CN112215442B - Method, system, device and medium for predicting short-term load of power system - Google Patents
Method, system, device and medium for predicting short-term load of power system Download PDFInfo
- Publication number
- CN112215442B CN112215442B CN202011351479.5A CN202011351479A CN112215442B CN 112215442 B CN112215442 B CN 112215442B CN 202011351479 A CN202011351479 A CN 202011351479A CN 112215442 B CN112215442 B CN 112215442B
- Authority
- CN
- China
- Prior art keywords
- data
- load
- power system
- network model
- preset
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention belongs to the field of power system scheduling control, and discloses a method, a system, equipment and a medium for predicting short-term load of a power system, which comprise the following steps: acquiring load data and load influence data of a preset time period before a time period to be predicted of a power system; acquiring load influence data of a power system in a period to be predicted; and predicting through a preset load prediction network model according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of a preset period before the period to be predicted of the power system to obtain a load prediction result of the power system in the period to be predicted. The load prediction network model based on the time convolution network model can perform parallel data processing, so that the load prediction speed of the power system is greatly increased; meanwhile, the load prediction network model based on the time convolution network model is adopted for prediction, compared with the existing LSTM model, more historical information can be reserved, and the high accuracy requirement of the short-term load prediction of the power system is further guaranteed.
Description
Technical Field
The invention belongs to the field of power system scheduling control, and relates to a power system short-term load prediction method, a power system short-term load prediction system, power system short-term load prediction equipment and a power system short-term load prediction medium.
Background
The load prediction of the power system analyzes the change condition of the historical power load through research, and makes a previous estimation and conjecture on the power demand according to the change trend of influence factors such as economy, weather and the like, so that the power demand has great influence on the operation of the power system. With the deep promotion of electric power marketization, the relationship between system load prediction and the supply and demand of the electric power market is more and more compact, and the safety and economy of power grid operation are directly influenced. The load prediction is divided into long-term, medium-term, short-term and ultra-short-term prediction, wherein the short-term prediction refers to prediction of the load at each moment in the day, and plays an important role in power failure planning and power generation planning.
At present, short-term prediction is mainly based on deep learning theories such as a recurrent neural network (recurrent neural network), a long-term memory network (LSTM), and the like to perform load prediction. With the rapid development of deep learning, related algorithms are increasingly applied to the field of power grid scheduling control, wherein a recurrent neural network is also successfully applied to load prediction of a power system, and the recurrent neural network models sequence data and uses historical information and current input as current output influence factors together so as to represent time sequence change information of data, for example, open documents: short-term load prediction model based on Attention-LSTM in the electric power market [ J ] Power grid technology, 2019(5): 1745-: the method and the system for predicting the load of the power system based on the LSTM, which are provided by CN111697560A, both effectively improve the accuracy of load prediction.
However, the long-time and short-time memory networks are adopted in the methods, however, the long-time and short-time memory networks adopt a time series model, and the time t depends on the time t-1, and cannot be executed in parallel, so that the prediction speed is slow, and some historical information is gradually omitted by an internal forgetting gate, which reduces the prediction accuracy.
Disclosure of Invention
The invention aims to overcome the defects of low speed and low accuracy of short-term load prediction of the conventional power system in the prior art, and provides a method, a system, equipment and a medium for predicting the short-term load of the power system.
In order to achieve the purpose, the invention adopts the following technical scheme to realize the purpose:
in a first aspect of the present invention, a method for predicting a short-term load of an electrical power system includes the steps of:
acquiring load data and load influence data of a preset time period before a time period to be predicted of a power system;
acquiring load influence data of a power system in a period to be predicted;
according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of the power system in the preset period before the period to be predicted, predicting through a preset load prediction network model to obtain a load prediction result of the power system in the period to be predicted;
the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system.
The method for predicting the short-term load of the power system is further improved as follows:
the load impact data comprises meteorological data or meteorological data and holiday data.
The specific method for training the time convolution network model by adopting the historical data of the power system comprises the following steps: acquiring historical data of the power system, wherein the historical data comprises historical load data and historical load influence data; dividing historical data of the power system into a training set and a test set; training a time convolution network model through a training set to obtain an initial load prediction network model; and testing the initial load prediction network model through the test set to obtain a test result, evaluating the initial load prediction network model according to the test result, and taking the initial load prediction network model meeting the preset evaluation index as the load prediction network model.
The specific method for dividing the historical data of the power system into the training set and the test set comprises the following steps: preprocessing historical data of the power system to obtain preprocessed data; and dividing the preprocessed data into a training set and a test set according to a preset proportion.
