CN106447091A - Regional meteorological condition similarity-based large power network load prediction method - Google Patents

Abstract

The invention discloses a regional meteorological condition similarity-based large power network load prediction method. The method comprises the steps of firstly obtaining load prediction values of regions in a to-be-predicted day and a load and meteorological history data of a sample space recently; aggregating the regions with similar meteorological conditions according to a probability distance-based synchronous back-substitution elimination technology, dividing an aggregated region into q sub-regions with relatively large meteorological condition difference, then calculating an average proportional coefficient of the q sub-regions at a t moment point in a historical data sample space, and predicting proportional coefficients of the sub-regions at the same t moment point in the to-be-predicted day; and predicting a whole network system load at the t moment point by using the q sub-regions, building an optimal comprehensive model at the t moment point for q different prediction values, performing solving to obtain a final prediction result of the whole network system load at the t moment point, and building optimal comprehensive models for T moment points in the whole to-be-predicted day, thereby obtaining a whole day load prediction sequence. According to the method, the short-term load prediction accuracy of a power system can be improved.

Description

Large power grid load prediction method based on regional meteorological condition similarity

Technical Field

The invention relates to the technical field related to load prediction of an electric power system, in particular to a large power grid load prediction method based on regional meteorological condition similarity, which is used for short-term load prediction of the electric power system.

Background

In order to ensure the dynamic balance of the generating power and the load power of the power system, scientific prediction must be made on the load of the power system. The load prediction is an important work of a dispatching center and a power grid development planning department, and the result of the load prediction has important guiding value on the aspects of power grid operation, control, dispatching, planning, construction and the like, and is the basis of power grid scientific development and scientific dispatching.

The load prediction technology level is improved, the planned power utilization management is facilitated, the reasonable arrangement of the power grid operation mode and the unit maintenance plan is facilitated, the coal saving, the oil saving and the power generation cost reduction are facilitated, and the economic benefit and the social benefit of the power system are facilitated to be improved. Therefore, load prediction has become an important content for realizing modernization of power system management.

The current method for predicting the load of the whole network system by using the regional load is a 'subnet accumulation method', and the prediction process of the 'subnet accumulation method' is shown in fig. 1. The prediction can be basically divided into the following three steps:

1. various factors influencing short-term load prediction are fully considered in each area, a proper prediction method is selected, and then short-term load prediction is carried out on each area according to historical data of each area; and obtaining 96-point load prediction results of each subnet.

2. And summarizing the load prediction results of the regions, and accumulating the 96-point prediction data of each region to obtain the accumulated sum of each time point.

3. And calculating the power consumption and the network loss of 96 points of the day to be predicted, and correcting the accumulated sum to obtain a final power network load prediction result of the Anhui.

Because the method needs to predict the subnets of each area, and for short-term load prediction, the load stability of each area is different, the prediction difficulty is different, and meanwhile, factory power consumption and network loss data also need to be predicted, so when the subnets are accumulated by using the predicted loads of all the areas, the accuracy effect of the load prediction of the whole network system is possibly not ideal.

In order to solve the above problems, the applicant of the present invention proposes a method for predicting a load of a whole grid based on comprehensive evaluation of predicted values of regional loads in an invention patent application with an application number of CN201310648023.9, which improves accuracy of short-term load prediction of a power system to a certain extent. However, when the above method is used for dividing regions, the regions are divided by administrative regions where geographic locations are located, but the change trends of meteorological conditions of the administrative regions are different, especially in summer when the meteorological conditions change drastically, so that the ratio of the load of each administrative region to the load of the whole network may change greatly in consecutive days, which is not favorable for predicting the load proportionality coefficient of each administrative region to the load of the whole network; meanwhile, if there are more administrative areas under a large power grid, the calculation amount of the optimal comprehensive model is large. The above problems can be solved by aggregating each administrative area using meteorological conditions. In addition, the overall load cardinality of the meteorological area is large, the load stability of each administrative area is high, and the whole network load prediction is facilitated.

Disclosure of Invention

The invention provides a large power grid load prediction method based on regional meteorological condition similarity, aiming at the problem that when the large power grid load is predicted based on regional load prediction value comprehensive evaluation, the large power grid is divided into regions by adopting administrative regions where geographic positions are located, so that the short-term load prediction accuracy of a power system is improved. Meanwhile, the method can also select part of regional load prediction values to predict the whole network load, and can avoid the influence on the whole network load prediction caused by the fact that some regional load prediction specializes to report regional prediction results in time.

