A transformer substation power energy scheduling method and system

Abstract

The present application relates to a kind of power energy scheduling method and system of transformer area, including the following contents: the influencing factor of determining transformer area electric power load;Obtain the influencing factor data of prediction day, and historical sample with influencing factor data and historical load data, to form sample data;The historical load data of screening is decomposed;Based on the load component obtained by decomposition, respectively obtain the load component of prediction day;According to the load component of prediction day, transformer area electric power energy scheduling is carried out.The present application can provide reference for the dispatching of intelligent energy in transformer area.

Description

Technical Field

[0001] This invention relates to load forecasting technology within a transformer substation area, and more specifically to a method and system for dispatching power energy within a transformer substation area. Background Technology

[0002] Power system load forecasting is a crucial basis and key link in power dispatch automation. With the widespread integration of new energy devices, such as electric vehicles, energy dispatch within distribution areas is becoming increasingly important. Distribution area load forecasting is a vital reference for energy dispatching in smart energy service systems. However, in practice, the following shortcomings have been found in distribution area power load forecasting methods:

[0003] First, the sample selection is haphazard, lacking effective means to select similar days, and it does not consider the impact of real-world factors such as holidays, day type, meteorological factors, economic factors, social factors, and residential electricity consumption characteristics on sample selection and load forecasting, making it difficult to use methods such as cluster analysis to establish load forecasting models;

[0004] Secondly, current load forecasting is mostly focused on power load in large areas such as cities and provinces, with little research on load forecasting at the transformer substation level. In fact, load forecasting in small areas, represented by transformer substations, is characterized by strong randomness in load changes and many influencing factors. Current large-area load forecasting methods are not entirely applicable to transformer substation load forecasting, resulting in a lack of effective means to forecast and analyze transformer substation loads with the above characteristics.

[0005] Finally, existing machine learning models for power load forecasting lack a basis for selecting initial values ​​of kernel parameters and learning parameters, such as those represented by support vector machines, and fail to consider the statistical characteristics of power load. This results in a lack of specificity in the selection of initial values ​​of learning parameters, increases the number of iterations of the algorithm, and makes it easy to get trapped in local solutions, so that the optimal solution obtained may not be the global solution. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for power dispatching within a distribution transformer area. This invention can provide a reference for smart energy dispatching within a distribution transformer area.

[0007] The technical solution of the present invention is as follows.

[0008] A method for dispatching power resources in a distribution area includes the following steps:

[0009] Based on the pre-set influencing factors of the power load of the transformer area and the obtained influencing factor values ​​for the predicted day, historical load data for several similar days are selected from the historical data.

[0010] The selected historical load data is then decomposed;

[0011] Based on the load components obtained from the decomposition, the predicted daily load components are obtained respectively;

[0012] Power dispatching for the distribution area is performed based on the load components of the predicted day.

[0013] Preferably, the factors influencing the power load of the transformer area include meteorological factors and day type factors.

[0014] Preferably, the step of selecting historical load data for several similar days from historical data based on pre-set influencing factors of the power load of the transformer area and the obtained influencing factor values ​​for the predicted day includes:

[0015] Historical sample data are formed by obtaining corresponding historical data based on the pre-defined factors affecting the power load of the transformer area.

[0016] The sample data is normalized, and the normalized sample data is then clustered based on the influencing factor values ​​of the predicted date to obtain similar dates.

[0017] Obtain the historical load data corresponding to the similar days.

[0018] Preferably, the step of performing cluster analysis on the normalized sample data based on the influencing factor values ​​of the predicted date to obtain similar days includes:

[0019] Construct a normalized matrix based on the normalized sample data;

[0020] A similarity coefficient matrix is ​​established based on the normalized matrix data to reflect the correlation between the influencing factor values ​​of the predicted date and the influencing factor data of historical samples;

[0021] By setting a threshold to filter the similarity coefficient matrix, similar days can be obtained.

[0022] Preferably, the normalized matrix data is as follows:

[0023]

[0024] In the formula, X N×M For the historical sample normalization matrix, x ij Let M be the values ​​of the j influencing factors for the i-th sample, M be the number of influencing factors, and N be the number of historical daily samples selected.

[0025] Preferably, the decomposition of the selected historical load data includes:

[0026] The selected historical load data was decomposed in three levels to obtain three low-frequency components and three high-frequency components.

