Power load prediction method based on short memory grey model

Abstract

The invention discloses a power load prediction method based on a short-memory grey model. The method comprises the following steps: constructing a power load short-memory grey prediction model; converting the power load short memory grey prediction model into an algebraic model; establishing a model parameter identification algorithm by combining deep learning based on the historical process data of the power load and the optimal variance criterion; and the power load short memory grey prediction model is used to realize power load rolling prediction. The electric power grey model prediction method with the memory attenuation characteristic is designed, the calculation complexity and the prediction precision are both considered, the physical process of the electric power demand in the electric power system can be reflected better, and the electric power load analysis is more accurate.

Description

Technical Field

[0001] The invention relates to the technical field of power system load forecasting, and in particular to a power load forecasting method based on a short memory grey model. Background Art

[0002] Modern society is increasingly dependent on electricity supply. With the rapid development of science and technology and economy, the demand for electricity is growing rapidly, and the contradiction between energy supply and demand is becoming increasingly prominent. Although the demand for electricity is gradually increasing, it has a certain elasticity due to factors such as seasonal changes and economic development fluctuations. However, as a secondary energy source, electricity cannot be stored in large quantities. Therefore, it is of great significance to predict electricity demand, adjust electricity supply, and ensure the balance of electricity supply and demand.

[0003] Electricity demand is affected by many factors such as industrial structure, seasonal changes, and economy, and these factors are affected by many other factors and change, and are extremely complex. It is difficult to establish an electricity demand forecasting model based on industry, season, economy and other factors.

[0004] The existing technology has studied the grey model and proposed the GM (1,1) model, DGM (1,1) model, FGM (1,1) model, DTDFF-DGM (1,1) model, etc. The research results are applied to power forecasting. These methods use historical data weighting to predict power demand and reduce the prediction error. However, from a system perspective, the system output depends on all historical inputs, but the outputs at different times have different weights, and this weight decays over time. Similarly, power demand forecasting depends on historical data, but different historical data have different weights and different degrees of dependence. Therefore, the establishment of a power grey model prediction method with memory decay characteristics can better reflect the physical process of power demand and provide more accurate analysis.

[0005] Although all historical information will affect the prediction due to the memory characteristics, as time goes by, the memory decays, and the historical information beyond a certain time range can be discarded, taking into account both the computational complexity and the prediction accuracy. Based on this, the present invention proposes a power load forecasting method based on a short memory grey model. Summary of the invention

[0006] The purpose of the present invention is to provide a power load forecasting method based on a short memory grey model. In view of the problem that the traditional grey model does not take into account the characteristics of the system response process and ignores the attenuation characteristics of historical information, thereby affecting the accuracy of power load forecasting, starting from the system response characteristics, the memory decay characteristics are analyzed, and different weights are given to historical information according to the memory decay characteristics. A power load forecasting method based on a short memory grey model is proposed.

[0007] In order to achieve the above functions, the present invention designs a power load forecasting method based on a short memory grey model. For the target power system, the following steps S1 to S4 are executed to complete the power load rolling forecasting:

[0008] Step S1: Collecting the power load data of the target power system in the historical time period, and constructing a power load short-memory grey prediction model for predicting the power load of the target power system at a future moment;

[0009] Step S2: The power load short memory grey prediction model is converted into an algebraic model by representing the power load short memory grey prediction model through a matrix;

[0010] Step S3: Based on the power load data of the target power system in the historical time period and the optimal variance criterion, a cost function is constructed, and a model parameter identification algorithm is established in combination with deep learning to obtain the parameters to be identified of the power load short-memory grey prediction model;

[0011] Step S4: Collect the power load data of the target power system, and use the power load short-memory grey prediction model to implement a rolling power load prediction for the target power system.

[0012] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0013] 1. The present invention proposes a power load forecasting method based on a short memory grey model. The method studies the memory characteristics of the forecasting model according to the system response characteristics, and assigns different weights to historical data. The weight depends on the system response characteristics. The more recent the data, the greater the weight, and the more distant the data, the smaller the weight. This method is more in line with the physical mechanism of power load changes and the prediction is more accurate.

[0014] 2. The present invention utilizes the memory decay characteristics. The older the historical data, the smaller the weight is, and the smaller the impact on the output is. Based on this, appropriate choices are made and a memory model of a certain length is intercepted for prediction. This method can not only ensure the prediction accuracy, but also reduce the complexity of the prediction algorithm.