The technical problem to be solved by the invention can be realized by the following technical scheme:

a large power grid load prediction method based on regional meteorological condition similarity is characterized by comprising the following steps:

(1) acquiring load predicted values of M administrative areas reported by the load predicted value on a day to be predicted; acquiring historical data in a recent sample period as a historical data sample space, wherein load data in the historical data are actual loads of the whole network and M administrative regions;

(2) acquiring at least 2 pieces of historical meteorological condition data in M administrative areas in a sample period, and dividing the M administrative areas into q meteorological areas with larger meteorological condition difference through the historical meteorological condition data;

(3) dynamically predicting the proportionality coefficients of q meteorological areas at the same time point t of the day to be predicted by an exponential smoothing method to obtain a proportionality coefficient matrix C of the q meteorological areas at the time point t_t；

(4) Respectively predicting the load of the whole network system at the time point t by using the selected q regions to obtain q different prediction results;

(5) establishing an optimal comprehensive model of the t time point for q different prediction results, calculating the optimal weight of each selected meteorological area, and obtaining the final prediction result L of the load of the whole network system of the t time point_all,t：

Wherein,for the optimal weight of the load of the whole network system predicted by the meteorological area k at the time point t,the load of the whole network system at the t time point predicted by the kth meteorological area;

(6) respectively establishing an optimal comprehensive model for T time points of the day to be predicted all day to obtain an all-day load prediction sequence (L)_all,1,L_all,2,…,L_all,T) Predicting a sequence (L) with said all-day load_all,1,L_all,2,…,L_all,T) And (4) predicting the result of the load of the whole network.

In the invention, the at least 2 pieces of historical meteorological condition data are 2 or any combination of more than 2 of the highest temperature, the lowest temperature, the average temperature, the human body comfort level, the weather type, the wind speed, the temperature-humidity index and the cold index of the administrative region in one sample time period.

The at least 2 historical meteorological condition data are the highest temperature, the lowest temperature and the average temperature of the administrative regions in a sample period, and a meteorological data matrix X is formed according to the historical meteorological condition data of the M administrative regions:

X＝(X₁,X₂,…X_i,…,X_M)

wherein,

X_ifor the ith administrative region within a sample periodA set of n-day weather condition data,

the highest temperature of the ith administrative region within a sample period of n days;

the minimum temperature of the ith administrative area for n days within a sample period;

average temperature for the ith administrative area over a sample period of n days.

In the invention, a synchronous back substitution elimination technology based on probability distance is adopted to aggregate the matrix X to form q meteorological areas with larger meteorological condition difference:

first, M column vectors in a matrix X are assigned a probability p of occurrence_sSet S represents the first administrative region, let ξ be 1/M (S is 1,2, …, M)_sRepresenting the s-th administrative region, DT, in the matrix X_s,s'Represents the distance between the administrative region s and the administrative region s ', and has a value of vector 2 norm between the administrative region s and the administrative region s', and the vector DA is set to [1,2, …, M ═ 1]^TAnd DB is a square matrix of M × M with all elements being 0.

The basic steps of synchronous back-substitution elimination are as follows:

1) calculating the distance DT between each pair of administrative regions_s,s'，DT_s,s'＝||ξ_s,ξ_s'||。

2) For each administrative region k, finding out the administrative region r with the shortest distance from the administrative region k, namely DT_k(r)＝minDT_k,s'。

3) Calculating PD_k(r)＝p_k*DT_k(r), k ∈ S, finding the administrative region index h in k, so that PD is_h＝minPD_k,k∈S。

4) Let S equal S- ξ_DA(h)And p is_r＝p_r+p_h. DB (h, h) ═ da (r), and da (h) is empty.

5) Repeat 2) -4) until the number of remaining administration areas is q.

The DA only has q elements left, and corresponds to the square matrix DB (DA (i), DA (i)) 0, i ═ 1,2, … q, that is, q meteorological areas with large meteorological condition differences need to be reserved.

The administrative region searching steps of the respective aggregation of q meteorological regions with large meteorological condition differences in the DA are as follows:

1) let i be 1 and P be da (i).

2) M satisfying the requirement is found so that DB (m, m) becomes P, and then P becomes [ P, m ] for the administrative area included in the meteorological area i.