[0027] Select the lowest frequency component from the low-frequency components as the base charge;

[0028] Select the two lower-frequency components from the high-frequency components;

[0029] The remaining highest frequency component is treated as a noise component and smoothed to remove noise.

[0030] The lowest frequency component reflects the load variation trend over a day; the two selected high frequency components describe the load variation trend over a short time scale.

[0031] Preferably, the step of using the least squares support vector machine method to obtain the predicted daily load components from the decomposed load components includes:

[0032] The selected low-frequency and high-frequency components are respectively used to obtain the load components corresponding to the low-frequency and high-frequency components using the least squares support vector machine method.

[0033] Based on the remaining high-frequency components, the load components are reconstructed to obtain the predicted daily load components.

[0034] Preferably, the least squares support vector machine model is as follows:

[0035]

[0036] In the formula, y represents the predicted load at a certain point in a day, w is the weight vector, T is the transpose, and x includes two parts of data: a normalized vector of influencing factors for each time point on the predicted day, and x... i Let be the input vector at a certain moment on the i-th sample day. x i Nonlinear mapping to a high-dimensional space, where b is a bias constant and α i Let K(x, x) be the Lagrange multiplier, n be the number of similar day samples selected, and K(x, x) be the Lagrange multiplier. i ) is the kernel function;

[0037] in, The normalized historical load data for the day preceding the corresponding time of the i-th sample day is represented by one of the selected low-frequency components or two high-frequency components. Let M be the normalized vector of the M influencing factors at the corresponding time on day i; M is the number of influencing factors.

[0038] Kernel function K(x,x) i The formula for calculating ) is as follows:

[0039]

[0040] In the formula, p is the kernel function parameter.

[0041] Furthermore, the method for predicting the power load of the distribution area also includes evaluating the predicted load using relative error and absolute error.

[0042] The present invention also provides a power dispatching system for distribution areas, including a screening module, a decomposition module, a prediction module, and a dispatching module;

[0043] The filtering module filters out historical load data for several similar days from historical data based on pre-set influencing factors of the power load of the distribution area and the obtained influencing factor values ​​for the predicted day.

[0044] The decomposition module is used to decompose the selected historical load data;

[0045] The prediction module obtains the predicted daily load components based on the decomposed load components.

[0046] The scheduling module performs power energy scheduling for the distribution area based on the predicted load components.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] This invention employs a power dispatching method for distribution transformer areas, comprising: selecting historical load data for several similar days from historical data based on pre-set influencing factors of distribution transformer area power load and obtained influencing factor values ​​for the predicted day; decomposing the selected historical load data; obtaining the load components for the predicted day based on the decomposed load components; and performing power dispatching for distribution transformer areas according to the load components for the predicted day. This invention solves the technical difficulties faced by distribution transformer area power load forecasting technology, such as the large number of influencing factors, strong randomness of load changes, and large load fluctuations. It provides a reference for smart energy dispatching within the distribution transformer area and is a beneficial and effective exploration that can be widely applied in the field of load forecasting technology within the distribution transformer area. Attached Figure Description

[0049] Figure 1 This is a flowchart of the power load forecasting method for transformer substations according to the present invention;

[0050] Figure 2 This is a flowchart of one embodiment of the transformer area power load prediction method based on similar day wavelet support vector machine of the present invention;

[0051] Figure 3 This is a schematic diagram of one embodiment of the historical load data decomposition and reconstruction of the present invention;

[0052] Figure 4 This is a schematic diagram of the three-level decomposition of historical load data for seven sample days in a certain area based on the Db4 wavelet basis;

[0053] Figure 5 These are the prediction results for each load component using the LSSVM model;

[0054] Figure 6 It compares the actual load on the predicted day with the predicted load;

[0055] Figure 7 It is the relative error of load forecasting (RAR). Detailed Implementation

[0056] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0057] Example 1:

[0058] This invention provides a method for power dispatching in a distribution area, including as follows: Figure 1 The content shown is as follows:

[0059] First, based on the pre-set influencing factors of the power load of the transformer area and the obtained influencing factor values ​​for the predicted day, historical load data for several similar days are selected from the historical data;

[0060] Then, the selected historical load data is decomposed;

[0061] Secondly, the predicted daily load components are obtained based on the load components obtained from the decomposition.