The specific method for preprocessing the historical data of the power system to obtain the preprocessed data comprises the following steps: acquiring the type of each datum in the historical data of the power system; when the type of the data is a continuous variable, carrying out normalization processing on the data to obtain preprocessed data; and when the type of the data is a discrete variable, encoding the data, and mapping the data into a binary vector to obtain preprocessed data.
Before training the time convolution network model through the training set, the method further comprises the following steps: acquiring the data type of each data in the training set; and when the data type is sparse characteristic data, performing dimensionality reduction on the data to a preset dimensionality through an artificial neural network.
And when the dimension reduction processing is carried out on the preprocessed data through the artificial neural network, carrying out random inactivation processing on each hidden layer of the artificial neural network.
The specific method for evaluating the initial load prediction network model according to the test result comprises the following steps: and obtaining the average absolute percentage error and the mean square error of the test result according to the historical data and the test result of the power system, wherein when the average absolute percentage error and the mean square error are both smaller than the preset average absolute percentage error and the preset mean square error, the initial load prediction network model meets the preset evaluation index.
And when the time convolution network model is trained through the training set, all convolution layers of the time convolution network model are subjected to random inactivation treatment.
And when the time convolution network model is trained through the training set, carrying out batch standardization processing on convolution results output by the current convolution layer in the time convolution network model, and then inputting the convolution results into the next convolution layer.
When the time convolution network model is trained through the training set, the convolution result output by the last convolution layer in the time convolution network model is weighted according to the attention mechanism, the convolution result after the weighting processing is subjected to full-connection mapping, a mapping result is obtained, and the time convolution network model is optimized according to the mapping result.
When the prediction is carried out through the preset load prediction network model, the convolution result output by the current convolution layer in the load prediction network model is input into the next convolution layer after batch standardization processing.
When the prediction is carried out through the preset load prediction network model, the convolution result output by the last convolution layer in the load prediction network model is weighted according to the attention mechanism, and the weighted convolution result is subjected to full-connection mapping to obtain the load prediction result of the power system in the period to be predicted.
In a second aspect of the present invention, a system for predicting a short-term load of an electrical power system includes:
the first data acquisition module is used for acquiring load data and load influence data of a preset time period before a time period to be predicted of the power system;
the second data acquisition module is used for acquiring load influence data of the power system in a period to be predicted;
the prediction module is used for predicting through a preset load prediction network model according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of the power system in the preset period before the period to be predicted to obtain a load prediction result of the power system in the period to be predicted;
the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system.
In a third aspect of the present invention, a computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the power system short-term load prediction method when executing the computer program.
In a fourth aspect of the present invention, a computer readable storage medium stores a computer program which, when executed by a processor, implements the steps of the above-described power system short-term load prediction method.
Compared with the prior art, the invention has the following beneficial effects:
the method for predicting the short-term load of the power system comprises the steps of obtaining load data and load influence data of a preset time period before a time period to be predicted of the power system and the load influence data of the time period to be predicted of the power system, predicting through a preset load prediction network model according to the load influence data of the time period to be predicted of the power system and the load data and the load influence data of the preset time period before the time period to be predicted of the power system, wherein the preset load prediction network model is obtained by training a time convolution network model through historical data of the power system, and based on parallel execution characteristics of the time convolution network model, the load prediction network model can perform parallel execution data processing, so that the speed of load prediction of the power system is greatly increased. Meanwhile, as the existing LSTM model has a forgetting gate, some historical information can be gradually omitted by the forgetting gate in the prediction process; the convolution layer of the time convolution network model is in a causal relationship with the convolution layer, so that more historical information can be reserved by adopting the time convolution network model compared with the existing LSTM model, and the high accuracy requirement of the short-term load prediction of the power system is further ensured.
Drawings
FIG. 1 is a block diagram of a method for predicting short-term load of a power system according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for constructing a load prediction network model according to an embodiment of the present invention;
FIG. 3 is a block diagram of a load prediction network model according to an embodiment of the present invention;
FIG. 4 is a block diagram of sub-blocks in a time convolutional network model according to an embodiment of the present invention;
fig. 5 is a block diagram of a short-term load prediction system of an electric power system according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The invention is described in further detail below with reference to the accompanying drawings:
referring to fig. 1, in an embodiment of the present invention, a method for predicting a short-term load of an electrical power system is provided, where historical data of a preset time period before a time period to be predicted of the electrical power system is obtained, and then the obtained historical data is used as a prediction basis and is input into a load prediction network model which is constructed in advance and trained and is based on a time convolution network model to perform prediction, so as to obtain a load prediction result of the time period to be predicted of the electrical power system. Specifically, the power system short-term load prediction method comprises the following steps.