[0015] 3. The present invention intercepts a certain length of historical information for prediction based on the memory characteristics, and this length is fixed, which is convenient for realizing rolling prediction of power load and for rapid realization using FFT algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a power load forecasting method based on a short memory grey model provided according to an embodiment of the present invention;

[0017] Figure 2 is a flowchart of constructing a short-memory grey prediction model for power load provided in an embodiment of the present invention;

[0018] Figure 3 is a flow chart of converting a short memory grey prediction model of electric load into an algebraic model according to an embodiment of the present invention;

[0019] Figure 4 is a simulation diagram of a short-memory grey prediction model of electric load provided according to an embodiment of the present invention;

[0020] Figure 5 is a flow chart of a model parameter identification algorithm provided according to an embodiment of the present invention;

[0021] Figure 6 is a flow chart of rolling prediction of power load provided according to an embodiment of the present invention;

[0022] Figure 7 It is a comparison chart between actual value and predicted value of power load in a city provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0024] An electric load forecasting method based on a short memory grey model provided by an embodiment of the present invention is directed to a target power system, referring to Figure 1 , execute the following steps S1 to S4 to complete the rolling forecast of power load:

[0025] Step S1: Collecting the power load data of the target power system in the historical time period, and constructing a power load short-memory grey prediction model for predicting the power load of the target power system at a future moment;

[0026] Reference Figure 2 , the specific steps of step S1 are as follows:

[0027] Step S1.1: Assume that the sampling interval is T, and collect the power load data of the target power system as follows:

[0028] x(0),x(1),x(2),…,x(n)

[0029] In the formula, x(0), x(1), x(2), …, x(n) represent the power load at the initial time, 1T, 2T, …, nT respectively. For the power load at (n+1)T, the power load memory grey prediction model is established as follows:

[0030] x(n+1)=(b1(nT) β +b2)*x(n)

[0031] Where, (nT) βrepresents the sampling at time nT, b1, b2 and β are the parameters to be identified, and the symbol * represents the convolution operation;

[0032] Step S1.2: Expand the power load memory grey prediction model into a cumulative expression as follows:

[0033]

[0034] The above relationship reflects the memory characteristics of the power load memory grey prediction model for historical information and the different memory weights of historical information at different times. The power load changes slowly, and β is less than 0, otherwise the power load demand will diverge exponentially, which is an unstable system and cannot be predicted.

[0035] Step S1.3: Since β is less than 0, the memory weight decays with time, and the weight of historical information after a certain period of time can be ignored, which reduces the computational complexity within the permitted range of prediction accuracy. According to the memory decay characteristics, the power load data within the historical time period with a memory length of L is taken to obtain the short-memory grey prediction model of power load:

[0036]

[0037] Wherein, St represents the starting point of the short-memory grey prediction model of power load. When nL≥0, St=nL; when nL<0, St=0; L represents the memory length.

[0038] Step S2: Refer to Figure 3 , the power load short memory grey prediction model is transformed into an algebraic model by representing the power load short memory grey prediction model through a matrix;

[0039] The specific steps of step S2 are as follows:

[0040] Step S2.1: The power load memory grey prediction model is expressed as follows using a matrix:

[0041]

[0042] Among them, the memory matrix satisfy:

[0043]

[0044] in, Represents the memory matrix The element at row i, column j:

[0045]

[0046] Step S2.2: The power load short memory grey prediction model is expressed as follows using a matrix:

[0047]

[0048] Among them, the short memory matrix satisfy:

[0049]

[0050] in, Represents a short memory matrix The element at row i and column j;

[0051]

[0052] Thus, the short-memory grey prediction model of power load is obtained.

[0053] In one embodiment, let h(t)=t -0.3 , y2(t)=[h(t)(u(t)-u(t-3))]*[t -0.3 (u(t)-u(t-3))], y3(t)=[h(t)(u(t))]*[t -0.3 (u(t))], e1(t)=y3(t)-y2(t), where u(t) represents the unit step function, t -0.3 (u(t)-u(t-3)) represents the interception of t -0.3 A historical information with a length of 3 from the current time forward, refer to Figure 4 ,The simulation results show that: due to the memory decay characteristics, ,the short memory model obtained by intercepting a period of historical ,information will not significantly affect the error accuracy.

[0054] Step S3: Refer to Figure 5 , based on the power load data of the target power system in the historical time period and the optimal variance criterion, a cost function is constructed, and a model parameter identification algorithm is established in combination with deep learning to obtain the parameters to be identified of the power load short-memory grey prediction model;

[0055] The specific steps of step S3 are as follows:

[0056] Step S3.1: Construct the cost function Cost(x) as follows:

[0057]

[0058] in, is the power load forecast value, obtained by solving step S2.2, and x(n) is the actual power load data collected; that is:

[0059]

[0060] Step S3.2: Identify the short memory matrix The parameters to be identified in make the cost function optimal.