[0062] Next, power dispatching for the distribution area is carried out based on the load components of the predicted day.

[0063] The specific process is as follows: Figure 2 As shown,

[0064] In practice, it has been found that the power load of the distribution area is affected by a variety of factors, such as meteorological factors and day type factors. Meteorological factors include temperature, wind force and weather type, while day type factors include normal days, holidays and week types.

[0065] The accuracy of power load forecasting in the distribution area is closely related to both the aforementioned influencing factors and the selection of similar days. In this embodiment of the invention, a similar day division model is established based on the influencing factors. Since different influencing factors have different physical dimensions, it is necessary to normalize the sample data, including historical load data and influence type data.

[0066] 1. Based on the pre-defined influencing factors of the power load in the transformer substation and the obtained influencing factor values ​​for the predicted day, select historical load data for several similar days from historical data, including:

[0067] The process involves acquiring influencing factor data and historical influencing factor data for the predicted date to form sample data; normalizing the sample data and performing cluster analysis on the normalized sample data to obtain similar dates; and acquiring the historical load data corresponding to the similar dates, which specifically includes the following:

[0068] S11: Determine the factors affecting the power load of the transformer area;

[0069] In this embodiment, the influencing factors for determining the power load of the distribution area include: daily hourly temperature, daily hourly wind force, weather type, and day type; wherein, the weather type includes sunny, cloudy, overcast, rain / snow, and the day type includes weekday, weekend (Saturday / Sunday), and public holiday;

[0070] S12: Obtain the influencing factor data for the forecast date, as well as historical samples with influencing factor data and historical load data, to form sample data;

[0071] In this embodiment, historical samples from the 30 days prior to the prediction date and historical samples from the same time each day (every 5-minute time unit) within this range for the past 5 years are initially selected. Each historical sample data includes the historical load of the transformer area every 5 minutes and has M (M is 4 in this embodiment) influencing factor data established in step S11. The predicted load every 5 minutes on the prediction date also has M (M is 4) influencing factor data. For N historical samples at a certain time on the prediction date, there is an N×M influencing factor input matrix:

[0072]

[0073] In the formula, X N×M Let x be the historical sample matrix. ij Let x represent the values ​​of the j-th influencing factor for the i-th sample, M be the number of influencing factors, and N be the number of historical days. i1 ,x i2 ,…x iM (M=4) represents the daily hourly temperature, daily hourly wind force, weather type (sunny, cloudy, overcast, rain / snow), and day type (weekday, weekend (Saturday / Sunday), holiday) of the i-th sample, respectively;

[0074] S13: Original data x of influencing factors in S12 ij Standardization and normalization processes were performed.

[0075] In this embodiment, the influencing factor data x ij Standardization and normalization processes were performed. The standardization and normalization formulas used for temperature and wind speed indicators are shown below. S represents the average values ​​of temperature and wind speed, respectively. j Let x″ be the standard deviation of each of the temperature and humidity indices.iij For the normalized data of temperature and wind influence factors, x′ ij x′ is the standardized data for temperature and wind speed. jmax Let x′ be the maximum value of each of the temperature and wind speed indicators after standardization among N samples. jmin The minimum values ​​after standardization of temperature and wind speed indicators:

[0076]

[0077]

[0078]

[0079]

[0080] The normalized sample data constitutes an N×M normalized matrix X″. N×M The normalized forms for weather type and day type are shown in Table 1 below:

[0081] Table 1

[0082]

[0083] S14: Perform cluster analysis on the normalized sample data to establish a similar day selection model;

[0084] In this embodiment, the normalized data matrix X″ in S13 is taken. N×M The matrix contains N samples, each with M features. For example, in S12, M is 4. For X″ ij Establish the similarity coefficient matrix R ij =r(ij) N×N Each similarity coefficient r ij Reflects the degree of correlation between sample i and sample j:

[0085]

[0086]

[0087]

[0088] x″ ik Let x″ be the normalized value of the k-th influencing factor for the i-th sample. jk Let s be the normalized value of the k-th influencing factor for the j-th sample. k The normalized sample variance of the Mth influencing factor. To determine the average value of the Mth influencing factor sample after normalization, a threshold α is set, and the similarity coefficient matrix R... ij =r(ij)N×N In this embodiment of the invention, a threshold of 0.8 is set for classification. If α is greater than 0.8, sample i is considered to be related to sample j as 1, otherwise it is 0. Furthermore, in this embodiment of the invention, similar days are screened and suitable samples are selected by the correlation between the influencing factors of the predicted day and the influencing factors of historical samples.

