CN106971238A - The Short-Term Load Forecasting Method of Elman neutral nets is obscured based on T S - Google Patents

Abstract

The invention relates to a short-term power load forecasting method based on T-S fuzzy Elman neural network, comprising the following steps: obtaining historical load data of a power system in a certain area, processing abnormal data of the historical load data; Analysis and quantification, normalize the corrected data; determine the input and output data of the neural network, introduce a delay unit in the rule layer, and use the output of the rule layer, that is, the activation intensity of all the rules at the previous moment, as the input information at the current moment , so as to establish a T-S fuzzy Elman neural network, use the trained T-S fuzzy Elman neural network to predict, and denormalize the predicted data to obtain the final forecast load value. The invention can well fit the nonlinear, dynamic and time-varying characteristics of the electric load system, has high prediction accuracy, and can be widely used in the short-term load prediction of the electric power system.

Description

Short-term Power Load Forecasting Method Based on T-S Fuzzy Elman Neural Network

technical field

The invention relates to a short-term power load forecasting method based on T-S fuzzy Elman neural network, which belongs to the field of power system load forecasting.

Background technique

Power system load forecasting can be divided into medium and long-term load forecasting, short-term load forecasting and ultra-short-term load forecasting according to the classification of forecasting period. Among them, short-term load forecasting refers to the research on load forecasting for each time period of each day in the next day to a week. Short-term load forecasting is very important in the research of power system load forecasting. The accuracy of its forecasting directly affects the safe, economical and stable operation of the power system and the realization of scientific management and dispatching of the power grid. At present, the backpropagation method (BP algorithm) of the artificial neural network algorithm is mainly used, which has been widely used in the research direction of power system load forecasting. Since the power system load is easily affected by climate, economic and other factors, it is dynamic, and the BP neural network converts the dynamic modeling problem into a static modeling problem, which will make the network run into local minimum and one-way propagation. No questions about feedback etc.

Contents of the invention

The technical problem to be solved by the invention is: to better fit the nonlinear, dynamic and time-varying characteristics of the electric load system.

In order to solve the problems of the technologies described above, the technical solution of the present invention provides a kind of short-term power load forecasting method based on T-S fuzzy Elman neural network, it is characterized in that, comprises the following steps:

Step 1. Obtain the historical load data of the power system in a certain area, and process the abnormal data of the historical load data;

Step 2. Analyze and quantify the factors affecting the electric load, and normalize the corrected load data;

Step 3. Determine the input and output data of the neural network, in which the weather characteristics, temperature, date type and load value of t-1 hour of the forecast day and the n-1 and n-2 days t and t-1 of the forecast time and the load value of hour t+1 as input data, the load value of the hour t hour of the predicted day is the output data, and determine the optimal number of neurons in the hidden layer, introduce a delay unit in the regular layer, and convert the regular layer The output of the previous moment is the activation intensity of all the rules as the current input information, so as to establish a T-S fuzzy Elman neural network;

Step 4. Use the historical load data, weather parameter data and date type data of the two months before the forecast date for training, use the trained T-S fuzzy Elman neural network for forecasting, and denormalize the forecasted data to obtain the final forecast load value;

Preferably, in the step 1, historical electricity load data of a certain region is obtained, and the data samples thereof come from a SCADA system.

Preferably, the abnormal data processing in step 1 includes horizontal processing, vertical processing or curve fitting.

Preferably, the factors affecting the load in the step 2 include temperature, weather characteristics and date type, and quantify them according to the degree of influence of these factors on the load.

Preferably, in the step 2, the load data is normalized, and the normalization formula is used to normalize the load data to [0, 1], so that they are at the same magnitude level, so as to speed up the convergence of the neural network.

Preferably, in the determination of the optimal number of hidden layer neurons in step 3, the hidden layer of the network is a single hidden layer, and the number of neurons is determined according to empirical formulas and training effects.

Preferably, in said step 4, the predicted data is denormalized to obtain the final forecasted load value, and the denormalized formula can be obtained according to the deformation of the normalized formula, and the final data is Load data of the actual order of magnitude.

