A distributed photovoltaic power short-term prediction method based on an improved similar time method

Abstract

The application relates to a distributed photovoltaic power short-term prediction method based on an improved similar time method. By constructing an improved similar time prediction model, the correlation degree parameters and simple correlation coefficient linear weights of various influence factors in the three-dimensional factors of meteorology, load and time are calculated to obtain the comprehensive influence correlation of the historical effective output time, the weights of the influence factors are determined, the screening indexes of the similar time are condensed, and a new idea is provided for the photovoltaic power short-term prediction technology of the distributed photovoltaic power station.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of distributed photovoltaic power short-term prediction method based on improved similar time method. BACKGROUND

[0002] The increasingly tense situation of world energy and environment brings the problem such as the contradiction between power supply and demand, so the development and utilization of traditional energy are more limited. In recent years, with the large-scale construction of new energy stations in China, the scale of distributed power grid connection is getting larger and larger, in order to realize the accurate scheduling of power grid, the power department needs to master the output data of new energy station in real time on the one hand, and needs to predict the power generation on the other hand.

[0003] Distributed photovoltaic power station usually refers to the use of dispersed resources, small-scale installation, arranged in user area near power generation system. Photovoltaic output is closely related to irradiance, and is also affected by short-term weather changes such as cloud cover and precipitation, which has strong intermittency and randomness, which has a great influence on power system planning and power supply quality, and this influence will become more and more prominent with the increase of photovoltaic grid connection capacity. Therefore, in the face of high proportion of distributed photovoltaic access to power grid, accurate prediction of short-term power generation of distributed photovoltaic has great significance for power grid operation and dispatching.

[0004] For the prediction of photovoltaic power output, the current prediction types can be divided into two categories: direct prediction and indirect prediction. The former estimates temperature, radiation and other environmental factors first, and then uses them as input of the model and estimates photovoltaic power output by combining certain algorithm. The latter directly uses historical photovoltaic output data as input of the prediction model, which has the advantage of relatively low cost, without the need to install a large number of equipment for environmental monitoring in advance, but its disadvantage is that it needs a large amount of historical data to support. Therefore, the similar day method is currently used for data preprocessing in the field of short-term prediction, which is a relatively simple and feasible method. This method uses meteorological characteristics to select similar days as model input for prediction, which greatly reduces the amount of historical data required. The existing similar day method mainly studies the overall similarity between historical sample days and the day to be measured, that is, the maximum, mean and variance of the influencing factors are calculated by selecting part of the data of the day, and the parameters obtained by weighting are used as characteristic values representing the overall situation of the historical day. It does not analyze the influence degree difference of different types of factors on the output situation at different times of the historical day, which is difficult to improve the pertinence and accuracy of the prediction.

[0005] The similar time prediction method is mainly used in short-term prediction field by analyzing the meteorological factors and photovoltaic power output value of the effective output time of historical days and studying the output difference of different time periods. The current similar time method mainly establishes a prediction model for two types of distributed photovoltaic power stations: ① The newly-built power station with less historical data often uses multiple methods for comprehensive analysis, that is, the historical data is directly used as the prediction value according to the fluctuation degree of photovoltaic output or the prediction is carried out by using artificial intelligence algorithm ② The conventional power station with more historical data selects a large number of influence factors and output data and then inputs the model for prediction. The former has the problems of long model training time, poor accuracy and adaptability and the like due to the elimination of heterogeneous data and the input of a large amount of redundant information into the model. The latter improves the prediction effect by reducing the dimension through correlation analysis of the data, but the weight of each meteorological factor is not determined in the similarity analysis. In the two cases, each type of meteorological factor and photovoltaic output value is input into the model as the main influence factor, the screening is not strong due to the lack of concise screening index, the number of similar time is insufficient or the correlation is weak, thereby affecting the prediction accuracy. SUMMARY

[0006] The purpose of the present application is to provide a distributed photovoltaic power short-term prediction method based on improved similar time method, (1) The correlation degree parameter and simple correlation coefficient of each type of influence factor in the three-dimensional factors of meteorology, load and time are linearly weighted to obtain the comprehensive influence correlation degree of the effective output time of the historical day, which provides a new idea for the weight determination of the influence factor and provides a concise screening index for the selection of similar time.