S1: load data and load influence data of a preset time period before a time period to be predicted of the power system are obtained.
Specifically, in this embodiment, the load data and the load influence data of the power system are acquired from the database of the power system by querying, where the load influence data may be a meteorological load of the power system, and may also additionally include holiday data, and preferably, in this embodiment, the load influence data includes the meteorological data and holiday data on the basis of the load data in consideration of influences of factors such as the meteorological load and holiday on the load of the power system.
The meteorological data can comprise data such as weather, temperature, humidity, wind power, wind direction, precipitation and body sensing temperature, the holiday data can comprise holidays such as spring festival, New year's day and double holiday and ordinary workdays, and the holidays, the three-day holidays and the seven-day holidays are divided into a working day, a double-holiday, a three-day holiday and a seven-day holiday through preprocessing.
S2: load influence data of a power system in a period to be predicted are obtained.
Specifically, in the embodiment, meteorological data and holiday data of a to-be-predicted time period of the power system are obtained, wherein the meteorological data are meteorological prediction data and can be obtained through a meteorological prediction center; holiday data may be determined based on actual forecasted regional conditions.
S3: according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of the power system in the preset period before the period to be predicted, predicting through a preset load prediction network model to obtain a load prediction result of the power system in the period to be predicted; the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system.
In this embodiment, the short-term load prediction of the power system is performed through the load prediction network model, and first, a method for constructing the load prediction network model by training the time convolution network model with the historical data of the power system is provided, which is shown in fig. 2 and 3 and specifically includes the following steps.
S301: historical data of the power system is acquired.
Specifically, in this embodiment, historical data of the power system in the last 5 years is acquired at 24 points per day at 1 hour intervals. Also, considering that the power system load has continuity and periodicity, a history load 24 hours or an integral multiple thereof before the time to be predicted may be generally used as an input.
Preferably, the weather data and the holiday data of the power system at the same time are read from the database in consideration of the influence of weather changes on air conditioning load and the influence of holidays on electricity consumption behavior of the user.
S302: historical data of the power system is divided into a training set and a test set.
Specifically, historical data of the power system is preprocessed to obtain preprocessed data; and dividing the preprocessed data into a training set and a test set according to a preset proportion.
When the historical data of the power system is preprocessed, different processing modes are selected according to the types of the data in the historical data, and specifically, the types of the data in the historical data of the power system are obtained; when the type of the data is a continuous variable, carrying out normalization processing on the data to obtain preprocessed data; and when the type of the data is a discrete variable, encoding the data, and mapping the data into a binary vector to obtain preprocessed data.
For example, when the data is continuous variable data such as temperature, precipitation, wind power, historical load, humidity and sensible temperature, the data is converted into the range of [0,1] by adopting maximum and minimum normalization processing according to the following formula:
wherein the content of the first and second substances,
which represents the original data of the image data,
represents the minimum value in the raw data,
which represents the maximum value in the original data,
the normalized data is represented.
When the data is discrete variable data such as date type, weather, wind direction, etc., in this embodiment, One-Hot encoding is adopted, and the discrete value is mapped to a binary vector. Here, taking the date type as an example, the date type = [ 'three-day vacation', 'seven-day vacation', 'two holiday', 'workday' ], and when the date type of the sample is 'three-day vacation', 'seven-day vacation', 'two holiday', 'workday', respectively, then the date feature codes are [1,0,0,0], [0,1,0,0], [0,0,1,0], [0,0,0,1] in this order.
In the embodiment, the preprocessing data of nearly five years are divided into training sets and testing sets according to the ratio of 8:2, and the training sets and the testing sets are respectively used for training a time convolution network model and testing the time convolution network model.
After the data is divided into a training set and a test set, preprocessing data in the training set and the test set are combined into a data sample for training or testing, when the data sample is combined, a time period to be predicted is determined, the time period to be predicted can be regarded as a certain time period randomly selected in historical time, and based on the continuity of the load of the power system in time, therefore, the load data and the load influence data in a preset time period before the time period to be predicted can influence the load data in the time period to be predicted, and therefore, the load data and the load influence data in the preset time period before the time period to be predicted can be used as an influence factor for prediction.
Then each data sample in the training set and the test set may contain: load data and load influence data in a preset time period before the time period to be predicted, and load influence data in the time period to be predicted. Then, since the load data of the period to be predicted is known in fact due to the historical data, the known load data can be used as the label of the data sample, and the prediction process is optimized based on the error between the label and the predicted load data of the period to be predicted.