3) Finding l that satisfies the requirement such that DB (l, l) ═ m, P ═ P, l ], the search process until the diagonal elements of DB are not equal to l;

4) i +1, P da (i), repeat 2) -4) until i q.

In the present invention, in the step (1), the historical data is preprocessed as follows:

order: l (d, t) is the load value at the time t on day d, L (d, t)₁) And L (d, t)₂) Two times t adjacent to the time t on day d₁、t₂Load value of L (d)₁T) and L (d)₂T) is the load value at the time point t on two days adjacent to d;

a) handling of missing data

If the load value L (d, t) at the time t on day d is missing, L (d, t) is obtained by using the formula (1):

L(d,t)＝αL(d,t₁)+αL(d,t₂)+βL(d₁,t)+βL(d₂,t) (1)

in formula (1), α and β are coefficients, α > β, α + β ═ 1;

b) processing of dead-end data

Defining as the allowable deviation rate of load, rho (d, t) as the actual deviation rate at the time point t on day d, when rho (d, t) is greater than or equal to, judging L (d, t) as bad data, using for bad dataThe substitution is carried out as follows:

in the present invention, the step (3) is represented as:

C_t＝(C_1,t,C_2,t,…,C_q,t)

wherein,

C_1,tthe scale factor of the 1 st meteorological area at the time point t of the day to be predicted;

C_2,tthe scale coefficient of the 2 nd meteorological area at the time point t of the day to be predicted;

C_q,tand the proportionality coefficient of the qth meteorological area at the time point t of the day to be predicted.

The exponential smoothing method in the step (3) is that the establishment of an exponential smoothing model is as follows:

in formula (3):indicating a meteorological area i inThe time t is the predicted value of the load proportion of the whole network system,representing the actual value of the load proportion of the meteorological area i in the previous j days in the whole network at the time t; n is the number of days in a sample period; lambda [ alpha ]_jDenotes a weight coefficient, λ_j＝λ(1-λ)^j-1λ is a constant, and 0 < λ < 1.

In the present invention, in the step (4), q different prediction results are obtained, and are represented as:

wherein,

the load of the whole network system at the t time point predicted by utilizing the 1 st meteorological area;

the load of the whole network system at the t time point predicted by utilizing the 2 nd meteorological area;

is the total network system load of the t time point predicted by utilizing the qth meteorological area, andL_q,tload prediction value of the qth meteorological area at time point t, C_q,tAnd predicting the proportionality coefficient of the qth meteorological area at the time point t.

In the invention, the optimal comprehensive model in the step (5) is established according to the following method:

an objective function of the load predicted value of the whole network system at the time point t is represented by an equation (4), and equations (5) to (6) are constraint conditions of the objective function:

in formula (4):showing the predicted value L of the load of the whole network system predicted by the area k at the time point t of the j day_all,t,jRepresenting the actual load value of the whole network system at the time point t on the jth day;

calculating the optimal weight of the selected q meteorological areas at the time point t by the formula (4), the formula (5) and the formula (6)Then, again according toAnd weighting to obtain the total network system load predicted value of the t time point.

The large power grid load prediction method based on the regional meteorological condition similarity has the following beneficial effects:

1. the method reduces the influence of the area which has larger load fluctuation and is difficult to predict in the cumulative method of the subnet in the prior art on the load prediction of the whole network; the prediction of power consumption of the power plant and power grid loss is avoided.

2. The load prediction values of all areas need to be known in advance in the subnet cumulative addition, and the method only selects part of the load prediction values of the areas to predict the load of the whole network, thereby avoiding that the load prediction of the whole network of the provincial and control center is influenced when some area load prediction specializes to report the area prediction result in time.

3. The areas are divided without simply adopting administrative areas, the administrative areas are processed based on the meteorological condition similarity, the influence caused by the fact that the ratio of the load of each administrative area to the load of the whole network changes greatly in consecutive days when the meteorological conditions of each administrative area change greatly, particularly in summer with severe meteorological condition changes, is avoided, and the prediction of the load ratio coefficient of each meteorological area to the whole network is facilitated by dividing the meteorological areas; meanwhile, if there are more administrative areas under a large power grid, the calculation amount of the optimal comprehensive model is large. Aggregating the various administrative areas using meteorological conditions can solve this problem. In addition, the overall load cardinality of the meteorological area is large, the load stability of each administrative area is high, and the whole network load prediction is facilitated.