[0007] To achieve the above purpose, the technical scheme of the present application is as follows: a distributed photovoltaic power short-term prediction method based on improved similar time method, comprising the following steps:

[0008] (1) The comprehensive influence correlation degree limit value G of the initialization setting M and the minimum number N of similar time M ;

[0009] (2) Calculate the comprehensive influence correlation degree of each time in the to-be-predicted sample and each time in the similar day sample;

[0010] (3) Determine the number of similar times according to the size relationship satisfied by the calculated comprehensive influence correlation degree;

[0011] (4) The weighted method and the extrapolation method are used respectively to predict the number of similar times, and then the prediction result is obtained by taking the average of the two;

[0012] (5) Compare the actual value with the prediction value of each effective output time of the historical day sample and calculate the relative error sum of squares of the two;

[0013] (6) judging whether the relative error square sum reaches the minimum value: if yes, obtaining the photovoltaic power prediction value of the time instant j of the day to be measured; if no, changing the initial value and re-computing.

[0014] In an embodiment of the present application, step (2) is specifically implemented as follows:

[0015] (2.1) data acquisition and preprocessing

[0016] 1) By means of fuzzy ranking method, the photovoltaic output condition is converted by using the proportion of the average value of photovoltaic output at the time instant in the whole day effective output as the weight value, which is directly taken as the quantization data of the time factor; the proportion of the time instant j in the whole day 11 time instants is measured by using the average value of photovoltaic power at the time instant j in m historical day samples, and the proportion Z is calculated according to formula (1) j :

[0017]

[0018] In the formula: m ij represents the photovoltaic power at the time instant j in the i-th historical day sample;

[0019] 2) The m historical samples at the time instant j are sorted according to the power size, and when the photovoltaic power takes the maximum value, Z j is given, the weight value is m, when the second maximum value is taken, the weight value is m-1, and so on, when the minimum value is taken at a certain time instant, Z j is given, the weight value is 1; and the quantization data N ij of the time factor is obtained:

[0020] N ij = h ij Z j (2)

[0021] In the formula: the intermediate variable h ij represents the weight value given to each time instant to be measured;

[0022] 3) Characteristic symbol definition

[0023] Define S ij ,T ij ,D ij ,H ij ,W ij ,G ij ,P ij ,N ij as the environmental air pressure, environmental air temperature, wind grade, total irradiance, scattered irradiance, sensor working temperature, average current, time factor of the time instant j of the i-th day (i=0 represents the day to be measured) before the day to be measured, and m is the number of historical days;

[0024] 4) Data normalization

[0025] The measurement scale of different dimensional data is limited within a predetermined range, and 8 characteristic matrices f(u ij ) are calculated according to formula (3):

[0026]

[0027] In the formula, u=S, T, D, H, W, G, P, N; C is a constant; u ij is the type of influencing factor;

[0028] (2.2) Screening of influencing factors

[0029] The daily average values of environmental air pressure, environmental air temperature, wind grade, total irradiance, scattered irradiance, sensor operating temperature, average current, and time factor of 8 power influencing factors in m historical samples are calculated as independent variables, and the daily average value of photovoltaic output power in the corresponding historical sample day is calculated as the dependent variable;

[0030] The simple correlation coefficients of each influencing factor are calculated according to the path analysis method; the absolute value 0.2 of the simple correlation coefficient is taken as the threshold value, and the data greater than the threshold value is selected as the main influencing factor; the remaining 7 influencing factors after screening are: environmental air pressure, environmental air temperature, total irradiance, scattered irradiance, sensor operating temperature, average current, and time factor.

[0031] (2.3) Calculation of correlation degree

[0032] 1) According to the principle of "near large and far small" of time, the correlation degree is measured, that is, the farther the distance from the predicted day, the smaller the correlation, and vice versa; according to formula (4), the time correlation degree b i is linearly depicted:

[0033]

[0034] In the formula, i represents the i-th day before the predicted day, and b i represents the time correlation degree of the i-th day;

[0035] 2) The difference between the average currents at any two times is used to measure the correlation degree of the output at the two times, and the larger the difference, the smaller the correlation, and vice versa; according to formula (5), the average current correlation degree d ij between the i-th day before the predicted day and the same time j of the predicted day is calculated:

[0036]

[0037] 3) The TOPSIS method is used to define the optimal meteorological factor T Y and the worst meteorological factor T C, two types of reference meteorological factors are calculated according to formula (6) and (7):

[0038]

[0039]

[0040] In the formula: T Y1 —T Y5 and T C1 —T C5 are the optimal distance and the worst distance of five influencing factors of photovoltaic power and meteorological factors, i.e. ambient pressure, ambient temperature, total irradiance, scattered irradiance and sensor operating temperature, respectively.