[0061] In step S3.2, the cuckoo optimization search algorithm is used to identify the short memory matrix The parameters to be identified are b1, b2 and β in the figure. The specific steps are as follows:

[0062] Step S3.2.1: Generate M bird nest locations for the parameters to be identified P{b1, b2, β}, and calculate the cost function corresponding to the bird nest location according to the bird nest location and step S3.1;

[0063] Step S3.2.2: Update the nest position according to the following formula:

[0064] AP i g =P i g +Le′vy(λ)·Sp

[0065] Where Sp = [Sp1 Sp2 Sp3] represents the step vector, P i g represents the location of the i-th nest in the g-th generation, AP i g represents the updated position of the i-th bird's nest in the g-th generation, Sp1, Sp2, Sp3 correspond to the step lengths of the parameters to be identified b1, b2 and β respectively; Le′vy(λ) represents a random walk with a step length that follows the Le′vy distribution; represents a random walk of each bird's nest position near the original position;

[0066] Step S3.2.3: Calculate the cost function corresponding to the bird's nest position according to the updated bird's nest position in step S3.2.2 and step S3.1; compare the cost function corresponding to the updated bird's nest position with the saved optimal cost function, select the minimum cost function for update, and globally update the optimal bird's nest position and the corresponding cost function;

[0067] Step S3.2.4: Generate new nest location:

[0068]

[0069] Among them, JP i g represents a new solution for local search, and represents two random nest positions r1 and r2 in the g-th generation of bird nests, is a random number uniformly distributed in the interval [0,1], H(·) represents the Heaviside function;

[0070] Step S3.2.5: Calculate the cost function corresponding to the bird's nest position according to the bird's nest position generated in step S3.2.4 and step S3.1; compare the cost function corresponding to the updated bird's nest position with the saved optimal cost function, select the minimum cost function for update, and locally update the optimal bird's nest position and the corresponding cost function;

[0071] Step S3.2.6: Jump to step S3.2.2 and repeat the search until the cost function is less than the given threshold, end the loop, save the final bird's nest position as the optimal bird's nest position, and use the bird's nest position parameters as the identification results of the parameters to be identified b1, b2 and β.

[0072] Step S4: Refer to Figure 6 , collect the power load data of the target power system, and use the power load short-memory grey prediction model to realize the power load rolling prediction for the target power system.

[0073] The specific steps of step S4 are as follows:

[0074] Step S4.1: Collect the power load data of the target power system, and retain the power load data in the historical time period with a memory length of L. When the power load data in the historical time period is less than L, the insufficient part is filled with 0;

[0075] Step S4.2: Identify the parameters b1, b2 and β to be identified according to step S3, and obtain a short-memory grey prediction model of power load based on the power load data in the current historical time period;

[0076] Step S4.3: Obtain the power load forecast value based on the power load data forecast in the current historical time period Shift the power load data in the historical time period to the left by one data point, remove the leftmost data point, and Move it to the rightmost end (if there is a new measured value x(n+1), use the measured value to move it to the rightmost end, and the measured value has a higher priority than the predicted value), and get a new set of power load data within the historical time period with a memory length of L; then identify the parameters b1, b2 and β of the power load short-memory grey prediction model based on the new data, make the next prediction based on the identification result, and realize the rolling prediction of power load in a cycle.

[0077] The following is an application embodiment of the present invention:

[0078] The collected data of power grid load in a certain city from 2001 to 2011 are shown in Table 1:

[0079] Table 1

[0080] Serial number years Electricity consumption (billion kWh) 1 2001 532.05 2 2002 641.27 3 2003 644.36 4 2004 841.26 5 2005 860.35 6 2006 944.36 7 2007 1120.51 8 2008 1215.46 9 2009 1326.61 10 2010 1549 11 2011 1751

[0081] The short memory grey prediction model of power load with memory length L=6 is used for rolling prediction. The simulation results of predicted value and actual value are as follows: Figure 7 shown.

[0082] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A power load forecasting method based on short memory grey model, characterized in that: For the target power system, execute the following steps S1 to S4 to complete the power load rolling forecast: Step S1: Collect the power load data of the target power system in the historical time period, and construct a short-memory grey prediction model for predicting the power load of the target power system at a future time; Step S2: The power load short memory grey prediction model is converted into an algebraic model by representing the power load short memory grey prediction model through a matrix; Step S3: Based on the power load data of the target power system in the historical time period and the optimal variance criterion, a cost function is constructed, and a model parameter identification algorithm is established in combination with deep learning to obtain the parameters to be identified of the power load short-memory grey prediction model; Step S4: Collect the power load data of the target power system, and use the power load short-memory grey prediction model to implement a rolling power load prediction for the target power system.