S303: and training the time convolution network model through a training set to obtain an initial load prediction network model.
Specifically, an initial time convolution network model is selected at the beginning of training, and a general time convolution network model comprises an input layer, a time convolution network layer and an output layer.
In particular toIn this embodiment, the input layer is used for inputting the preprocessed data, the time length is set as T, and the preprocessed data is expressed as
Wherein X represents dense feature data, S represents sparse feature data, and the dense feature data are data of dense features, such as continuous variable features of temperature, body-sensing temperature, humidity, historical load and the like; the sparse feature data is data of sparse features, such as discrete variable features of date type, weather, wind direction and the like. And carrying out cascade processing on the dense characteristic data and the sparse characteristic data to be used as final input data.
Referring to fig. 4, the time convolutional network layer includes a plurality of sub-blocks, the number of which can be set according to actual needs, in this embodiment, the number of the sub-blocks is set to 5, each sub-block is composed of two convolutional layers, and the ReLU is used as an activation function. Wherein the convolution operation of the convolution layer is one-dimensional convolution, the expansion coefficient is d, d increases exponentially with the deepening of the number of the sub-blocks,
and n is the number of sub-blocks. And in order to ensure that the length of the output sequence is the same as that of the input sequence, a cutting module is added after the convolution operation, and redundant filling elements are cut.
The output layer is composed of three layers of full-connection networks and is used for highly purifying data output by the time convolution network layer and then mapping the data into the output of the whole time convolution network model, namely the load value to be predicted through full connection.
Then, on the basis of the constructed time convolution network model, training is carried out through a training set to obtain an initial load prediction network model. During the training process of the time convolution network model, the optimizer can select Adam (adaptive moment estimation optimizer) to perform iterative optimization of the time convolution network model. Adam combines the advantages of the adaptive optimization algorithms AdaGrad and RMSProp, and calculates the updating step length by comprehensively considering the first moment estimation of the gradient, namely the mean value of the gradient, and the second moment estimation, namely the variance of the gradient which is not centralized.
Preferably, when the preprocessed data are sparse feature data, the data can be subjected to dimensionality reduction processing to a preset dimensionality through an artificial neural network. Wherein, the artificial neural network comprises an input layer, an output layer and two hidden layers, each layer uses a ReLU (Rectified Linear Unit) as an activation function, adds nonlinear mapping, inputs a vector S into the artificial neural network, and finally outputs a vector C with the length of m,
wherein m is a preset dimension.
Preferably, random inactivation treatment is added, and when dimension reduction treatment is carried out on the preprocessed data through the artificial neural network, random inactivation treatment is carried out on each hidden layer of the artificial neural network. The random inactivation is a method for optimizing the artificial neural network with the deep structure, and the weights of part of nodes of each hidden layer are randomly inactivated by randomly zeroing the partial weights or outputs of the hidden layers in the learning process, so that the interdependency among the nodes is reduced, the regularization of the neural network is realized, the structural risk of the neural network is reduced, and the overfitting is prevented.
Similarly, when the time convolution network model is trained through the training set, all convolution layers of the time convolution network model are subjected to random inactivation treatment, so that over-training fitting is avoided.
Preferably, in the training process of the time convolution network layer, the distribution of the output data of each convolution layer changes gradually, and the difference from the distribution rule of the input data is larger and larger, so that a batch standardization processing layer is added before the input of each convolution layer to perform batch standardization processing on the input data, and the batch standardization processing is also called batch normalization, and is a technology for improving the performance and stability of the neural network, and provides zero-mean/unit variance input data for any layer in the neural network. And further, the expectation and variance of the input data of each convolution layer are fixed, the distribution of the data is stabilized, the aim of accelerating the convergence speed of the time convolution network is fulfilled, and the phenomena of overfitting and gradient explosion are relieved to a certain extent.
In this embodiment, in the training process of the time convolution network model, the initial learning rate of Adam is set to 0.0001, the learning rate attenuation factor is 0.7, when the learning rate is high, the oscillation occurs in the training process, and the oscillation is reduced by setting the attenuation factor. The training iteration times are set to be 100 times, 64 samples are selected each time, in order to avoid overfitting, a regular function is added to a mean square error loss function of a time convolution network model, a time length is set to be 72 in a data input layer, namely load data, meteorological data and date data 72 hours before a prediction moment are input, 52-dimensional sparse features are subjected to dimension reduction, a dimension reduction part is a four-layer artificial neural network, the number of input channels is 52, the number of channels of a hidden layer is 32 and 16 respectively, and the number of output channels is 8. The time convolution network layer part comprises five sub-blocks, the sizes of convolution kernels are all 2, in convolution operation, the step length is 1, the filling coefficient is equal to the expansion coefficient, the random deactivation rate of each convolution layer is 0.2, the number of output channels of each sub-block is 64, 128, 256 and 512, the expansion coefficients are 1, 2, 4, 8 and 16, three layers of full-connection networks are used in the output layers, and the number of the output channels is set to be 128, 64 and 1.