[0041] 4) The TOPSIS method is optimized by using the simple correlation coefficient as the weight, and the positive distance M Zij and the negative distance M Fij are calculated according to formula (8) and (9):

[0042]

[0043] In the formula: r1-r5 are the simple correlation coefficients of ambient pressure, ambient temperature, total irradiance, scattered irradiance and sensor operating temperature, respectively.

[0044]

[0045] The closeness M cij of the meteorological characteristics at each time to the optimal meteorological characteristics is calculated according to formula (10):

[0046]

[0047] In the formula: 0≤M cij ≤1, the smaller the positive distance M Zij , the larger M cij , and the closer to the optimal meteorological factor;

[0048] 5) The comprehensive meteorological correlation g ij of the similar day sample and the comprehensive meteorological correlation of each time of the day to be predicted is calculated according to formula (11):

[0049]

[0050] (2.4) Calculation of comprehensive influence correlation

[0051] The simple correlation coefficient is weighted, and the correlation of various influencing factors in the three-dimensional factors of meteorology, load and time is combined, and the comprehensive influence correlation G ij is calculated according to formula (12):

[0052]

[0053] wherein r b , r d represent simple correlation coefficients of time factor and load factor respectively.

[0054] In an embodiment of the present application, the specific implementation of calculating the simple correlation coefficient of each influencing factor in step (2.2) based on the path analysis method is as follows:

[0055] (2.2.1) Calculation method of direct path coefficient

[0056] wherein x a and y represent independent variable and dependent variable respectively, both of which contain m groups of data; the direct path relationship r a between x 1,a and y is calculated according to formula (13):

[0057]

[0058] wherein x a,t at is the tth sample value of the ath influencing factor; y t at is the tth sample value of y; b a is the partial regression coefficient; and m is the number of selected historical days.

[0059] (2.2.2) Calculation method of indirect path coefficient

[0060] First, the correlation coefficient r a between any two independent variables x a+1 and x 2a,a+1 is calculated according to formula (14):

[0061]

[0062] The indirect path coefficient r a of x a+1 on y is calculated according to formula (15): 2a,a+1

[0063] r 2a,a+1 = r a,a+1 r 1,a+1 (15)

[0064] (2.2.3) Calculation method of simple correlation coefficient

[0065] Combining the characteristics of the two types of path coefficients in (2.1) and (2.2), the influence of the influencing factor itself on the prediction result and the influence of the influencing factor on the prediction result through other types of factors are comprehensively measured, and the simple correlation coefficient r a of x a and y is calculated according to formula (16): ​

[0066] r a = r 1,a +∑ o≠a r 2a,o (16)

[0067] In the formula: r 2a,o is the indirect path coefficient of the a-th influence factor with respect to the o-th influence factor other than a.

[0068] Compared with the prior art, the present application has the following beneficial effects:

[0069] (1) The present application proposes a distributed photovoltaic power short-term prediction method based on improved similar time method, which uses the correlation degree parameters and simple correlation coefficients of various influence factors in the three-dimensional factors of meteorology, load and time to obtain the comprehensive influence correlation of the historical effective output time, provides a new idea for the weight determination of influence factors, and provides a concise screening index for the selection of similar time.

[0070] (2) The present application constructs a prediction model based on improved similar time method, which provides theoretical support for the research on photovoltaic power short-term prediction of distributed photovoltaic power station. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 The present application is a flowchart of a distributed photovoltaic power short-term prediction method based on improved similar time method. DETAILED DESCRIPTION

[0072] The technical solutions of the present application will be specifically described below in combination with the drawings.

[0073] As shown in the drawings, the present application is a distributed photovoltaic power short-term prediction method based on improved similar time method, and the specific steps are as follows: Figure 1

[0074] (1) The limit value GM of the comprehensive influence correlation and the minimum number NM of similar time are initialized and set;

[0075] (2) The comprehensive influence correlation of each time in the predicted sample and each time in the similar day sample is calculated;

[0076] (3) The number of similar times is determined according to the size relationship of the calculated comprehensive influence correlation;

[0077] (4) The prediction is carried out by using the weighting method and the extrapolation method respectively according to the obtained number of similar times, and then the prediction result is obtained by taking the average of the two;

[0078] (5) The actual value is compared with the predicted value of each effective output time of the historical day sample, and the relative error sum of squares of the two is calculated; ​

[0079] (6)Judge whether the relative error square sum reaches the minimum value: if yes, the photovoltaic power prediction value of the time j is obtained; if not, change the initial value and recalculate.

