Power system load forecasting refers to fully considering some important system operating characteristics and work decisions of the power system Under the conditions of , natural conditions and social influence, study or use mathematical methods to systematically process historical load data, and determine the load value at a specific time in the future under the premise of meeting certain precision and accuracy. Improving the technical level of load forecasting is conducive to the management of power planning, the rational arrangement of the power system operation mode and maintenance plan, the realization of energy saving and cost reduction, and the formulation of a reasonable power system construction plan to improve the economic and social benefits of the power system .

      Short-term load forecasting plays a key role in the entire power dispatching and electricity retailing companies' participation in the spot market , it provides basic data for basic power generation planning, computer online power grid control, system security analysis, day-ahead market quotations, intraday (real-time) market transactions, etc. Reduce the deviation between the reported electricity consumption and the actual electricity consumption.

      In terms of scale, load forecasting can be divided into grid-level load forecasting and user-level load forecasting . Power grid companies mainly use load forecasting at the grid level, which is of decisive significance to the stable and safe operation of the entire power system and the improvement of system operation economy. For electricity retail companies, load forecasting at the customer level is more relevant. Electricity retail companies need to make power load forecasts for their own user combinations and single users, and then carry out other actions based on the results of this forecast.

      Compared with the grid level, load forecasting at the user side level is more difficult. Its load curve fluctuates violently and has great randomness, which is difficult to predict. Although user-level load forecasting is more difficult, it is unavoidable for all electricity sales companies and virtual power plant companies, especially for short-term (more than one day and within two weeks) and ultra-short-term (within one day) user-level load forecasting. The following is to focus on short-term and ultra-short-term load forecasting methods.

● Basic methods of load forecasting

      Influence, in the time series, it is a non-stationary random process, but most of the factors affecting the system load have regularity, thus laying the foundation for realizing effective forecasting. At present, there are many methods for short-term load forecasting. The relatively new algorithms mainly include neural network method, time series method, regression analysis method, support vector machine method, fuzzy prediction method and so on. The core issue of power load forecasting research is how to use the existing historical data to establish a forecasting model to predict the load value at a time or in the future. Therefore, the reliability of historical data information and the forecasting model affect the accuracy of short-term load forecasting. main factor. With the gradual establishment of the current power system management information system and the improvement of the weather forecast level, it is no longer difficult to accurately obtain various historical data. Therefore, the core issue of short-term load forecasting is the level of the forecasting model. The following are various loads A brief description of the forecasting method:

Neural network method (artificial intelligence)

      The neural network method is currently the most advanced load forecasting method. As a kind of artificial intelligence algorithm, neural network has been widely used in image recognition, natural language processing, machine translation, automatic driving and so on. The most important artificial intelligence algorithms of well-known domestic and foreign artificial intelligence companies such as Google, Baidu, Ali, and iFLYTEK are neural networks. Similarly, neural networks are also widely used in the energy field, not only in power load forecasting, but also in power spot market price forecasting, wind power generation forecasting and other fields.

The application of neural network method in load forecasting is mainly divided into artificial neural network (Artificial Neural Networks, hereinafter referred to as ANN) and recursive Neural Networks (Recurrent Neural Networks, also known as recurrent neural networks, hereinafter referred to as RNN). And long-short-term memory network (Long-Short Term Memory hereinafter referred to as LSTM) and other specialized neural networks suitable for processing long and short cycle characteristics.

      The neural network method selects the load of the past period of time as a training sample, constructs a suitable network structure, and uses After a certain training algorithm trains the network to meet the accuracy requirements, this neural network is used as a load forecasting model. Practice has proved that artificial neural network has better accuracy in short-term forecasting. The advantage of artificial neural network is that it can adapt to a large number of non-structural and inaccurate laws. It has the characteristics of information memory, autonomous learning, knowledge reasoning and optimized calculation, as well as strong computing power, complex mapping ability, and fault tolerance. And various intelligent processing capabilities, especially its learning and adaptive functions are not available in other algorithms. The shortcomings of the neural network method lie in the construction of the model structure, the optimization of the learning speed, and the local minimum point.

Time series method

      The historical data of electric load is an ordered collection that is sampled and recorded at certain time intervals. Therefore, it is a time series. The time series method is a relatively mature algorithm developed in the short-term load forecasting of the power system. According to the historical data of the load, a mathematical model describing the change of the power load with time is established, and the load forecasting method is established on the basis of the model. expression, and forecast the future load. The advantage of the time series method is that it requires less data and less workload; the calculation speed is faster; it reflects the continuity of recent load changes. The shortcomings of the time series method are that the modeling process is relatively complicated and requires high theoretical knowledge; the model has high requirements for the stationarity of the original time series, and is only suitable for short-term forecasts with relatively uniform load changes; it does not consider the influence of load changes. Insufficient consideration of uncertain factors (such as weather, holidays, etc.), when the weather changes greatly or encounters holidays, the prediction error of the model is large.

Regression analysis

      The regression analysis and forecasting method is to find the self- The correlation between variables and dependent variables and their regression equations are used to determine the model parameters and to infer the load value in the future. Its advantages are simple calculation principle and structure, fast prediction speed, good extrapolation performance, and better prediction for situations that have not occurred in history. The disadvantages are that the requirements for historical data are relatively high, and the linear method is used to describe more complex problems. The structure is too simple and the accuracy is low; the model cannot describe various factors that affect the load in detail, and the model initialization is difficult and requires a wealth of information. experience and high skill.

support vector machine

      Support Vector Machine (SVM) is a method based on statistical learning theory that can be implemented in limited samples The machine learning method that satisfies the VC dimension theory and the principle of structural risk minimum under certain conditions has outstanding advantages such as strong generalization ability, global optimization and fast calculation speed. However, the selection of its optional parameters and kernel functions is usually determined mainly by experience, and there are relatively large human factors. At the same time, it lacks the ability to deal with fuzzy phenomena, and model errors will cause a gap between the regression value and the actual value.

Fuzzy prediction

     Fuzzy forecasting method is a new technology of load forecasting based on fuzzy mathematics theory, fuzzy mathematics The concept can describe some fuzzy phenomena in the power system, such as the key factors in load forecasting: the judgment of weather conditions, the division of load date types, etc. Applying fuzzy methods to load forecasting can better deal with the uncertainty of load changes sex. At present, there are mainly the following methods for applying fuzzy theory to load forecasting: fuzzy clustering method, fuzzy similarity priority ratio method and fuzzy maximum closeness method.

From the point of view of practical application, the accuracy of the simple fuzzy method for short-term load forecasting is difficult to meet the requirements; at the same time, it is required to provide more Historical data has difficulties in practical applications; its advantage is that the forecast results can be described in the form of forecast intervals and probabilities.

Other traditional methods

     There are many traditional methods of electric load forecasting not mentioned above, and the effects of these methods are similar to those of Compared with the new algorithm such as the neural network method, there is no advantage, so it will not be repeated here.