[0089] 2. For example Figure 2 As shown, the historical load data selected in step 1 is decomposed;

[0090] In this embodiment, S represents the historical load data of the selected similar days. After layer-by-layer decomposition, three low-frequency components A1, A2, and A3 and three high-frequency components D1, D2, and D3 are obtained sequentially from high to low. Among them, the lowest frequency component A3 in the low-frequency part is used as the low-frequency part of the actual load, reflecting the load change trend of a day as the base load part. The lower frequencies D3 and D2 in the high-frequency part are used as the high-frequency part of the actual load, describing the load change trend over a shorter time scale. The high-frequency component D1 has a strong randomness characteristic and is treated as clutter noise. In this embodiment, the Least Squares Support Vector Machine (LSSVM) method is used to obtain the corresponding components of the predicted daily load for A3, D3, and D2. The high-frequency component D1, as the clutter component of the predicted daily load, is processed by mean filtering.

[0091] 3. The predicted daily load components are obtained by using the least squares support vector machine method on the actual load components A3, D3, and D2 obtained from wavelet decomposition in step 2.

[0092] In this embodiment, the following method is used for each of A3, D3, and D2:

[0093] For sample (x) i ,y i The mathematical model used by the Least Squares Support Vector Machine (LSSVM) for (i = 1, 2, ..., n) is as follows:

[0094]

[0095] st0≤α i ≤C,i=1,2,…,n (10)

[0096]

[0097] In the formula, y is the predicted load at a certain point in a day, w is the weight vector, T is the transpose, and x is the normalized vector of each influencing factor at each time point on the predicted day. i Let be the input vector at a certain moment on the i-th sample day. x i Nonlinear mapping to a high-dimensional space, where b is a bias constant and α iK(x,x) is a Lagrange multiplier. i ) is the kernel function. A kernel function that satisfies the Mercer condition is further defined in this embodiment of the invention. Let be the input vector at a certain moment on the i-th sample day, consisting of two parts of data, where The normalized historical load data for the day preceding the i-th sample day is one of A3, D3, and D2, and the normalized vector of the four influencing factors for the i-th day corresponding to the time in step S12 of step 1. y i For the historical load corresponding to the i-th sample day, normalized historical load data x for the corresponding time of the day before the prediction date is input. 0 The input vector x = (x_t) is composed of the normalized vectors of the four influencing factors at the corresponding time on the prediction date. 0 ,x 1 ,…,x 4 Based on the decision function described above, the predicted load for each of the 288 time periods of the predicted day is obtained. The selected kernel function is the following radial basis function:

[0098]

[0099]

[0100] In this embodiment, the relative error (RAR) and absolute error (AE) are used as the evaluation indicators for the prediction error, as follows:

[0101]

[0102]

[0103] In the above formula, For a 5-minute forecast load, P t To correspond to the actual load over 5 minutes, the forecast period is 288 time units from 0 to 24 hours of the forecast day.

[0104] In this embodiment of the invention, the kernel parameters p and learning parameters C of the least squares support vector machine load prediction model in step 3 are determined based on the normal distribution characteristics of the power load and combined with the particle swarm optimization algorithm. The regression coefficients α and b are determined using the SMO algorithm that satisfies the support vector duality constraint. The specific SMO algorithm is as follows:

[0105] S31: Establish the optimization objective and constraints of the SMO algorithm. For the least squares support vector machine model in step 3, if the kernel function satisfies... Then solve the following dual problem:

[0106]

[0107]

[0108] y represents the predicted load at a certain point in a day, and α represents the load. i For a given x, a Lagrange multiplier is used. i y i Kernel function K(x) i ,x j There must exist an optimal solution. At the same time, it can be found that:

[0109]

[0110] S32: For the kernel parameter p in formula (12), given that the power load of the distribution area generally follows a normal distribution, the initial value can be determined by the sample standard deviation:

[0111]

[0112]

[0113] p=3σ (21)

[0114] This invention determines that the ratio of the kernel parameter p to the sample standard deviation б is 3, and the initial value of the learning parameter C is 10;

[0115] S33: The particle swarm optimization algorithm is used to optimize the kernel parameter p and the learning parameter C. The minimum relative error RAR and the minimum absolute error AE are used as the search direction of the particles, and the number of particles is selected as 20.