[0080] 1Simple correlation coefficient calculation based on path analysis method

[0081] There are interactions among various factors affecting distributed photovoltaic power. This interaction not only shows the influence of the influencing factors on the prediction results, but also reflects the influence of the factors on the prediction results through other types of factors. Based on the path analysis method, the simple correlation coefficient is defined to comprehensively measure the direct and indirect influence degree in the interaction.

[0082] 1.1 Calculation method of direct path coefficient

[0083] x a and y represent the independent variable and the dependent variable, respectively, and both contain m groups of data. According to formula (1), the direct path relationship r a between x 1,a and y is calculated.

[0084]

[0085] In the formula: xa ,t is the tth sample value of the ath influencing factor; y t is the tth sample value of y; b a is the partial regression coefficient; and m is the number of selected historical days.

[0086] 1.2 Calculation method of indirect path coefficient

[0087] First, according to formula (2), the correlation coefficient r a between any two independent variables x a+1 and x 2a,a+1 is calculated.

[0088]

[0089] According to formula (3), the indirect path coefficient r a of x a+1 on y through x 2a,a+1 is calculated.

[0090] r 2a,a+1 = r a,a+1 r 1,a+1 (3)

[0091] 1.3 Calculation method of simple correlation coefficient

[0092] Combining the characteristics of the two types of path coefficients, we comprehensively measure the influence of the influencing factor itself on the prediction result and the influence of the influencing factor on the prediction result through other types of factors, and calculate x according to equation (4). a The simple correlation coefficient r with y a .

[0093] r a =r 1,a +∑ o≠a r 2a,o (4)

[0094] In the formula: r 2a,o Let be the indirect path coefficient of factor a with respect to factor o (excluding factor a).

[0095] 2. Determination of similarity time screening indicators

[0096] 2.1 Data Acquisition and Preprocessing

[0097] 1) Using the fuzzy sorting method, the photovoltaic output is calculated by using the proportion of the average photovoltaic output at historical moments to the total effective output for the whole day as the weight, and this is directly used as the quantitative data of the time factor. The average photovoltaic power at time j in m historical day samples is used to measure the proportion of that time among the 11 times of the whole day, and the proportion Z is calculated according to formula (5). j .

[0098]

[0099] Where: m ij Let represent the photovoltaic power at time j in the i-th historical day sample.

[0100] 2) Sort the m historical samples at time j by power magnitude, and assign Z when the photovoltaic power reaches its maximum value. j When the weight is m, the weight is changed to m-1 when the second largest value is obtained, and so on. When the photovoltaic power reaches its minimum value at a certain moment, Z is assigned a weight. j The weight is 1. This is used to obtain the quantified data N of the time factor. ij .

[0101] N ij =h ij Z j (6)

[0102] In the formula: intermediate variable h ij This indicates that weights are assigned to each time point to be measured.

[0103] 3) Definition of feature symbols

[0104] Define S ij ,T ij D ijH ij ,W ij ,G ij ,P ij ,N ij is the environmental pressure, environmental temperature, wind level, total irradiance, scattering irradiance, sensor working temperature, average current, time factor of the i-th (i = 0 represents the day to be tested) day j of the day before the day to be tested, and m is the number of historical days.

[0105] 4) Data normalization

[0106] The measurement scale of different dimensional data is limited within a certain range, and eight characteristic matrices f(u ij ) are calculated according to formula (7).

[0107]

[0108] In the formula: u = S, T, D, H, W, G, P, N; C is a constant; u ij is the type of influencing factor.

[0109] 2.2 Selection of influencing factors

[0110] The daily average values of the environmental pressure, environmental temperature, wind level, total irradiance, scattering irradiance, sensor working temperature, average current, and time factor of the m historical samples are calculated as independent variables, and the daily average value of the photovoltaic output power in the corresponding historical sample day is taken as the dependent variable.

[0111] The simple correlation coefficients of the above-mentioned influencing factors are calculated according to the path analysis. The absolute value 0.2 of the simple correlation coefficient is taken as the threshold value, and the data greater than the threshold value is selected as the main influencing factor. The remaining seven influencing factors after screening are: environmental pressure, environmental temperature, total irradiance, scattering irradiance, sensor working temperature, average current, and time factor.

[0112] 2.3 Calculation of correlation degree

[0113] 1) According to the principle of "near large and far small" of time, the correlation degree is measured, that is: the farther from the day to be predicted, the smaller the correlation, and vice versa.

[0114] According to formula (8), the time correlation degree b i is linearly depicted by the time distance.

