating the expectation and variance of a comprehensive prediction result with weight considered; (4) calculating a new weighted interval prediction result according to the expectation and variance of the comprehensive prediction result obtained in the step (3), and ending calculation. According to the method, aiming at the multiple interval prediction results acquired through different approaches and different prediction methods, a new interval prediction result is formed according to the probability distribution information of multiple prediction results preserved currently or user tendency, and then limitation of a single prediction algorithm interval prediction result is avoided and practicality of prediction is improved greatly.

Description

The weighting integration method predicted the outcome between a kind of multi-region

Technical field

The present invention relates to the weighting integration method predicted the outcome between a kind of multi-region, belong to forecast analysis technical field.

Background technology

Classical forecast is often based on determining that numerical value carries out analysis modeling, predicting the outcome of obtaining also is determine numerical value, along with Forecasting Methodology is on the increase, there is scholar to propose to solve many covers method combined of determining to predict the outcome, be used for obtaining predicting the outcome of a set of relatively better final fixed number value really.

But determine that the prediction of numerical value is unfavorable for that subsequent applications considers to predict the outcome the risk of error, for example power industry is in order to ensure the coulomb balance that generates electricity, not only consider to meet the load prediction determined during establishment generation schedule, also will ensure that power load still has regulating power under certain surging condition simultaneously.Therefore researcher proposes multiple interval prediction method, and the expectation that not only providing predicts the outcome determines provides the waving interval predicted the outcome under certain confidence level simultaneously, makes the information that predicts the outcome more comprehensive.

Along with new interval prediction method constantly proposes, many cover interval prediction results will be formed, urgently by a set of for these TB control newly, can to predict the outcome the interval prediction result of information by the many covers of concentrated expression.

Summary of the invention

For the deficiency that prior art exists, the object of the invention is to provide the weighting integration method predicted the outcome between a kind of multi-region, for the many covers interval prediction result obtained by different approaches, different Forecasting Methodology, retaining under the probability distribution information or the tendentious condition of consideration integration of overlapping more and predicting the outcome, define a set of new interval prediction result, thus avoid the limitation of Individual forecast algorithm interval prediction result, greatly improve the practical level of prediction work.

To achieve these goals, the present invention realizes by the following technical solutions:

The weighting integration method predicted the outcome between a kind of multi-region of the present invention, specifically comprises following step:

(1) obtain many covers interval prediction result of same forecasting object, comprise prediction in each interval prediction result and expect and variance;

(2) different weight is arranged to different interval prediction result;

(3) expectation and the variance of the integrated forecasting result considering weight is calculated;

(4) according to expectation and the variance of integrated forecasting result described in step (3), calculate new weighting interval prediction result, calculate and terminate.

In step (1), obtain many cover interval prediction results by statistics, regression analysis.

In step (2), according to overlapping the probability distribution information that predicts the outcome or the tendentiousness according to user, different weight is arranged to different interval prediction result more.

In step (3), the expectation of the integrated forecasting result of described consideration weight and the computing method of variance as follows:

$\mathrm{\μ}=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i{\mathrm{\μ}}_i}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i}$

$\mathrm{\σ}=\sqrt{\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left(a_i{\mathrm{\σ}}_i^2\right)}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i}+\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left(a_i\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left(a_j^2{(u_i-u_j)}^2\right)\right)}{{\left(\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i\right)}^3}}$

Wherein, N is the tricks predicted the outcome, a _ibe the weight coefficient of the i-th cover result, μ _i, σ _ibe respectively the expectation and variance of the i-th cover interval prediction result, a _jfor the weight coefficient of jth cover result, μ _j, σ _jbe respectively the expectation and variance of jth cover result, μ, σ are respectively the expectation and variance of integrated forecasting result.

In step (4), the computing method of described new weighting interval prediction result are as follows:

According to the confidence level α in integrated results interval, look into standardized normal distribution table and obtain t _α/2, then new forecast interval is [μ-t _α/2σ, μ+t _α/2σ].

Above-mentioned confidence level α gets 0.90 or 0.95.

Beneficial effect of the present invention is as follows:

1) the present invention can be a set of new interval prediction result by overlapping interval prediction TB control more, subsequent applications does not need again in the face of many covers predict the outcome, overlap the integration predicted the outcome owing to comprehensively employing more simultaneously, reduce the forecasting risk of Individual forecast method.

2) forecast interval that the present invention can obtain different prediction algorithm in integration process arranges different weights, this weight can be determined according to the precision of prediction (probability distribution) of this algorithm in history, the Forecasting Methodology of precision high (probability is large) uses larger weight, and final precision of prediction can be increased; This weight can also be determined according to the tendentiousness to algorithms of different of prognosticator.

3) the present invention is the principle design based on the probability density weighted stacking that predicts the outcome more, for the computing method obtained under normal distribution assumed condition, its validity is through engineering verification, and computation process extends to other field, is weighted integration for many cover compartmental results.

Accompanying drawing explanation

Fig. 1 is the weighting integration method principle schematic predicted the outcome between a kind of multi-region of the present invention.