[0116] S34: Iteratively solve for the optimal values ​​of p and C, reconstruct the load prediction value for each component, and use a relative error of less than 7% as the termination condition for iterative solution; otherwise, continue to execute steps S31, S32, S33, and S34.

[0117] 4. Perform power energy dispatching for the distribution area based on the predicted daily load components.

[0118] This invention establishes a similar day selection model using cluster analysis, which fully considers real-world influencing factors such as weather and day type. By using a similarity coefficient matrix to select similar days, it can effectively select similar days and ensure the rationality and effectiveness of the selected samples.

[0119] This invention targets the load of a transformer substation and uses the Db4 wavelet basis to perform a three-level decomposition of the original load. By extracting high-frequency and low-frequency components, the characteristics of the power load in the transformer substation are effectively analyzed.

[0120] Example 2

[0121] To clearly illustrate the power load prediction method for transformer substations based on similar day wavelet support vector machines provided in this embodiment of the invention, the following detailed explanation is provided in conjunction with the power load of a certain transformer substation:

[0122] like Figures 3-4 As shown, the forecast date for a certain transformer area is December 13, 2018, and the target is to forecast the power load for a total of 288 time periods, with each period lasting 5 minutes.

[0123] The first step is to input sample data, normalize the sample data, establish a similar day segmentation model, and select typical days:

[0124] In this embodiment of the invention, a total of 288 sampling points are used for the daily load. Each 5-minute interval serves as a sample data unit for the power load of the distribution area. Taking the load information for the 100th time period of the predicted day as an example, sample information for the 100th time period of each day from November 13, 2018 to December 12, 2018 is collected. Sample information for the previous five years for the 100th time period of each day within this time range is also collected. Thus, for the predicted load of the 100th time period of the predicted day, a total of 180 historical sample information points are collected. Each historical sample information point covers the load, temperature, wind force, weather type, and day type for that time period. The sample information for the predicted day includes temperature, wind force, weather type, and day type. The weather type includes sunny, cloudy, overcast, and rain / snow, and the day type includes weekday, weekend (Saturday / Sunday), and public holiday.

[0125] The influencing factor data and load data were normalized to form a normalized influencing factor input matrix. The normalized data for daily type and weather conditions are shown in Table 2 below.

[0126] Table 2

[0127]

[0128] In this invention, the influencing factors for the 100th time period of the predicted date December 13, 2018 are: temperature 7℃, wind force level 1, weather type is cloudy, and day type is a normal workday (Thursday). The normalized input vector of the influencing factors is (0.81, 0.56, 0.3, 0.2).

[0129] Input matrix X″ of load influencing factors after normalization of the selected 180 samples 180×180 Establish the similarity coefficient matrix R ij =r(ij) 180×180 Each similarity coefficient r ij Reflects the degree of correlation between sample i and sample j:

[0130]

[0131]

[0132]

[0133] A threshold α is set to classify the similarity coefficient matrix. In this invention, the threshold is set to 0.8. If α is greater than 0.8, sample i is considered to be related to sample j as 1; otherwise, it is 0. Furthermore, this invention filters similar days by the degree of correlation between the influencing factors of the predicted day and the influencing factors of historical samples. Finally, seven similar days are selected as sample days, as shown in Table 3.

[0134] Table 3

[0135]

[0136] The second step involves performing a three-level decomposition on the historical load data of the seven similar days selected in the first step using the Db4 wavelet basis. S represents the original load data for the seven sample days. Through this layer-by-layer decomposition, three low-frequency components (A1, A2, A3) and three high-frequency components (D1, D2, D3) are obtained. A3, representing the low-frequency component of the actual load, reflects the load variation trend over a single day and serves as the base load. D2 and D3, representing the high-frequency component of the actual load, describe the load variation trend over a shorter time scale. The high-frequency component D1, exhibiting strong randomness, is treated as noise. This invention employs the Least Squares Support Vector Machine (LSSVM) method to obtain the corresponding components for the predicted daily load from A3, D3, and D2. The high-frequency component D1, as a component of the predicted daily load, undergoes mean filtering. Figure 2 The high-frequency components D1, D2, D3 and low-frequency component A3 are obtained from the decomposition of the original load S of similar days. The low-frequency component A3 shows the daily load variation trend and the peak and trough show periodic changes in the seven sample days. The high-frequency component D3 reflects the high-frequency details of the daily load variation well, and D2 reflects the high-frequency randomness of the load variation well. D3 and D2 can represent the random operation of high-power electrical equipment and energy equipment in the transformer area, while D1 shows the high-frequency noise characteristics of the original load.