Drawings

The invention is further described below in conjunction with the appended drawings and the detailed description.

Fig. 1 is a flow chart of a subnet accumulation algorithm.

FIG. 2 is a flow chart of the method of the present invention.

Fig. 3 is a comparison graph of load prediction effect of a provincial power grid 2016, 5, 18 and 18.

FIG. 4 is a comparison graph of the final all-day load prediction curve and the actual load curve solved by the optimal comprehensive model.

Detailed Description

In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further described below by combining the specific drawings.

The method is mainly characterized in that the problem of region division of administrative regions with geographic positions is found by analyzing the basic operation mode situation of prediction of the large power grid load based on the regional load prediction value, and the large power grid load prediction method based on the regional meteorological condition similarity is provided to improve the short-term load prediction accuracy of the power system.

Referring to fig. 2, the large power grid load prediction method based on the similarity of regional meteorological conditions includes the following steps:

the method includes the steps that historical data in a recent sample period (for example, n days) are obtained as a historical data sample space, load data in the historical data are actual loads and predicted loads of a whole network and M administrative regions in the whole network, wherein the administrative regions refer to regions divided by geographic positions, and for example, when the range of the whole network is a province, the administrative regions can be the administrative region division performed by city in the province.

For the historical data, preprocessing can be performed as follows:

order: l (d, t) is the load value at the time t on day d, L (d, t)₁) And L (d, t)₂) Two times t adjacent to the time t on day d₁、t₂Load value of L (d)₁T) and L (d)₂T) is the load value at the time point t on two days adjacent to d;

a) handling of missing data

If the load value L (d, t) at the time t on day d is missing, L (d, t) is obtained by using the formula (1):

L(d,t)＝αL(d,t₁)+αL(d,t₂)+βL(d₁,t)+βL(d₂,t) (1)

in formula (1), α and β are coefficients, α > β, α + β ═ 1;

b) processing of dead-end data

Defining as the allowable deviation rate of load, rho (d, t) as the actual deviation rate at the time point t on day d, when rho (d, t) is greater than or equal to, judging L (d, t) as bad data, using for bad dataThe substitution is carried out as follows:

then, at least 2 pieces of historical meteorological condition data in M administrative areas in a sample time period are obtained, the historical meteorological condition data preferably adopt three pieces of historical meteorological condition data, namely the highest temperature, the lowest temperature and the average temperature of the administrative areas in the sample time period, of course, the historical meteorological condition data can also adopt human body comfort, weather types, wind speeds, temperature and humidity indexes, cold indexes and the like due to the requirement of improving the prediction accuracy, and the historical meteorological condition data which are not data in the weather types and the like can be converted into data through data processing. For convenience of illustration, the present embodiment is exemplified by three pieces of historical meteorological condition data, i.e., the highest temperature, the lowest temperature, and the average temperature of the administrative area.

Forming a meteorological data matrix X according to historical meteorological condition data of M administrative areas:

X＝(X₁,X₂,…X_i,…,X_M)

wherein,

X_ia set of n days of meteorological condition data for the ith administrative area within a sample period,

the highest temperature of the ith administrative region within a sample period of n days;

the minimum temperature of the ith administrative area for n days within a sample period;

average temperature for the ith administrative area over a sample period of n days.

Thus, the X matrix has a row vector number of 3 × n and a column vector number of M.

In the invention, a synchronous back-substitution elimination technology based on probability distance is adopted to aggregate the matrix X, and M administrative areas are divided into q meteorological areas with larger meteorological condition difference:

first, M column vectors in a matrix X are assigned a probability p of occurrence_sSet S represents the first administrative region, let ξ be 1/M (S is 1,2, …, M)_sRepresenting the s-th administrative region, DT, in the matrix X_s,s'Represents the distance between the administrative region s and the administrative region s ', and has a value of vector 2 norm between the administrative region s and the administrative region s', and the vector DA is set to [1,2, …, M ═ 1]^TAnd DB is a square matrix of M × M with all elements being 0.