[0115]

[0116] In the formula: i represents the i-th day before the day to be predicted, and b i represents the time correlation degree of the i-th day.

[0117] 2) The difference between the average current of any two time points is used to measure the correlation between the output of the two time points. The greater the difference, the smaller the correlation, and vice versa. The correlation d between the i-th day of the day to be predicted and the j-th time point of the day to be predicted is calculated according to formula (9) ij .

[0118]

[0119] 3) The TOPSIS method is used to define the optimal weather factor T Y and the worst weather factor T C , and the two types of reference weather factors are calculated according to formulas (10) and (11).

[0120]

[0121]

[0122] In the formula, T Y1 -T Y5 and T C1 -T C5 are the optimal distance and the worst distance of the five influencing factors (ambient pressure, ambient temperature, total irradiance, scattered irradiance, and sensor operating temperature) of photovoltaic power and weather factors.

[0123] 4) The simple correlation coefficient is used as the weight to optimize the TOPSIS method, and the positive distance M Zij and the negative distance M Fij are calculated according to formulas (12) and (13).

[0124]

[0125] In the formula, r1-r5 are the simple correlation coefficients of ambient pressure, ambient temperature, total irradiance, scattered irradiance, and sensor operating temperature, respectively.

[0126]

[0127] The closeness M cij of the weather characteristics at each time point to the optimal weather characteristics is calculated according to formula (14).

[0128]

[0129] In the formula, 0≤M cij ≤1, the smaller the positive distance M Zij , the larger M cij , and the closer to the optimal weather factor.

[0130] 5) The comprehensive weather correlation g ij of the similar day sample and each time point of the day to be predicted is calculated according to formula (15).

[0131]

[0132] 2.4 Calculation of comprehensive influence correlation

[0133] The comprehensive influence correlation G is calculated according to formula (16) by using the simple correlation coefficient for weighting, combining the correlation of various influence factors of the three-dimensional factors of meteorology, load and time. ij .

[0134]

[0135] In the formula, r b , r d respectively represent the simple correlation coefficients of the time factor and the load factor.

[0136] The above is the preferred embodiment of the present application. Any change made according to the technical solution of the present application, as long as the function does not exceed the scope of the technical solution of the present application, belongs to the protection scope of the present application.