[0137] The third step involves using the Least Squares Support Vector Machine (LSSVM) method to obtain the predicted daily load components from the actual load components A3, D3, and D2 obtained from the wavelet decomposition in the second step. In this example, the mathematical model used for the LSSVM of the selected samples is as follows:

[0138]

[0139] st0≤α i ≤C,i=1,2,…,7

[0140]

[0141] Furthermore, in this example, The input vector for the i-th sample day includes the historical load normalized data from the day before the i-th sample day. Normalized data of the four influencing factors on day i in step two y i For the historical load of the i-th sample day, normalized historical load data x from the day before the prediction date is input. 0 The vector x = (x_t) is formed by the normalized data of the four influencing factors on the prediction date. 0 ,x 1 ,…,x 4 Based on the decision function described above, the predicted load for the predicted date is obtained. The selected kernel function is the following radial basis function:

[0142]

[0143]

[0144] The relative error (RAR) and absolute error (AE) are used as evaluation metrics for the prediction results, as follows:

[0145]

[0146]

[0147] The fourth step, regarding the solution steps for the kernel parameters p and C of the least squares support vector machine model in the third step, is as follows:

[0148] S1: Establish the optimization objective and constraints of the SMO algorithm, and solve the following dual problem for the least squares support vector machine model in step 3:

[0149] F(α)=min

[0150] st

[0151] For a given x i y i Kernel function K(x) i ,x j There must exist an optimal solution. At the same time, it can be found that:

[0152]

[0153] S2: The initial value of the kernel parameter p can be determined using the sample standard deviation.

[0154]

[0155]

[0156] p = 3σ

[0157] This invention fully utilizes the normal distribution characteristics of power load. When the ratio of the kernel parameter p to the standard deviation б of the kernel function is 3, 90% of the load information can be retained, and the initial value of the learning parameter C is determined to be 10.

[0158] S3: The particle swarm optimization algorithm is used to optimize the kernel parameter p and the learning parameter C. The minimum relative error RAR and the minimum absolute error AE are used as the search direction for particles. The number of particles is selected as 20. Given that the particle swarm optimization algorithm is very mature, [following the previous steps]... Figure 7 The specific details will not be elaborated upon in this example.

[0159] S4: Iteratively solve for the optimal values ​​of p and C, reconstruct the load forecast value for each component, and use a relative error of less than 7% as the termination condition for iterative solution; otherwise, continue to execute steps S1, S2, S3, and S4.

[0160] In this example, the learning parameters p and C of the least squares support vector machine model for low-frequency load component A3, high-frequency load components D2, and D3 are shown in Table 4 below:

[0161] Table 4

[0162] Load components p C a3 0.21 1.12 d3 0.052 0.18 d2 0.021 0.072

[0163] In this embodiment, the predicted load and actual load for the 288 time periods of the transformer area on December 13 are as follows: Figure 5 As shown, the relative error RAR is as follows: Figure 6 As shown.

[0164] This invention employs a least-squares support vector machine-based power load prediction model, taking into account the normal distribution characteristics of power load and establishing an initial value for the kernel parameter p. This effectively addresses nonlinear problems, improves the pertinence of the particle swarm optimization algorithm in solving for kernel parameters, and significantly enhances the accuracy of transformer area load prediction.

[0165] Example 3:

[0166] This invention provides a power dispatching system for distribution areas based on similar day wavelet support vector machines, comprising the following devices and modules:

[0167] The system includes monitoring equipment for real-time power data testing, a database for storing historical power data, a filtering module, a decomposition module, a prediction module, and a scheduling module.