The basic steps of synchronous back-substitution elimination are as follows:

1) calculating the distance DT between each pair of administrative regions_s,s'，DT_s,s'＝||ξ_s,ξ_s'||。

2) For each administrative region k, finding out the administrative region r with the shortest distance from the administrative region k, namely DT_k(r)＝minDT_k,s'。

3) Calculating PD_k(r)＝p_k*DT_k(r), k ∈ S, finding the administrative region index h in k, so that PD is_h＝minPD_k,k∈S。

4) Let S equal S- ξ_DA(h)And p is_r＝p_r+p_h. DB (h, h) ═ da (r), and da (h) is empty.

5) Repeat 2) -4) until the number of remaining administration areas is q.

The DA only has q elements left, and corresponds to the square matrix DB (DA (i), DA (i)) 0, i ═ 1,2, … q, that is, q meteorological areas with large meteorological condition differences need to be reserved.

The administrative region searching steps of the respective aggregation of q meteorological regions with large meteorological condition differences in the DA are as follows:

1) let i be 1 and P be da (i).

2) M satisfying the requirement is found so that DB (m, m) becomes P, and then P becomes [ P, m ] for the administrative area included in the meteorological area i.

3) Finding l that satisfies the requirement such that DB (l, l) ═ m, P ═ P, l ], the search process until the diagonal elements of DB are not equal to l;

4) i +1, P da (i), repeat 2) -4) until i q.

In fact, the essence of dividing M administrative areas into q meteorological areas with large meteorological condition differences is that the M administrative areas with small meteorological condition differences are aggregated (i.e. the administrative areas with small meteorological condition differences are regarded as the same meteorological area), and q meteorological areas are formed based on the meteorological condition differences, and each meteorological area includes one or several meteorological areas.

After the meteorological areas are obtained, calculating the average proportionality coefficients of the q meteorological areas at the time point tDynamically predicting the proportionality coefficients of q meteorological areas at the same time point t of the day to be predicted by an exponential smoothing method to obtain a proportionality coefficient matrix C of the q meteorological areas at the time point t_t(ii) a Expressed as:

C_t＝(C_1,t,C_2,t,…,C_q,t)

wherein,

C_1,tthe scale factor of the 1 st meteorological area at the time point t of the day to be predicted;

C_2,tthe scale coefficient of the 2 nd meteorological area at the time point t of the day to be predicted;

C_q,tand the proportionality coefficient of the qth meteorological area at the time point t of the day to be predicted.

The exponential smoothing method is that an exponential smoothing model is established as follows:

in formula (3):the predicted value of the meteorological area i occupying the load proportion of the whole network system at the time t is shown,representing the actual value of the load proportion of the meteorological area i in the previous j days in the whole network at the time t; n is the number of days in a sample period; lambda [ alpha ]_jDenotes a weight coefficient, λ_j＝λ(1-λ)^j-1λ is a constant, λ is more than 0 and less than 1, and λ is usually a constant between 0.7 and 0.9 to ensure that the weight of recent data is large and the weight of distant data is small.

Respectively predicting the load of the whole network system at the time point t by using the selected q regions to obtain q different prediction results, wherein the prediction results are expressed as:

wherein,

the load of the whole network system at the t time point predicted by utilizing the 1 st meteorological area;

the load of the whole network system at the t time point predicted by utilizing the 2 nd meteorological area;

is the total network system load of the t time point predicted by utilizing the qth meteorological area, andL_q,tload prediction value of the qth meteorological area at time point t, C_q,tAnd predicting the proportionality coefficient of the qth meteorological area at the time point t.

Establishing an optimal comprehensive model of the t time point for q different prediction results, calculating the optimal weight of each selected meteorological area, and obtaining the final prediction result L of the load of the whole network system of the t time point_all,t：

Wherein,for the optimal weight of the load of the whole network system predicted by the meteorological area k at the time point t,the load of the whole network system at the t time point predicted by the kth meteorological area;

the optimal comprehensive model is established according to the following method:

an objective function of the load predicted value of the whole network system at the time point t is represented by an equation (4), and equations (5) to (6) are constraint conditions of the objective function:

in formula (4):showing the predicted value L of the load of the whole network system predicted by the meteorological area k at the time point of the j day t_all,t,jAnd (4) representing the actual value of the load of the whole network system at the time point t on the j th day.

For the solution of the equations (5) to (6), firstly, the virtual prediction residual v of the predicted total network system load value of each meteorological area is defined_kjtVirtual prediction residual sum of squaresAnd the covariance of virtual prediction results of predicting the load of the whole network system in certain two meteorological areasThe following were used:

the objective functions of equations (4) - (6) at time t are transformed into the matrix form as follows:

wherein:

the formula (7) is a standard form of a quadratic programming problem, and the optimal weight of the selected q meteorological areas at the time point t is directly calculatedThen, again according toAnd weighting to obtain the total network system load predicted value of the t time point.