Claims

1. A short-term forecasting method for distributed photovoltaic power based on an improved similarity-time method, characterized in that, Includes the following steps: (1) Initial setting of the comprehensive influence correlation limit value G M The minimum number N of similar times M ; (2) Calculate the correlation between the combined influence of each time point in the sample to be predicted and each time point in the sample of similar days; the specific implementation is as follows: (2.1) Data Acquisition and Preprocessing 1) Using the fuzzy sorting method, the photovoltaic output is calculated by using the proportion of the average photovoltaic output at historical moments to the total effective output for the whole day as the weight, and this is directly used as the quantitative data of the time factor; the average photovoltaic power at time j in m historical day samples is used to measure the proportion of time j in the 11 moments of the whole day, and the proportion Z is calculated according to formula (1). j : Where: m ij This represents the photovoltaic power at time j in the i-th historical day sample; 2) Sort the m historical samples at time j by power magnitude, and assign Z when the photovoltaic power reaches its maximum value. j When the weight is m, the weight is changed to m-1 when the second largest value is obtained, and so on. When the photovoltaic power reaches its minimum value at a certain moment, Z is assigned a weight. j The weight is set to 1; and the quantified data N of the time factor is obtained accordingly. ij : N ij =h ij Z j (2) In the formula: intermediate variable h ij This represents the weight assigned to each time point being measured; 3) Definition of feature symbols Define S ij ,T ij D ij H ij W ij G ij ,P ij N ij The eight influencing factors are ambient air pressure, ambient temperature, wind force level, total irradiance, diffuse irradiance, sensor operating temperature, average current, and time factor at time j on the i-th day before the measurement date. i = 0 represents the day before the measurement date, and m is the number of historical days. 4) Data normalization By limiting the measurement scale of data with different dimensions to a predetermined range, eight feature matrices f(u) are calculated according to equation (3). ij ): In the formula: u = S, T, D, H, W, G, P, N; C is a constant; u ij Types of influencing factors; (2.2) Screening of influencing factors The daily average values ​​of eight power-influencing factors—ambient air pressure, ambient temperature, wind force level, total irradiance, diffuse irradiance, sensor operating temperature, average current, and time factor—are calculated for each of the m historical samples and used as independent variables. The daily average value of photovoltaic output power in the corresponding historical sample days is used as the dependent variable. The simple correlation coefficients of each influencing factor were calculated using the path analysis method. The absolute value of the simple correlation coefficient was 0.2, which was used as a threshold. Data with values ​​greater than this threshold were selected as the main influencing factors. The seven influencing factors that remained after screening were: ambient air pressure, ambient temperature, total irradiance, diffuse irradiance, sensor operating temperature, average current, and time factor. (2.3) Calculation of relevance 1) The correlation is measured according to the principle of "nearer is larger, farther is smaller", that is: the farther away from the day to be predicted, the smaller the correlation, and vice versa; according to formula (4), the time correlation is linearly characterized by time distance b. i : In the formula: i represents the i-th day before the date to be predicted, b i Indicates the time relevance on day i; 2) The difference between the average current at any two times is used to measure the correlation between the output at the two times. The larger the difference, the smaller the correlation, and vice versa. The correlation d between the average current at the same time j on the i-th day before the forecast date and the same time j on the forecast date is calculated according to formula (5). ij : 3) Define the optimal meteorological factor T using the TOPSIS method. Y And worst weather factor T C The two types of baseline meteorological factors are calculated according to equations (6) and (7): In the formula: T Y1 —T Y5 and T C1 —T C5 The optimal and worst distances for photovoltaic power and five meteorological factors are: ambient air pressure, ambient temperature, total irradiance, diffuse irradiance, and sensor operating temperature. 4) Optimize the TOPSIS method using simple correlation coefficients as weights, and calculate the positive distance M according to equations (8) and (9). Zij and the reverse distance M Fij : In the formula: r1-r5 are simple correlation coefficients of ambient air pressure, ambient air temperature, total irradiance, diffuse irradiance, and sensor operating temperature, respectively; The degree of similarity between the meteorological characteristics at each time point and the optimal meteorological characteristics is calculated according to equation (10). cij ; Where: 0≤M cij ≤1, positive distance M Zij The smaller M is cij The larger the value, the closer it is to the optimal meteorological factors; 5) The comprehensive meteorological correlation g between similar daily samples and the time of the day to be predicted is calculated according to formula (11). ij : (2.4) Calculation of the overall influence correlation Using simple correlation coefficients for weighting, and combining the correlations of various influencing factors in the three dimensions of meteorology, load, and time, the comprehensive influence correlation G is calculated according to equation (12). ij : In the formula: r b r d These represent the simple correlation coefficients of the time factor and the load factor, respectively. (3) Determine the number of similar moments based on the magnitude of the calculated correlation between the comprehensive impacts; (4) Based on the number of similar times obtained, predictions are made using the weighted method and the extrapolation method respectively, and the prediction results are obtained by taking the average of the two methods. (5) Compare the actual values ​​with the predicted values ​​at each effective output time of the historical daily sample and calculate the sum of squares of the relative errors between the two. (6) Determine whether the sum of squares of relative error has reached the minimum value: if so, obtain the predicted value of photovoltaic power at time j on the day to be measured; if not, change the initial value and recalculate.

2. The method for short-term prediction of distributed photovoltaic power based on the improved similarity time method according to claim 1, characterized in that, In step (2.2), the specific implementation method for calculating the simple correlation coefficients of each influencing factor based on the path analysis method is as follows: (2.2.1) Calculation method of direct path factor x a y and y represent the independent and dependent variables, respectively, each containing m sets of data; x is calculated according to equation (13). a The direct path relationship r with y 1,a : In the formula: x a,t Let y be the t-th sample value of the a-th influencing factor; t Let y be the t-th sample value; b a is the partial regression coefficient; m is the number of historical days selected; (2.2.2) Calculation method of indirect path factor First, calculate any two independent variables x according to equation (14). a With x a+1 The correlation coefficient r between them 2a,a+1 : Calculate x according to equation (15) a Through x a+1 Indirect path coefficient r for y 2a,a+1 : r 2a,a+1 ＝r a,a+1 r 1,a+1 (15) (2.2.3) Calculation method of simple correlation coefficient Combining the characteristics of the two types of path coefficients (2.1) and (2.2), and comprehensively weighing the influence of the influencing factor itself on the prediction result and the influence of the influencing factor on the prediction result through other types of factors, x is calculated according to equation (16). a The simple correlation coefficient r with y a : r a ＝r 1,a +∑ o≠a r 2a,o (16) In the formula: r 2a,o Let be the indirect path coefficient of factor a with respect to factor o (excluding factor a).

Images