Elman neural network is a typical dynamic neuron network, which has the ability to adapt to time-varying characteristics. At the same time, T-S fuzzy control enables the output of the system to be expressed as a linear combination of input variables, so the system can well fit the power load system Non-linear, dynamic and time-varying characteristics.

Due to the adoption of the above-mentioned technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention combines T-S fuzzy control with Elman neural network and applies it to short-term power load forecasting, and the model has both Elman The advantages of neural network and fuzzy control not only have a strong dynamic nonlinear fitting ability, but also better simulate the dynamic process of error feedback correction, which can better fit the nonlinear, dynamic and time-sensitive characteristics of the power load system. Due to the characteristics of variability and high forecasting accuracy, it can be widely used in short-term load forecasting of power systems.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is a structural diagram of the T-S fuzzy Elman neural network of the present invention.

detailed description

Below in conjunction with specific embodiment, further illustrate the present invention. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the teachings of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

Embodiments of the present invention relate to a short-term power load forecasting method based on T-S fuzzy Elman neural network, as shown in Figure 1, comprising the following steps:

(1) Obtain the historical load data of the power system in a certain area, and process the abnormal data of the historical load data; among them, the historical load data samples mainly come from the SCADA system; the abnormal data is due to the interference of other factors, and the system will be incomplete data, and there are erroneous data.

(2) Analyze and quantify the factors that affect the power load, and normalize the corrected data; among them, the factors that affect the power load include temperature, weather characteristics, and date types, etc., according to the degree of influence of these factors on the load. Quantization processing; normalization is to use the normalization formula to normalize the load data to [0, 1], so that it is at the same level of magnitude and speed up the convergence of the neural network.

(3) Determine the input and output data of the neural network, and determine the optimal number of neurons in the hidden layer, introduce a delay unit in the rule layer, and use the output of the rule layer, that is, the activation intensity of all the rules at the previous moment, as the current moment The input information is used to establish a T-S fuzzy Elman neural network; among them, the input and output data are the weather characteristics, temperature, date type and t-1 hour load value of the forecast day and the n-1 and n-2 days of the forecast time The load values at hours t, t-1, and t+1 are used as input data, and the load values at the hour t of the forecast day are output data; the number of neurons in the hidden layer is determined according to the empirical formula and the effect of training; The delay unit takes the output of the rule layer, that is, the activation intensity of all the rules at the previous moment, as the input information at the current moment, and can also be regarded as a memory unit.

(4) Use the historical load data, weather parameter data and date type data of the two months before the forecast date for training, use the trained T-S fuzzy Elman neural network for forecasting, and denormalize the forecasted data to obtain the final forecast Load value; wherein, the denormalization is the denormalization formula obtained by deforming the normalization formula, so as to obtain the load data of the actual order of magnitude.

The present invention will be further described below by taking the historical load data of a prefecture-level city in a certain coastal area as the research object as an example. The specific steps are as follows:

Acquisition of historical load data and data processing

In the current power system, the load data of the power system mainly comes from the SCADA system. Due to various reasons, the data of the SCADA system is not complete and there are some wrong data. For these abnormal data, it is necessary to adopt certain methods to detect and correct the data.

Vertical processing of data: The load of the power system has a certain periodicity, and the load at the same time on similar days has a certain similarity, and its variation range is kept within a certain range. If it exceeds this range, it can be judged as abnormal data. Through the vertical processing of the load, some abnormal data can be detected. By smoothing the mutation load, the data fluctuation can be reduced.

Horizontal processing of data: In the load data of the power system, the load at similar times will not change drastically. Therefore, the maximum error range of the data can be set based on the load data at two moments before and after as the benchmark value. If the absolute value of the difference between the load value and the load data at two moments before and after exceeds the threshold, it can be determined that the load value is bad data.

Quantify the influencing load factors and normalize the corrected data

In the electricity market, the load of the power system is affected by many factors, not only by power load demand, weather conditions, seasonality, and geographical factors, but also by factors such as national economy, politics, and residents' living habits. The main factors affecting this coastal area are temperature, weather characteristics and date type. When the temperature changes within a certain range, the influence on the load change is basically similar, so a quantized value between [0, 1] is given at different temperatures. For the weather characteristics, the region has four distinct seasons, so according to the historical load data and seasonal characteristics, a quantitative value between [0, 1] from large to small is given from sunny to cloudy to heavy rain. For the date type, it is mainly quantified according to working days and rest days.