For the loads of the whole network system predicted by the load predicted values of the meteorological areas, different prediction effects are shown at different time points, so that the time points of the day to be predicted are treated differently, and comprehensive models are respectively established, so that the weights of the loads of the whole network system predicted by the load predicted values of the meteorological areas at the time points are different, and the prediction effects of the loads of the whole network system at the different time points are reflected. Respectively establishing an optimal comprehensive model for T time points of the day to be predicted all day to obtain the all day load forecastSequencing (L)_all,1,L_all,2,…,L_all,T) Predicting a sequence (L) with said all-day load_all,1,L_all,2,…,L_all,T) And (4) predicting the result of the load of the whole network.

The following is a detailed example to continue the process of the present invention:

in the method, the meteorological information of each administrative area is aggregated by adopting a synchronous back substitution elimination technology based on the probability distance, and q meteorological areas with larger meteorological condition difference are formed. Taking the prediction of the whole network load of a province as an example, the province comprises 16 cities (respectively represented by city 1, city 2, … and city 16, the city numbering criterion is that each city is numbered from north to south and from east to west), the 16 cities are divided into 6 meteorological areas with large meteorological condition difference according to a meteorological condition similarity judging method, and the specific implementation is carried out according to the following steps:

1. data reading: and selecting 2016, 5 and 18 days as days to be predicted, and respectively obtaining actual load data of the whole network and each city 30 working days before the predicted days, meteorological data of each city 30 days before and the predicted value of the load of the days to be predicted reported by each city.

2. Dividing weather-like regions: and aggregating the weather information of 16 cities by adopting a synchronous back-substitution elimination technology based on probability distance, and dividing the 16 cities into 6 weather areas with large weather condition difference.

Dividing regions	Including the city of land
		Weather region 1	City of land 1, city of land 2, city of land 4
Weather region 2	City 3, city 5, city 7
		Weather region 3	6, 8 and 9
Weather region 4	10, 13 and 15
		Weather region 5	City of land 11, city of land 12
Weather region 6	City of land 14, city of land 16

3. And (3) proportional coefficient calculation: an exponential smoothing method is used for predicting a proportionality coefficient C of loads of 6 meteorological areas at a time point t-1 of a day to be predicted to the load of the whole network_t. Due to the existence of the loss of the power grid, the sum of the proportionality coefficients of all meteorological areas at any time point is less than 1.

4. And (3) predicting the load of the whole network in each meteorological area: the market load prediction data of each place read by the step 1 areThe following 6 total network load prediction results were obtained:

5. optimal comprehensive model of single time point: and predicting the final total network load predicted value of the time point t being 1. Using the objective function solving method of the formulas (4) to (6) to obtain the optimal weight matrix W of the 6 total network load prediction results as follows: w ═ 0.1530.1540.2030.0670.1430.281.

Therefore, the final total network load predicted value of the time point when t is 1On the other hand, the actual load at the time point when t is 1 on day 18/5 is 15069.9MW, and the prediction accuracy reaches 99.17%.

6. And (4) load prediction of multiple points all day. And (5) repeating the steps 1 to 5 for the load prediction of other points of the predicted day with t from 2 to 96, so as to obtain the load predicted value sequence of the whole day.

Fig. 3 is 6 total network load prediction curves respectively predicted from 6 meteorological areas of 2016, 5, 18 and 4 days, and fig. 4 is a comparison graph of the final total day load prediction curve and the actual load curve solved by the optimal comprehensive model. By calculation, the daily accuracy of the method is 98.43%.