S304: and testing the initial load prediction network model through the test set to obtain a test result, evaluating the initial load prediction network model according to the test result, and taking the initial load prediction network model meeting the preset evaluation index as the load prediction network model.
Specifically, after the data in the test set is processed in the same way as the data in the training set, the data is predicted through an initial load prediction network model to obtain a test result, namely predicted load data of a time period to be predicted, actual load data of the time period to be predicted is found in historical data, the predicted load data is compared with the actual load data, and an average absolute percentage error is obtained through the following formula
And mean square error
：
Wherein the content of the first and second substances,
in order to be the actual load data,
in order to predict the load data,
the number of points, namely the time in the test set.
And evaluating the initial load prediction network model by taking the average absolute percentage error and the mean square error as evaluation indexes, meeting the preset evaluation indexes when the average absolute percentage error and the mean square error are both smaller than the preset average absolute percentage error value and the preset mean square error value, and taking the initial load prediction network model meeting the preset evaluation indexes as the load prediction network model, otherwise, continuing training the initial load prediction network model.
The load forecasting network model is obtained through the construction mode, and then forecasting is carried out through the load forecasting network model according to the load influence data of the power system in the period to be forecasted, and the load data and the load influence data of the power system in the preset period before the period to be forecasted, so that the load forecasting result of the power system in the period to be forecasted is obtained. Specifically, due to the continuity of the load of the power system in time, the load data and the load influence data in a preset time period before the time period to be predicted are used as an influence factor of the load data in the time period to be predicted, and then the load influence data in the time period to be predicted are combined and used for load data prediction in the time period to be predicted together. The load influence data of the time period to be predicted, and the load data and the load influence data of the power system in the preset time period before the time period to be predicted are preprocessed according to the preprocessing mode described in the training set part in the step S302, and then are combined to obtain a data sample, the data sample is used as the input of the load prediction network model, the input is performed on the load prediction network model, and finally, the load prediction result of the time period to be predicted of the power system is obtained.
In summary, according to the short-term load prediction method of the power system, load data and load influence data of a preset time period before a time period to be predicted of the power system and load influence data of the time period to be predicted of the power system are obtained, prediction is performed through a preset load prediction network model according to the load influence data of the time period to be predicted of the power system and the load data and load influence data of the preset time period before the time period to be predicted of the power system, wherein the preset load prediction network model is obtained by training a time convolution network model through historical data of the power system, and based on parallel execution characteristics of the time convolution network model, the load prediction network model can perform parallel data processing, so that the load prediction speed of the power system is greatly improved. Meanwhile, as the existing LSTM model has a forgetting gate, some historical information can be gradually omitted by the forgetting gate in the prediction process; the convolution layer of the time convolution network model is in a causal relationship with the convolution layer, so that more historical information can be reserved by adopting the time convolution network model compared with the existing LSTM model, and the high accuracy requirement of the short-term load prediction of the power system is further ensured.
In another embodiment of the present invention, a method for predicting a short-term load of a power system is provided, which includes the following steps in addition to the whole contents of the method for predicting a short-term load of a power system in the embodiment shown in fig. 1: when the time convolution network model is trained through the training set, the convolution result output by the last convolution layer in the time convolution network model is weighted according to the attention mechanism and the time, the convolution result after the weighting processing is subjected to full-connection mapping, the mapping result is obtained, and the time convolution network model is optimized according to the mapping result. When the prediction is carried out through a preset load prediction network model, the convolution result output by the last convolution layer in the load prediction network model is weighted according to the attention mechanism and time, and the weighted convolution result is subjected to full-connection mapping to obtain the load prediction result of the power system in the period to be predicted.
Specifically, the retention of effective information distributed in different time periods in long-time sequence input is realized by introducing an Attention mechanism, so that the feature extraction capability of the time convolution network is enhanced. The Attention mechanism is derived from a human visual Attention mechanism, namely, limited Attention is focused on key information, and all Attention is focused to the key point, so that resources are saved, the most effective information is quickly obtained, and for the Attention mechanism, the Attention mechanism is a weight parameter distribution mechanism and aims to assist a model to capture important information.