Embodiment

The technological means realized for making the present invention, creation characteristic, reaching object and effect is easy to understand, below in conjunction with embodiment, setting forth the present invention further.

The present embodiment contain adopt the inventive method electric system a few days ago 96 period loads (15 minutes intervals) overlap the conformity calculation process of forecast interval result more, its feature, object and advantage can be found out from the explanation of embodiment.

Dispatching of power netwoks department predicts electrical network power load next day every day, obtains many cover interval prediction results by methods such as statistics, regretional analyses, based on predicting the outcome the balancing work carrying out electrical network hair electricity, and establishment generation schedule.

In planning procedure, use is single cover load prediction compartmental results, instead of multiple interval, needs to integrate predicting the outcome between multi-region for this reason.

See Fig. 1, the weighting integration method predicted the outcome between a kind of multi-region, specifically comprises following step:

(1) obtain many cover interval prediction results, in result, comprise the expectation and covariance information that respectively predict the outcome.

(2) according to the tendentiousness of user, weight proportion is arranged to often kind of forecast interval.96 periods can identical proportion, also can distinguish to some extent.

(3) for 96 periods interval prediction result, comprehensive expectation and covariance information is obtained according to formula pointwise weighting.

Wherein, consider that the computing method of the expectation of integrated forecasting result of weight and variance are as follows:

$\mathrm{\μ}=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i{\mathrm{\μ}}_i}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i}$

$\mathrm{\σ}=\sqrt{\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left(a_i{\mathrm{\σ}}_i^2\right)}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i}+\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left(a_i\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left(a_j^2{(u_i-u_j)}^2\right)\right)}{{\left(\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i\right)}^3}}$

Wherein, N is the tricks predicted the outcome, a _ibe the weight coefficient of the i-th cover result, μ _i, σ _ibe respectively the expectation and variance of the i-th cover interval prediction result, a _jfor the weight coefficient of jth cover result, μ _j, σ _jbe respectively the expectation and variance of jth cover result, μ, σ are respectively the expectation and variance of integrated forecasting result.

(4) according to comprehensive expectation, variance, calculate between new complex zone, calculate and terminate.

Wherein, the computing method of new weighting interval prediction result are as follows:

According to the confidence level α (be known quantity, confidence level α gets 0.90 or 0.95) in integrated results interval, look into standardized normal distribution table and obtain t _α/2, then new forecast interval is [μ-t _α/2σ, μ+t _α/2σ].

More than show and describe ultimate principle of the present invention and principal character and advantage of the present invention.The technician of the industry should understand; the present invention is not restricted to the described embodiments; what describe in above-described embodiment and instructions just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications, and these changes and improvements all fall in the claimed scope of the invention.Application claims protection domain is defined by appending claims and equivalent thereof.

Claims (6)

1. the weighting integration method predicted the outcome between multi-region, is characterized in that, specifically comprises following step:

(1) obtain many covers interval prediction result of same forecasting object, comprise prediction in each interval prediction result and expect and variance;

(2) different weight is arranged to different interval prediction result;

(3) expectation and the variance of the integrated forecasting result considering weight is calculated;

(4) according to expectation and the variance of integrated forecasting result described in step (3), calculate new weighting interval prediction result, calculate and terminate.

2. the weighting integration method predicted the outcome between multi-region according to claim 1, is characterized in that,

In step (1), obtain many cover interval prediction results by statistics, regression analysis.

3. the weighting integration method predicted the outcome between multi-region according to claim 1, is characterized in that,

In step (2), according to overlapping the probability distribution information that predicts the outcome or the tendentiousness according to user, different weight is arranged to different interval prediction result more.

4. the weighting integration method predicted the outcome between multi-region according to claim 1, is characterized in that,

In step (3), the expectation of the integrated forecasting result of described consideration weight and the computing method of variance as follows:

$\mathrm{\μ}=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i{\mathrm{\μ}}_i}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i}$

$\mathrm{\σ}=\sqrt{\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left(a_i{\mathrm{\σ}}_i^2\right)}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i}+\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left(a_i\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\left(a_j^2{(u_i-u_j)}^2\right)\right)}{{\left(\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{a}_i\right)}^3}}$

Wherein, N is the tricks predicted the outcome, a _ibe the weight coefficient of the i-th cover result, μ _i, σ _ibe respectively the expectation and variance of the i-th cover interval prediction result, a _jfor the weight coefficient of jth cover result, μ _j, σ _jbe respectively _jthe expectation and variance of cover result, μ, σ are respectively the expectation and variance of integrated forecasting result.

5. the weighting integration method predicted the outcome between multi-region according to claim 4, is characterized in that,

In step (4), the computing method of described new weighting interval prediction result are as follows:

According to the confidence level α in integrated results interval, look into standardized normal distribution table and obtain t _α/2, then new forecast interval is [μ-t _α/2σ, μ+t _α/2σ].

6. the weighting integration method predicted the outcome between multi-region according to claim 5, is characterized in that,

Described confidence level α gets 0.90 or 0.95.