The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are given by way of illustration of the principles of the present invention, and that various changes and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (10)

1. The large power grid load prediction method based on the similarity of regional meteorological conditions is characterized by comprising the following steps of:

(1) acquiring load predicted values of M administrative areas reported by the load predicted value on a day to be predicted; acquiring historical data in a recent sample period, wherein load data in the historical data are actual loads of a whole network and M administrative regions;

(2) acquiring at least 2 pieces of historical meteorological condition data in M administrative areas in a sample period, and dividing the M administrative areas into q meteorological areas with larger meteorological condition difference through the historical meteorological condition data;

(3) dynamically predicting the proportionality coefficients of q meteorological areas at the same time point t of the day to be predicted by an exponential smoothing method to obtain a proportionality coefficient matrix C of the q meteorological areas at the time point t_t；

(4) Respectively predicting the load of the whole network system at the time point t by using the selected q regions to obtain q different prediction results;

(5) establishing an optimal comprehensive model of the t time point for q different prediction results, calculating the optimal weight of each selected meteorological area, and obtaining the final prediction result L of the load of the whole network system of the t time point_all,t：

$L_{all,t}=\underset{k=1}{\overset{q}{\&Sigma;}}{w}_t^k{L}_{all,t}^k,k=1,2,...,q$

Wherein,for the optimal weight of the load of the whole network system predicted by the meteorological area k at the time point t,the load of the whole network system at the t time point predicted by the kth meteorological area;

(6) for T time points of the whole day of the day to be predicted, respectivelyEstablishing an optimal comprehensive model to obtain an all-day load prediction sequence (L)_all,1,L_all,2,…,L_all,T) Predicting a sequence (L) with said all-day load_all,1,L_all,2,…,L_all,T) And (4) predicting the result of the load of the whole network.

2. The large power grid load prediction method based on regional meteorological condition similarity according to claim 1, wherein the large power grid load prediction method comprises the following steps: the at least 2 pieces of historical meteorological condition data are 2 or any combination of more than 2 of the highest temperature, the lowest temperature, the average temperature, the human body comfort level, the weather type, the wind speed, the temperature and humidity index and the cold index of the administrative region in one sample time period.

3. The large power grid load prediction method based on regional meteorological condition similarity according to claim 1, wherein the large power grid load prediction method comprises the following steps: the at least 2 historical meteorological condition data are the highest temperature, the lowest temperature and the average temperature of the administrative regions in a sample period, and a meteorological data matrix X is formed according to the historical meteorological condition data of the M administrative regions:

X＝(X₁,X₂,…X_i,…,X_M)

$X_i={\left(W_{i,n}^{\max }{W}_{i,n}^{\min }{W}_{i,n}^{ave}\right)}^T$

wherein,

X_ia set of n days of meteorological condition data for the ith administrative area within a sample period,

the highest temperature of the ith administrative region within a sample period of n days;

the minimum temperature of the ith administrative area for n days within a sample period;

average temperature for the ith administrative area over a sample period of n days.

4. The large power grid load prediction method based on regional meteorological condition similarity according to claim 3, wherein the large power grid load prediction method comprises the following steps: and aggregating the matrix X by adopting a synchronous back substitution elimination technology based on probability distance, and dividing M administrative areas into q meteorological areas with larger meteorological condition difference.

5. The large power grid load prediction method based on regional meteorological condition similarity according to claim 4, wherein the large power grid load prediction method comprises the following steps:

first, M column vectors in a matrix X are assigned a probability p of occurrence_sSet S represents the first administrative region, let ξ be 1/M (S is 1,2, …, M)_sRepresenting the s-th administrative region, DT, in the matrix X_s,s'Represents the distance between the administrative region s and the administrative region s ', and has a value of vector 2 norm between the administrative region s and the administrative region s', and the vector DA is set to [1,2, …, M ═ 1]^TAnd DB is a square matrix of M × M with all elements being 0.

The basic steps of synchronous back-substitution elimination are as follows:

1) calculating the distance DT between each pair of administrative regions_s,s'，DT_s,s'＝||ξ_s,ξ_s'||。

2) For each administrative region k, finding out the administrative region r with the shortest distance from the administrative region k, namely DT_k(r)＝minDT_k,s'。

3) Calculating PD_k(r)＝p_k*DT_k(r), k ∈ S, finding the administrative region index h in k, so that PD is_h＝minPD_k,k∈S。

4) Let S equal S- ξ_DA(h)And p is_r＝p_r+p_h. DB (h, h) ═ da (r), and da (h) is empty.

5) Repeat 2) -4) until the number of remaining administration areas is q.