2. The power load forecasting method based on the short memory grey model according to claim 1 is characterized in that: The specific steps of step S1 are as follows: Step S1.1: Assume that the sampling interval is T, and collect the power load data of the target power system as follows: x(0),x(1),x(2),…,x(n) In the formula, x(0), x(1), x(2), …, x(n) represent the power load at the initial time, 1T, 2T, …, nT respectively. For the power load at (n+1)T, the power load memory grey prediction model is established as follows: x(n+1)=(b1(nT) β +b2)*x(n) Where, (nT) β represents the sampling at time nT, b1, b2 and β are the parameters to be identified, and the symbol * represents the convolution operation; Step S1.2: Expand the power load memory grey prediction model into a cumulative expression as follows: Step S1.3: Take the power load data in the historical time period with a memory length of L and obtain the power load short-memory grey prediction model: Wherein, St represents the starting point of the short-memory grey prediction model of power load. When nL≥0, St=nL; when nL<0, St=0; L represents the memory length.

3. The power load forecasting method based on short memory grey model according to claim 1 is characterized in that: The specific steps of step S2 are as follows: Step S2.1: The power load memory grey prediction model is expressed as follows using a matrix: Among them, the memory matrix satisfy: in, Represents the memory matrix The element at row i, column j: Step S2.2: The power load short memory grey prediction model is expressed as follows using a matrix: Among them, the short memory matrix satisfy: in, Represents a short memory matrix The element at row i and column j; Thus, the short-memory grey prediction model of power load is obtained.

4. The power load forecasting method based on short memory grey model according to claim 1 is characterized in that: The specific steps of step S3 are as follows: Step S3.1: Construct the cost function Cost(x) as follows: in, is the power load forecast value, obtained by solving step S2.2, and x(n) is the actual power load data collected; that is: Step S3.2: Identify the short memory matrix The parameters to be identified in make the cost function optimal.

5. The power load forecasting method based on short memory grey model according to claim 4 is characterized in that: In step S3.2, the cuckoo optimization search algorithm is used to identify the short memory matrix The parameters to be identified are b1, b2 and β in the figure. The specific steps are as follows: Step S3.2.1: Generate M bird nest locations for the parameters to be identified P{b1, b2, β}, and calculate the cost function corresponding to the bird nest location according to the bird nest location and step S3.1; Step S3.2.2: Update the nest position according to the following formula: AP i g =P i g +Le′vy(λ)·Sp Where Sp = [Sp1 Sp2 Sp3] represents the step vector, P i g represents the location of the i-th nest in the g-th generation, AP i g represents the updated nest position of the i-th generation, Sp1, Sp2, Sp3 correspond to the step lengths of the parameters to be identified b1, b2 and β respectively; Le′vy(λ) represents the random walk whose step length follows the Le′vy distribution; Step S3.2.3: Calculate the cost function corresponding to the bird's nest position according to the updated bird's nest position in step S3.2.2 and step S3.1; compare the cost function corresponding to the updated bird's nest position with the saved optimal cost function, select the minimum cost function for update, and globally update the optimal bird's nest position and the corresponding cost function; Step S3.2.4: Generate new nest location: Among them, JP i g represents a new solution for local search, and represents two random nest positions r1 and r2 in the g-th generation bird nest, is a random number uniformly distributed in the interval [0,1], H(·) represents the Heaviside function; Step S3.2.5: Calculate the cost function corresponding to the bird's nest position according to the bird's nest position generated in step S3.2.4 and step S3.1; compare the cost function corresponding to the updated bird's nest position with the saved optimal cost function, select the minimum cost function for update, and locally update the optimal bird's nest position and the corresponding cost function; Step S3.2.6: Jump to step S3.2.2 and repeat the search until the cost function is less than the given threshold, end the loop, save the final bird's nest position as the optimal bird's nest position, and use the bird's nest position parameters as the identification results of the parameters to be identified b1, b2 and β.

6. The power load forecasting method based on short memory grey model according to claim 1 is characterized in that: The specific steps of step S4 are as follows: Step S4.1: Collect the power load data of the target power system, and retain the power load data in the historical time period with a memory length of L. When the power load data in the historical time period is less than L, the insufficient part is filled with 0; Step S4.2: Identify the parameters b1, b2 and β to be identified according to step S3, and obtain a short-memory grey prediction model of power load based on the power load data in the current historical time period; Step S4.3: Obtain the power load forecast value based on the power load data forecast in the current historical time period Shift the power load data in the historical time period to the left by one data point, remove the leftmost data point, and Move to the rightmost end to obtain a new set of power load data within a historical time period with a memory length of L; then identify the parameters b1, b2 and β of the power load short-memory grey prediction model based on the new data, make the next prediction based on the identification results, and realize the rolling prediction of power load in a cycle.

7. The power load forecasting method based on short memory grey model according to claim 6 is characterized in that: If there is a new power load measurement value x(n+1) in step S4.3, the measurement value x(n+1) is moved to the rightmost end of the power load data in the historical time period. The power load measurement value x(n+1) has a higher priority than the power load forecast value. priority.

Images