In the embodiment, the importance degree of each historical data on the short-term load of the power system at each moment is manually analyzed, the weight parameter of each historical data at each moment is set, then based on the set weight parameter, the convolution result output by the last convolution layer in the time convolution network model is weighted during training, and during prediction, the convolution result output by the last convolution layer in the load prediction network model is weighted, so that effective information of each historical data in each period is greatly reserved, and the accuracy of the prediction result is improved.
The following are embodiments of the apparatus of the present invention that may be used to perform embodiments of the method of the present invention. For details of non-careless mistakes in the embodiment of the apparatus, please refer to the embodiment of the method of the present invention.
Referring to fig. 5, in a further embodiment of the present invention, a power system short-term load prediction system is provided, which can be used to implement the above power system short-term load prediction method.
The first data acquisition module is used for acquiring load data and load influence data of a preset time period before a time period to be predicted of the power system; the second data acquisition module is used for acquiring load influence data of the power system in a period to be predicted; the prediction module is used for predicting through a preset load prediction network model according to load influence data of a to-be-predicted time period of the power system and load data and load influence data of a preset time period before the to-be-predicted time period of the power system to obtain a load prediction result of the to-be-predicted time period of the power system; the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system.
In yet another embodiment of the present invention, a terminal device is provided that includes a processor and a memory for storing a computer program comprising program instructions, the processor being configured to execute the program instructions stored by the computer storage medium. The Processor may be a Central Processing Unit (CPU), or may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable gate array (FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., which is a computing core and a control core of the terminal, and is adapted to implement one or more instructions, and is specifically adapted to load and execute one or more instructions to implement a corresponding method flow or a corresponding function; the processor provided by the embodiment of the invention can be used for the operation of the short-term load forecasting method of the power system, and comprises the following steps: acquiring load data and load influence data of a preset time period before a time period to be predicted of a power system; acquiring load influence data of a power system in a period to be predicted; according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of the power system in the preset period before the period to be predicted, predicting through a preset load prediction network model to obtain a load prediction result of the power system in the period to be predicted; the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system.
In still another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a Memory device in a terminal device and is used for storing programs and data. It is understood that the computer readable storage medium herein may include a built-in storage medium in the terminal device, and may also include an extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also, one or more instructions, which may be one or more computer programs (including program code), are stored in the memory space and are adapted to be loaded and executed by the processor. It should be noted that the computer-readable storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory.
One or more instructions stored in the computer-readable storage medium may be loaded and executed by the processor to implement the corresponding steps of the method for predicting the short-term load of the power system in the above embodiments; one or more instructions in the computer-readable storage medium are loaded by the processor and perform the steps of: acquiring load data and load influence data of a preset time period before a time period to be predicted of a power system; acquiring load influence data of a power system in a period to be predicted; according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of the power system in the preset period before the period to be predicted, predicting through a preset load prediction network model to obtain a load prediction result of the power system in the period to be predicted; the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (11)
1. A method for predicting short-term load of a power system is characterized by comprising the following steps:
acquiring load data and load influence data of a preset time period before a time period to be predicted of a power system; wherein the load influence data comprises meteorological data and holiday data;
acquiring load influence data of a power system in a period to be predicted;
according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of the power system in the preset period before the period to be predicted, predicting through a preset load prediction network model to obtain a load prediction result of the power system in the period to be predicted;
when the prediction is carried out through a preset load prediction network model, carrying out weighting processing on a convolution result output by the last convolution layer in the load prediction network model according to an attention mechanism and time, and carrying out full-connection mapping on the weighted convolution result to obtain a load prediction result of the power system in a period to be predicted;
the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system; the specific method comprises the following steps:
acquiring historical data of the power system, wherein the historical data comprises historical load data and historical load influence data;
dividing historical data of the power system into a training set and a test set;
acquiring the data type of each data in the training set; wherein the data type comprises dense feature data and sparse feature data;
when the data type is sparse characteristic data, performing dimensionality reduction on the data to a preset dimensionality through an artificial neural network;
training a time convolution network model through a training set to obtain an initial load prediction network model;
when the time convolution network model is trained through the training set, the convolution result output by the last convolution layer in the time convolution network model is weighted according to the attention mechanism and the time, the weighted convolution result is subjected to full-connection mapping to obtain a mapping result, and the time convolution network model is optimized according to the mapping result;
and testing the initial load prediction network model through the test set to obtain a test result, evaluating the initial load prediction network model according to the test result, and taking the initial load prediction network model meeting the preset evaluation index as the load prediction network model.