6. The large power grid load prediction method based on regional meteorological condition similarity according to any one of claims 1 to 5, wherein: in the step (1), the historical data is preprocessed according to the following method:

order: l (d, t) is the load value at the time t on day d, L (d, t)₁) And L (d, t)₂) Two times t adjacent to the time t on day d₁、t₂Load value of L (d)₁T) and L (d)₂T) is the load value at the time point t on two days adjacent to d;

a) handling of missing data

If the load value L (d, t) at the time t on day d is missing, L (d, t) is obtained by using the formula (1):

L(d,t)＝αL(d,t₁)+αL(d,t₂)+βL(d₁,t)+βL(d₂,t) (1)

in formula (1), α and β are coefficients, α > β, α + β ═ 1;

b) processing of dead-end data

Defining as the allowable deviation rate of load, rho (d, t) as the actual deviation rate at the time point t on day d, when rho (d, t) is greater than or equal to, judging L (d, t) as bad data, using for bad dataThe substitution is carried out as follows:

$\stackrel{\&OverBar;}{L}(d,t)=\frac{L(d_1,t)+L(d_2,t)}{2}---\left(2\right).$

7. the large power grid load prediction method based on regional meteorological condition similarity according to any one of claims 1 to 5, wherein: the step (3) is expressed as:

C_t＝(C_1,t,C_2,t,…,C_q,t)

wherein,

C_1,tthe scale factor of the 1 st meteorological area at the time point t of the day to be predicted;

C_2,tthe scale coefficient of the 2 nd meteorological area at the time point t of the day to be predicted;

C_q,tand the proportionality coefficient of the qth meteorological area at the time point t of the day to be predicted.

8. The large power grid load prediction method based on regional meteorological condition similarity according to claim 7, wherein the large power grid load prediction method comprises the following steps: the exponential smoothing method in the step (3) is that the establishment of an exponential smoothing model is as follows:

$C_t^i=\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&lambda;}}_j{C}_{t,j}^i,i=1,2,...,q$

$\left\{\begin{array}{c}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&lambda;}}_j=1\\ 0\&le;{\mathrm{\&lambda;}}_j\&le;1\end{array}\right.,j=1,2,...,n---\left(3\right)$

in formula (3):the predicted value of the meteorological area i occupying the load proportion of the whole network system at the time t is shown,representing the actual value of the load proportion of the meteorological area i in the previous j days in the whole network at the time t; n is the number of days in a sample period; lambda [ alpha ]_jDenotes a weight coefficient, λ_j＝λ(1-λ)^j-1λ is a constant, and 0 < λ < 1.

9. The large power grid load prediction method based on regional meteorological condition similarity according to any one of claims 1 to 5, wherein: in step (4), q different prediction results are obtained, which are expressed as:

$(L_{all,t}^1,L_{all,t}^2,...,L_{all,t}^q)$

wherein,

to utilize the 1 st gasThe load of the whole network system at the t time point predicted by the image area;

the load of the whole network system at the t time point predicted by utilizing the 2 nd meteorological area;

is the total network system load of the t time point predicted by utilizing the qth meteorological area, and

L_q,tload prediction value of the qth meteorological area at time point t, C_q,tAnd predicting the proportionality coefficient of the qth meteorological area at the time point t.

10. The large power grid load prediction method based on regional meteorological condition similarity according to any one of claims 1 to 5, wherein: the optimal comprehensive model in the step (5) is established according to the following method:

an objective function of the load predicted value of the whole network system at the time point t is represented by an equation (4), and equations (5) to (6) are constraint conditions of the objective function:

$\begin{array}{cc}\underset{{\mathrm{\&omega;}}_k}{min}& z^{\left(t\right)}=\underset{j=1}{\overset{n}{\&Sigma;}}\underset{t=1}{\overset{T}{\&Sigma;}}{(\underset{k=1}{\overset{q}{\&Sigma;}}{w}_t^k{L}_{all,t,j}^k-L_{all,t,j})}^2\end{array}---\left(4\right)$

$\begin{array}{cc}S.t.& \underset{k=1}{\overset{q}{\&Sigma;}}{w}_t^k=1\end{array}---\left(5\right)$

$w_t^k\&GreaterEqual;0,k=1,2,...,q---\left(6\right)$

in formula (4):showing the predicted value L of the load of the whole network system predicted by the area k at the time point t of the j day_all,t,jRepresenting the actual load value of the whole network system at the time point t on the jth day;

calculating the optimal weight of the selected q meteorological areas at the time point t by the formula (4), the formula (5) and the formula (6)Then, again according toAnd weighting to obtain the total network system load predicted value of the t time point.