Constructing the Model of T-S Fuzzy Elman Neural Network

The T-S fuzzy Elman neural network consists of five layers, which are input layer, membership function layer, rule layer, parameter layer, and output layer, with a total of n input nodes and one output node. Its topology is shown in Figure 2.

The input layer is the input data, including the weather characteristics, temperature, date type and t-1 hour load value of the forecast day, and the t, t-1 and t+1 hour loads of the n-1 and n-2 days at the forecast time value; membership function layer, each node in this layer represents a membership function, and the connection weight with the input layer is 1, using the Gaussian function as the membership function to calculate the membership degree corresponding to each node; the rule layer introduces a delay Unit, which takes the output of the rule layer, that is, the activation intensity of all the rules at the previous moment, as the input information at the current moment; the parameter layer, which belongs to the subsequent part of the T-S fuzzy neural network; the output layer, which realizes the defuzzification function of the T-S fuzzy system , to get the output of a linear combination.

Realization of Short-term Load Forecasting

Taking the data of May and June 2012 in this area as the original sample data, after abnormal data processing and normalization, a set of normal data for neural network learning is obtained, and the constructed T-S fuzzy neural network is used to , and finally get the load value of 24 hours a day on the forecast day. After inspection and comparison, the average absolute error can be controlled within 2%, and the MSE can be controlled within 0.6%. Therefore, the model can predict short-term power load data well.

Claims (7)

1. a short-term power load forecasting method based on T-S fuzzy Elman neural network, is characterized in that, comprises the following steps: Step 1. Obtain the historical load data of the power system in a certain area, and process the abnormal data of the historical load data; Step 2. Analyze and quantify the factors affecting the electric load, and normalize the corrected load data; Step 3. Determine the input and output data of the neural network, in which the weather characteristics, temperature, date type and load value of t-1 hour of the forecast day and the n-1 and n-2 days t and t-1 of the forecast time and the load value of hour t+1 as input data, the load value of the hour t hour of the predicted day is the output data, and determine the optimal number of neurons in the hidden layer, introduce a delay unit in the regular layer, and convert the regular layer The output of the previous moment is the activation intensity of all the rules as the current input information, so as to establish a T-S fuzzy Elman neural network; Step 4. Use the historical load data, weather parameter data and date type data of the two months before the forecast date for training, use the trained T-S fuzzy Elman neural network for forecasting, and denormalize the forecasted data to obtain the final forecast load value. 2. the short-term electric load forecasting method based on T-S fuzzy Elman neural network according to claim 1, is characterized in that, in the described step 1, obtains the electric historical load data of certain area, and its data sample comes from SCADA system. 3. The short-term power load forecasting method based on T-S fuzzy Elman neural network according to claim 1, characterized in that the abnormal data processing in the step 1 includes horizontal processing, vertical processing or curve fitting. 4. the short-term power load forecasting method based on T-S fuzzy Elman neural network according to claim 1, is characterized in that, the influence load factor in described step 2 comprises temperature, weather feature and date type, according to these factors to load The degree of influence will be quantified. 5. the short-term power load forecasting method based on T-S fuzzy Elman neural network according to claim 1, is characterized in that, in described step 2, load data normalization, use normalization formula load data normalization to [ 0, 1], making it at the same level of magnitude, speeding up the convergence of the neural network. 6. the short-term power load forecasting method based on T-S fuzzy Elman neural network according to claim 1, is characterized in that, in the number of optimal hidden layer neurons determined in the described step 3, the implicit of this network The layer is a single hidden layer, and the number of its neurons is determined according to the empirical formula and the effect of training. 7. the short-term power load forecasting method based on T-S fuzzy Elman neural network according to claim 1, is characterized in that, in described step 4, the data denormalization of prediction obtains final forecasted load value total, and its regression According to the deformation of the normalization formula, the anti-normalization formula can be obtained, and the final data is the load data of the actual order of magnitude.