2. The method for predicting the short-term load of the power system as claimed in claim 1, wherein the specific method for dividing the historical data of the power system into the training set and the test set is as follows:
preprocessing historical data of the power system to obtain preprocessed data;
and dividing the preprocessed data into a training set and a test set according to a preset proportion.
3. The method for predicting the short-term load of the power system as claimed in claim 2, wherein the method for preprocessing the historical data of the power system to obtain the preprocessed data comprises:
acquiring the type of each datum in the historical data of the power system;
when the type of the data is a continuous variable, carrying out normalization processing on the data to obtain preprocessed data;
and when the type of the data is a discrete variable, encoding the data, and mapping the data into a binary vector to obtain preprocessed data.
4. The method for predicting the short-term load of the power system according to claim 1, wherein when the dimension reduction processing is performed on the preprocessed data through the artificial neural network, each hidden layer of the artificial neural network is subjected to random inactivation processing.
5. The method for predicting the short-term load of the power system according to claim 1, wherein the specific method for evaluating the initial load prediction network model according to the test result comprises the following steps:
and obtaining the average absolute percentage error and the mean square error of the test result according to the historical data and the test result of the power system, wherein when the average absolute percentage error and the mean square error are both smaller than the preset average absolute percentage error and the preset mean square error, the initial load prediction network model meets the preset evaluation index.
6. The method according to claim 1, wherein each convolutional layer of the time convolutional network model is randomly deactivated when the time convolutional network model is trained by the training set.
7. The method according to claim 1, wherein when the time convolutional network model is trained by the training set, a convolution result output by a current convolutional layer in the time convolutional network model is input to a next convolutional layer after being subjected to batch normalization processing.
8. The method for predicting the short-term load of the power system according to claim 1, wherein when the prediction is performed through a preset load prediction network model, a convolution result output by a current convolution layer in the load prediction network model is input into a next convolution layer after batch standardization processing.
9. A power system short term load prediction system, comprising:
the first data acquisition module is used for acquiring load data and load influence data of a preset time period before a time period to be predicted of the power system; wherein the load influence data comprises meteorological data and holiday data;
the second data acquisition module is used for acquiring load influence data of the power system in a period to be predicted;
the prediction module is used for predicting through a preset load prediction network model according to the load influence data of the power system in the period to be predicted, and the load data and the load influence data of the power system in the preset period before the period to be predicted to obtain a load prediction result of the power system in the period to be predicted; when the prediction is carried out through a preset load prediction network model, weighting processing is carried out on a convolution result output by the last convolution layer in the load prediction network model according to an attention mechanism, and full-connection mapping is carried out on the convolution result after weighting processing to obtain a load prediction result of a power system in a period to be predicted;
the preset load prediction network model is obtained by training a time convolution network model by using historical data of the power system; the specific method comprises the following steps:
acquiring historical data of the power system, wherein the historical data comprises historical load data and historical load influence data;
dividing historical data of the power system into a training set and a test set;
acquiring the data type of each data in the training set; wherein the data type comprises dense feature data and sparse feature data;
when the data type is sparse characteristic data, performing dimensionality reduction on the data to a preset dimensionality through an artificial neural network;
training a time convolution network model through a training set to obtain an initial load prediction network model;
when the time convolution network model is trained through the training set, the convolution result output by the last convolution layer in the time convolution network model is weighted according to the attention mechanism and the time, the weighted convolution result is subjected to full-connection mapping to obtain a mapping result, and the time convolution network model is optimized according to the mapping result;
and testing the initial load prediction network model through the test set to obtain a test result, evaluating the initial load prediction network model according to the test result, and taking the initial load prediction network model meeting the preset evaluation index as the load prediction network model.
10. A computer arrangement comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, carries out the steps of the power system short term load prediction method according to any one of claims 1 to 8.