[0168] Among them, the detection equipment is used to acquire the influencing factor data of the predicted day, as well as historical samples with influencing factor data and historical load data, to form a database that stores historical power data and influencing factor data;

[0169] The filtering module determines the influencing factors of the power load of the transformer area based on the pre-set influencing factors and the obtained influencing factor values ​​for the predicted day. It then filters out historical load data for several similar days from historical data. The sample data is normalized, and cluster analysis is performed on the normalized sample data to establish a similar day division model.

[0170] The decomposition module performs a three-level decomposition of the selected historical load data using the Db4 wavelet basis.

[0171] The prediction module uses the least squares support vector machine method to obtain the predicted daily load components from the decomposed load components and reconstructs them to obtain the predicted load.

[0172] The dispatching module performs power energy dispatching for distribution areas based on predicted load trends.

[0173] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0174] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A method for dispatching power energy in a distribution area, characterized in that, include: Based on the pre-set influencing factors of the power load of the transformer area and the obtained influencing factor values ​​for the predicted day, historical load data for several similar days are selected from the historical data. The selected historical load data is then decomposed; Based on the load components obtained from the decomposition, the predicted daily load components are obtained respectively; Power dispatching for the distribution area is performed based on the load components of the predicted day. Based on the pre-set influencing factors of the power load in the distribution area and the obtained influencing factor values ​​for the predicted day, historical load data for several similar days are selected from historical data, including: Historical sample data are formed by obtaining corresponding historical data based on the pre-defined factors affecting the power load of the transformer area. The sample data is normalized, and the normalized sample data is then clustered based on the influencing factor values ​​of the predicted date to obtain similar dates. Obtain the historical load data corresponding to the similar days; The process of clustering the normalized sample data based on the influencing factor values ​​of the predicted date to obtain similar days includes: Construct a normalized matrix based on the normalized sample data; A similarity coefficient matrix is ​​established based on the normalized matrix data to reflect the correlation between the influencing factor values ​​of the predicted date and the influencing factor data of historical samples; Set a threshold to filter the similarity coefficient matrix and obtain similar days; The normalized matrix data is as follows: （1） In the formula, For the historical sample normalization matrix, For the first one sample j The values ​​of each influencing factor, M is the number of influencing factors, and N is the number of historical day samples selected; The process of decomposing the selected historical load data includes: The selected historical load data was decomposed in three levels to obtain three low-frequency components and three high-frequency components. Select the lowest frequency component from the low-frequency components as the base charge; Select the two lower-frequency components from the high-frequency components; The remaining highest frequency component is treated as a clutter component and smoothed for noise reduction. The lowest frequency component reflects the load variation trend over a day; the two selected high frequency components describe the load variation trend over a short time scale. The method of obtaining predicted daily load components based on the decomposed load components includes: The selected low-frequency and high-frequency components are respectively used to obtain the load components corresponding to the low-frequency and high-frequency components using the least squares support vector machine method. Based on the remaining high-frequency components, the load components are reconstructed to obtain the predicted daily load components. The least squares support vector machine model is as follows: In the formula, Forecast load at a certain point in a day, For the weight vector, For transpose, This is the normalized vector of influencing factors for each time point corresponding to the predicted date. For the first The input vector at a certain moment on a sample day. Will Nonlinear mapping to high-dimensional space, b The bias constant is For Lagrange multipliers, The number of similar day samples selected. For kernel functions; in, , For the first The historical load normalized data of the previous day corresponding to each sample day is one of the selected low-frequency components or two high-frequency components. For the first The normalized vector of M influencing factors corresponding to the daily time; M is the number of influencing factors; Kernel function The calculation formula is as follows: In the formula, These are the parameters for the kernel function.

2. The method according to claim 1, characterized in that: The factors affecting the power load of the transformer area include meteorological factors and daily type factors.

3. The method according to claim 1, characterized in that: The method for predicting the power load of the distribution area also includes evaluating the predicted load using relative error and absolute error.

4. A system for implementing the power dispatching method for distribution transformer areas as described in any one of claims 1-3, characterized in that... It includes a screening module, a decomposition module, a prediction module, and a scheduling module; The filtering module filters out historical load data for several similar days from historical data based on pre-set influencing factors of the power load of the distribution area and the obtained influencing factor values ​​for the predicted day. The decomposition module is used to decompose the selected historical load data; The prediction module obtains the predicted daily load components based on the decomposed load components. The scheduling module performs power energy scheduling for the distribution area based on the predicted load components.

Images