11. A computer-readable storage medium, in which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for predicting a short-term load of an electric power system according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011351479.5A CN112215442B (en) | 2020-11-27 | 2020-11-27 | Method, system, device and medium for predicting short-term load of power system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011351479.5A CN112215442B (en) | 2020-11-27 | 2020-11-27 | Method, system, device and medium for predicting short-term load of power system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112215442A CN112215442A (en) | 2021-01-12 |
| CN112215442B true CN112215442B (en) | 2021-03-12 |
Family
ID=74068079
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011351479.5A Active CN112215442B (en) | 2020-11-27 | 2020-11-27 | Method, system, device and medium for predicting short-term load of power system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112215442B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113255973A (en) * | 2021-05-10 | 2021-08-13 | 曙光信息产业(北京)有限公司 | Power load prediction method, power load prediction device, computer equipment and storage medium |
| CN113743650B (en) * | 2021-08-04 | 2022-12-06 | 南方电网科学研究院有限责任公司 | Power load prediction method, device, equipment and storage medium |
| CN113822467A (en) * | 2021-08-24 | 2021-12-21 | 华南理工大学 | Graph neural network prediction method for electric power area load |
| CN113743674A (en) * | 2021-09-10 | 2021-12-03 | 中国电力科学研究院有限公司 | Energy storage output prediction method, system, equipment and medium based on deep learning |
| CN114374953B (en) * | 2022-01-06 | 2023-09-05 | 西安交通大学 | APP use prediction method and system under multi-source feature conversion base station based on Hadoop and RAPIS |
| CN116526479B (en) * | 2023-07-03 | 2023-10-13 | 国网北京市电力公司 | Method, device, equipment and medium for predicting power supply quantity |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210671A (en) * | 2019-06-04 | 2019-09-06 | 国家电网有限公司 | Power-system short-term load forecasting method, device and terminal device |
| CN110472779A (en) * | 2019-07-30 | 2019-11-19 | 东莞理工学院 | A kind of power-system short-term load forecasting method based on time convolutional network |
| CN111382906A (en) * | 2020-03-06 | 2020-07-07 | 南京工程学院 | Power load prediction method, system, equipment and computer readable storage medium |
| CN111507521A (en) * | 2020-04-15 | 2020-08-07 | 北京智芯微电子科技有限公司 | Method and device for predicting power load of transformer area |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9020874B2 (en) * | 2011-10-31 | 2015-04-28 | Siemens Aktiengesellschaft | Short-term load forecast using support vector regression and feature learning |
| US9118182B2 (en) * | 2012-01-04 | 2015-08-25 | General Electric Company | Power curve correlation system |
-
2020
- 2020-11-27 CN CN202011351479.5A patent/CN112215442B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210671A (en) * | 2019-06-04 | 2019-09-06 | 国家电网有限公司 | Power-system short-term load forecasting method, device and terminal device |
| CN110472779A (en) * | 2019-07-30 | 2019-11-19 | 东莞理工学院 | A kind of power-system short-term load forecasting method based on time convolutional network |
| CN111382906A (en) * | 2020-03-06 | 2020-07-07 | 南京工程学院 | Power load prediction method, system, equipment and computer readable storage medium |
| CN111507521A (en) * | 2020-04-15 | 2020-08-07 | 北京智芯微电子科技有限公司 | Method and device for predicting power load of transformer area |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112215442A (en) | 2021-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112215442B (en) | Method, system, device and medium for predicting short-term load of power system | |
| Tan et al. | Ultra-short-term industrial power demand forecasting using LSTM based hybrid ensemble learning | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| CN112116144B (en) | Regional power distribution network short-term load prediction method | |
| CN112734128B (en) | 7-day power load peak prediction method based on optimized RBF | |
| CN111680841B (en) | Short-term load prediction method, system and terminal equipment based on principal component analysis | |
| CN113743673B (en) | Power load prediction method during typhoon | |
| CN113657668A (en) | Power load prediction method and system based on LSTM network | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN116596044B (en) | Power generation load prediction model training method and device based on multi-source data | |
| CN114330934A (en) | Model parameter self-adaptive GRU new energy short-term power generation power prediction method | |
| CN111680712A (en) | Transformer oil temperature prediction method, device and system based on similar moments in the day | |
| CN116599050A (en) | Photovoltaic prediction method and related device based on self-attention mechanism | |
| CN114611757A (en) | Electric power system short-term load prediction method based on genetic algorithm and improved depth residual error network | |
| CN111697560B (en) | Method and system for predicting load of power system based on LSTM | |
| CN113570106A (en) | Power load prediction method and device based on expressway neural network | |
| Fatema et al. | Analysing and forecasting electricity demand and price using deep learning model during the COVID-19 pandemic | |
| CN115907131A (en) | Method and system for building electric heating load prediction model in northern area | |
| CN115271242A (en) | Training method, prediction method and device of photovoltaic power generation power prediction model | |
| CN114139788A (en) | Method and device for predicting power of offshore wind turbine generator | |
| CN113723670A (en) | Photovoltaic power generation power short-term prediction method with variable time window | |
| CN113780644A (en) | Photovoltaic output prediction method based on online learning | |
| CN113254857A (en) | SSA-ELM-based short-term power load prediction method | |
| CN110942195A (en) | Power load prediction method and device | |
| Lee et al. | Application of temperature prediction model based on LSTNet in telecommunication Room |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |