CN115983448B - Multi-energy load prediction method based on space-time diagram neural network - Google Patents
Multi-energy load prediction method based on space-time diagram neural network Download PDFInfo
- Publication number
- CN115983448B CN115983448B CN202211609240.2A CN202211609240A CN115983448B CN 115983448 B CN115983448 B CN 115983448B CN 202211609240 A CN202211609240 A CN 202211609240A CN 115983448 B CN115983448 B CN 115983448B
- Authority
- CN
- China
- Prior art keywords
- data
- space
- neural network
- load
- load prediction
- 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
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a multi-energy load prediction method based on a space-time diagram neural network, which comprises the following steps: selecting calendar features, and acquiring electric, thermal, cold load and weather feature data of corresponding time in the comprehensive energy system; performing abnormal data detection, linear interpolation and normalization processing on each characteristic data to obtain a time sequence data set; selecting weather and calendar features with strong correlation with electric, thermal and cold loads by using a gray correlation analysis method to obtain a multi-energy load prediction data set; dividing the multi-energy load prediction data set into a training set, a verification set and a test set, and training the space-time diagram neural network by utilizing the data in the training set and the verification set to obtain an optimal load prediction model; and inputting the data in the test set into an optimal load prediction model to obtain electric, thermal and cold load prediction results. The invention utilizes the space-time diagram neural network to ensure that the multi-energy load prediction model has stronger universality and higher prediction performance.
Description
Technical Field
The invention relates to the technical field of comprehensive energy systems, in particular to a multi-energy load prediction method based on a space-time diagram neural network.
Background
The comprehensive energy system plays an important role in the process of realizing the double-carbon target in China. The comprehensive energy system is an important technical means for improving the energy utilization rate and reducing the energy consumption cost, and along with the continuous development of the comprehensive energy utilization technology and the diversification of the energy consumption demands of users, the coupling characteristics among multiple loads are difficult to reflect by the existing single load prediction method, and the accurate multiple load prediction becomes a primary premise of optimizing the dispatching and economic operation of the comprehensive energy system.
The traditional load prediction method generally predicts different types of loads independently, and does not consider that different loads have complex coupling relations, so that the prediction precision of the multi-energy load is higher than that of a single load. The classical methods of short-term load prediction comprise a regression analysis method, a time sequence method and an exponential smoothing method, and the methods do not well reflect influence factors such as weather, holidays and the like.
In the aspect of comprehensive energy system load prediction, the following prediction methods exist: the method comprises an electric, cold and heat load prediction method in a multi-energy system based on vector autoregressive, a multi-energy load prediction model formed by CNN, GRU, GBRT, and a short-term energy prediction method for electric power, heat and fuel gas constructed by a Deep Belief Network (DBN) and a multi-task regression layer; recurrent Neural Networks (RNNs), LSTM or GRU have their own limitations. First, RNNs have problems of gradient extinction and gradient explosion when processing time-series data, and thus cannot be relied on for a long period of time. As LSTM adds more return gradient paths, as long as one path has no gradient vanishing, the gradient vanishing problem is improved, but the remote dependency capability of the model is improved, and the problem still exists. GRU has also been proposed to solve the gradient problem in long-term memory and back propagation, but still not solve the long-term memory and gradient problem, and the GRU has one less gating unit inside compared with LSTM, and the parameter amount is reduced. Most importantly, the inherent temporal characteristics of RNNs prevent parallelization of training. On the other hand, the implementation of the multitasking architecture is now essentially a simple "hard-wired" approach. The feature sharing layer and the specific task layer are directly connected, which cannot reflect different concerns of different subtasks about sharing features. The limitations cause that the existing energy prediction model has low prediction precision.
Disclosure of Invention
The invention aims at: the invention provides a multi-energy load prediction method based on a space-time diagram neural network, which fully captures the dependency relationship between electric, cold and heat loads and weather features and calendar features, effectively utilizes rich information between the electric, cold and heat loads and the weather features and calendar features, and improves the prediction performance of a multi-energy load prediction model.
In order to achieve the above purpose, the present invention provides the following technical solutions: a multi-energy load prediction method based on a space-time diagram neural network comprises the following steps:
s1: calendar characteristic data are firstly selected from the existing comprehensive energy system database, then electric, thermal and cold load historical data corresponding to time are obtained, and then weather characteristic data of the place where the comprehensive energy system is located are obtained;
s2: respectively carrying out data preprocessing on the acquired electric, thermal and cold load historical data and each meteorological characteristic data to obtain a time sequence data set;
s3: using a gray correlation analysis method, taking the time sequence data obtained in the step S2 as input to obtain correlation between electric, thermal and cold loads and correlation between each load and each weather feature and each calendar feature, selecting weather features and calendar features with correlation not smaller than the selected threshold according to the gray correlation analysis result and a preset correlation selection threshold, and combining the time sequence data corresponding to the selected weather features and calendar features with the time sequence data of each load feature to form a multi-energy load prediction data set;
s4: dividing data in the multi-energy load prediction data set into a training set, a verification set and a test set according to a preset proportion, and building a space-time diagram neural network load prediction model; training the space-time diagram neural network load prediction model by using the data in the training set and the verification set, and optimizing parameters and super parameters of the model in the process of training the space-time diagram neural network load prediction model to obtain an optimal space-time diagram neural network load prediction model;
s5: inputting the data in the test set into an optimal space-time diagram neural network prediction model to obtain normalized prediction results of electric, thermal and cold loads, and performing inverse normalization processing on the normalized prediction results of the electric, thermal and cold loads to obtain final prediction values of the electric, thermal and cold loads.
Further, the step S2 includes the following steps:
s201: according to the international measurement unit conversion method, the units of the acquired electric, thermal and cold load data are uniformly converted into KW serving as units; the conversion formula is as follows:
1KW＝3.4mBtu/h＝0.284ton-hr
wherein KW is electric load unit, mBtu/h is cold load unit, ton-hr is heat load unit;
s202: for unified electric, heat and cold load data and meteorological characteristic data, detecting time sequence abnormal values by using a box diagram quartile detection method and a first-order exponential smoothing method respectively; if the time sequence abnormal value is detected, setting the time sequence abnormal value as Nan, and filling Nan by adopting a linear interpolation method to obtain the sample data of electricity, heat, cold load and each meteorological characteristic;
s203: respectively carrying out normalization processing on the electric, thermal and cold load and sample data of each meteorological feature to obtain a time sequence data set of electric, thermal and cold load data, meteorological feature data and calendar feature data; the normalization formula is as follows:
wherein X is original data, X min Is the minimum value of the feature, X max The Xnom normalized data is the maximum value of the feature.
Further, the gray correlation analysis formula in S3 is as follows:
in the formula: alpha i As the correlation coefficient beta i For the degree of association, x 0 (k) The weather characteristic sequence is normalized; x is x i (k) The normalized load characteristic sequence; ρ is the resolution factor, taken to be 0.5.
Further, the space-time diagram neural network load prediction model in the step S4 includes a diagram learning module, a feature attention module, a diagram convolution module and a time sequence convolution module; the data in the training set is input into a drawing learning module for learning and training, and the drawing learning module extracts dynamic spatial relations between all features and loads from the training set to obtain an adjacent matrix; the time sequence convolution module extracts high-level time characteristics; the feature attention module calculates the weight of each feature and the mutual influence between the loads, fully excavates the correlation between each feature and the electric, thermal and cold loads respectively, and automatically selects weather and calendar features with strong correlation with the final predicted electric, thermal and cold loads; the time sequence convolution module, the characteristic attention module and the graph convolution module are used in a staggered mode to capture the dependency relationship among time, characteristics and space respectively; and residual connection is added between the input end of the time sequence convolution module and the output end of the graph convolution module, so that the problem of gradient disappearance is avoided.
Further, the training of the space-time diagram neural network load prediction model includes:
(1) Determining super parameters of a space-time diagram neural network load prediction model by adopting a grid search method, wherein the super parameters comprise the number of convolution channels, the learning rate, the layer number of a graph convolution module and the layer number of a time sequence convolution module;
(2) Inputting data in the training set into a space-time diagram neural network load prediction model, outputting predicted values of electric, thermal and cold loads, and performing learning training on the space-time diagram neural network load prediction model;
(3) And inputting the data in the verification set into a space-time diagram neural network load prediction model, and evaluating the prediction capability of the space-time diagram neural network load prediction model while adjusting the super parameters of the model.
Further, the inverse normalization processing formula in S5 is as follows:
X m ＝X n *(X max -X min )+X min
wherein Xn is multi-energy load prediction data based on space-time diagram neural network, X min Is the minimum value of the feature, X max And Xm is the predicted data after inverse normalization and is the maximum value of the characteristics.
Compared with the prior art, the multi-energy load prediction method based on the space-time diagram neural network has the following technical effects:
(1) In the invention, for electric, cold and heat load data and meteorological data in a comprehensive energy system, a box diagram quartile detection algorithm and a first-order exponential smoothing algorithm are used for detecting abnormal values of a time sequence, and a linear interpolation method is adopted for filling Nan, so that the quality and relative integrity of the data are effectively ensured;
(2) In the invention, gray correlation degree is used for analyzing the correlation degree between electric, thermal and cold loads and the correlation degree between each load and each weather feature and calendar feature respectively, and weather and calendar features with strong correlation with electric, thermal and cold loads are selected, so that the prediction precision can be effectively improved, and meanwhile, the training and reasoning time is reduced;
(3) The multi-energy load prediction method provided by the invention not only considers the internal coupling relation of the multi-energy load to perform joint prediction, but also captures the dependency relation among each load, weather and calendar characteristics in a time sequence by using the space-time diagram neural network, thereby improving the accuracy of multi-energy load prediction, leading the universality of the multi-energy load prediction model to be stronger and leading the prediction performance to be higher.
Drawings
FIG. 1 is a flow chart of a method of multi-energy load prediction based on a space-time neural network;
FIG. 2 is a flow chart of data preprocessing in a space-time diagram neural network-based multi-energy load prediction method;
FIG. 3 is a graph of gray correlation analysis results in a multi-energy load prediction method based on a space-time neural network;
FIG. 4 is a diagram of a space-time diagram neural network structure of a multi-energy load prediction method based on the space-time diagram neural network;
FIG. 5 is a block diagram of a graph convolution module in a space-time graph neural network;
FIG. 6 is a block diagram of a time sequential convolution module in a space-time diagram neural network;
FIG. 7 is a graph comparing electrical load prediction results of a multi-energy load prediction method based on a space-time neural network;
FIG. 8 is a graph comparing the results of the multi-energy load prediction method based on the space-time neural network;
fig. 9 is a comparison graph of thermal load prediction results of a multi-energy load prediction method based on a space-time neural network.
Detailed Description
For a better understanding of the technical content of the present invention, specific examples are set forth below, along with the accompanying drawings.
Aspects of the invention are described herein with reference to the accompanying drawings, which illustrate various exemplary embodiments. The embodiments of the present invention are not limited to the embodiments described in the drawings. It is to be understood that this invention is capable of being carried out by any of the various concepts and embodiments described above and as such described in detail below, since the disclosed concepts and embodiments are not limited to any implementation. Additionally, some aspects of the disclosure may be used alone or in any suitable combination with other aspects of the disclosure.
The invention provides a multi-energy load prediction method based on a space-time diagram neural network, which fully captures the dependency relationship among electricity, cold and heat loads, meteorological features and holiday features in a time sequence, effectively utilizes rich information among time and features and improves the prediction performance of a multi-energy load prediction model. As shown in fig. 1, the multi-energy load prediction method based on the space-time diagram neural network comprises the following steps:
the first step: calendar characteristic data are selected from the existing comprehensive energy system database, then electric, thermal and cold load historical data corresponding to time are obtained, and then weather characteristic data of the place where the comprehensive energy system is located are obtained.
In this embodiment, calendar features select holidays, weekends, months, days, hours; the load data is from the Tanpei school district of the state university of Arizona, america, and the cold, heat and electric load data from 2017, 1 month, 1 day, 0 time, 2019, 12 months, 31 days, 24 time are acquired from the network platform, and the time resolution is 1h. Weather data sources are weather data including dew point, wind speed, precipitation, wind direction, humidity, temperature, barometric pressure at the location of the sample school district in the website of the national weather data center.
And a second step of: and respectively carrying out data preprocessing on the acquired electric, thermal and cold load historical data and each meteorological characteristic data to obtain a time sequence data set.
In this example, as shown in FIG. 2, the electric load unit is KW, the cold load unit is mBtu/h, and the heat load unit is ton-hr in the obtained history load. Firstly, the electric, cold and heat loads are taken as a unified measurement unit by KW. Wherein, the conversion formula is as follows:
1KW＝3.4mBtu/h＝0.284ton-hr
and detecting abnormal values of the time sequence by using a box diagram four-bit detection algorithm and a first-order exponential smoothing algorithm for historical load data of a unified measurement unit and meteorological data of dew point, wind speed, precipitation, wind direction, humidity, temperature and atmospheric pressure. If the value is abnormal, the value is set as Nan, and the Nan is filled by adopting a linear interpolation method so as to ensure the quality and the relative integrity of the data. And finally, carrying out normalization processing on the electric, cold and heat load data and each meteorological characteristic data to obtain a time sequence data set of the electric, cold and heat load data, the meteorological data and the calendar data. The normalization formula is as follows:
wherein X is original data, X min Is the minimum value of the feature, X max The Xnom normalized data is the maximum value of the feature.
And a third step of: and (3) using a gray correlation analysis method, taking time sequence data as input to obtain correlation degrees between electric, thermal and cold loads and between each load and each weather feature and each calendar feature, selecting a threshold according to a gray correlation analysis result and a preset correlation degree, selecting weather features and calendar features with the correlation degree of each load not smaller than the selection threshold, and combining the time sequence data corresponding to the selected weather features and calendar features with the time sequence data of each load feature to form a multi-energy load prediction data set. The gray correlation analysis formula is as follows:
in the formula: alpha i As the correlation coefficient beta i For the degree of association, x 0 (k) The weather characteristic sequence is normalized; x is x i (k) The normalized load characteristic sequence; ρ is the resolution factor, taken to be 0.5.
In this embodiment, the load prediction features of the integrated energy system include 15 features in total, including electric load, cold load, heat load, dew point, wind speed, precipitation, wind direction, humidity, temperature, atmospheric pressure, holidays, weekends, months, days, and hours, the gray correlation analysis result is shown in fig. 3, 0-0.6 indicates weak correlation, 0.6-0.8 indicates general correlation, 0.8-1 indicates strong correlation, the load feature correlation selection threshold is set to 0.6, and the finally selected correlation features are 8 features in total.
Fourth step: dividing data in the multi-energy load prediction data set into a training set, a verification set and a test set according to a preset proportion, and building a space-time diagram neural network load prediction model; training the space-time diagram neural network load prediction model by using the data in the training set and the verification set, and optimizing parameters and super parameters of the model in the process of training the space-time diagram neural network load prediction model to obtain an optimal space-time diagram neural network load prediction model;
in this embodiment, the multi-energy load prediction dataset is calculated according to 6:2:2 are divided into training sets, validation sets and test sets. The built space-time diagram neural network load prediction model comprises a diagram learning module, a characteristic attention module, a diagram convolution module and a time sequence convolution module; the connection sequence between the modules is shown in fig. 4, and the method specifically comprises the steps of inputting training data to perform space-time diagram neural network load prediction model learning training and adjusting optimization model parameters:
(1) Determining the super parameters of a space-time diagram neural network load prediction model by adopting a grid search method;
according to the embodiment of the invention, python codes are adopted for writing, a space-time diagram neural network is built based on a pytorch framework, meanwhile, a SVR, RF, DBN, LSTM model is selected as a comparison model, and detailed parameters are shown in table 1:
TABLE 1
(2) Inputting data in training set into a drawing learning module for learning training, wherein the drawing learning module aims at learning fromDynamic spatial relationships between variables are extracted in the training set. The change of the numerical values of different variables at the crossing time can better reflect the spatial relationship among the variables, and considering the periodicity of the time sequence, we set a super-parameter period P to divide the training MTS intoSegments, after obtaining time series segments, we join the segments to obtain a four-dimensional tensorSubsequently, we use two-dimensional convolution and two fully connected layers to transform the four-dimensional tensor. The number of input channels is S and the number of output channels is the number of graphs we want to get. These figures constitute an atlas a.
The feature attention module adaptively captures dynamic associations between nodes in a feature dimension.
First, we first convert X into two feature spaces Q, K. The formula is as follows:
Q(X)＝(X·W Q1 )·W Q2
K(X)＝X·W K
then, a weight value of the feature attention is calculated, and the formula is as follows:
S＝Q(X)·K(X) T +b s
finally, the normalized attention alpha i，j Multiplying by X to calculate the output of attention while multiplying a by i，j Multiplying by the adjacency matrix further increases the weight of the important feature.
X′＝α i，j ·X
adj′＝α i，j ·αdj
Wherein X is the input of the attention module, W Q1 ∈R T ，W Q2 ER C×T ，W K ∈R C ，b s ∈R b×N×N Adj is the adjacency matrix.
The graph convolution module fuses the information of one node with the information of its neighbors to handle the spatial dependency of the features. As shown in fig. 5, the graph convolution module consists of two GCN modules that process the ingress and egress information through each node separately.
Where K is the propagation depth, H (k-1) Input hidden state representing previous layer output, H K The output hiding state of the current layer is represented, A is an adjacent matrix, and W is a convolution change parameter, so that the optimization can be trained.
The temporal convolution module expands the convolution filter to extract high-level temporal features. As shown in fig. 6, the module consists of two inflated convolution layers. An expansion convolution layer is followed by a tangential hyperbolic activation function and operates as a filter. The other layer is followed by a sigmoid activation function as a gate that controls the amount of information that the filter can pass to the next module.
(3) And inputting the data in the verification set into a space-time diagram neural network load prediction model, and evaluating the prediction capability of the space-time diagram neural network load prediction model while adjusting the super parameters of the model.
Fifth step: inputting the data in the test set into an optimal space-time diagram neural network prediction model to obtain normalized prediction results of electric, thermal and cold loads, and performing inverse normalization processing on the normalized prediction results of the electric, thermal and cold loads to obtain final prediction values of the electric, thermal and cold loads. Wherein, the inverse normalization processing formula is as follows:
X m ＝X n *(X max -X min )+X min
wherein Xn is multi-energy load prediction data based on space-time diagram neural network, X min Is the minimum value of the feature, X max And Xm is the predicted data after inverse normalization and is the maximum value of the characteristics.
In order to compare the load prediction effect, a prediction result evaluation index is established, and three indexes of an average absolute error MAE, a root mean square error RMSE and an average absolute percentage error MAPE are adopted to measure the deviation between a prediction value and an actual value. The specific error formula is shown as follows:
in the formula, the number of samples to be evaluated,representing predicted value, y i Representing the true value.
The MAE, RMSE and MAPE indices for each model on the test set were calculated as shown in tables 2, 3, 4, and we selected one month prediction curves in the test dataset for more visual display of the prediction results as shown in FIGS. 7, 8, 9. The result shows that the multi-energy load prediction method based on the space-time diagram neural network has high prediction precision, and the prediction precision is higher than that of a comparison model.
TABLE 2
TABLE 3 Table 3
TABLE 4 Table 4
While the invention has been described in terms of preferred embodiments, it is not intended to be limiting. Those skilled in the art will appreciate that various modifications and adaptations can be made without departing from the spirit and scope of the present invention. Accordingly, the scope of the invention is defined by the appended claims.
Claims (3)
1. A multi-energy load prediction method based on a space-time diagram neural network is used for capturing the dependency relationship between electric, thermal and cold loads in a comprehensive energy system and each meteorological characteristic and calendar characteristic respectively through the space-time diagram neural network to realize the optimization of the prediction performance of a multi-energy load prediction model, and is characterized by comprising the following steps:
s1: calendar characteristic data are firstly selected from the existing comprehensive energy system database, then electric, thermal and cold load historical data corresponding to time are obtained, and then weather characteristic data of the place where the comprehensive energy system is located are obtained;
s2: respectively carrying out data preprocessing on the acquired electric, thermal and cold load historical data and each meteorological characteristic data to obtain a time sequence data set;
s3: using a gray correlation analysis method, taking the time sequence data obtained in the step S2 as input to obtain correlation between electric, thermal and cold loads and correlation between each load and each weather feature and each calendar feature, selecting weather features and calendar features with correlation degree not smaller than the correlation degree selection threshold according to the gray correlation analysis result and a preset correlation degree selection threshold, and combining the time sequence data corresponding to the selected weather features and calendar features with the time sequence data of each load feature to form a multi-energy load prediction data set;
s4: dividing data in the multi-energy load prediction data set into a training set, a verification set and a test set according to a preset proportion, and building a space-time diagram neural network load prediction model; training the space-time diagram neural network load prediction model by using the data in the training set and the verification set, and optimizing parameters and super parameters of the model in the process of training the space-time diagram neural network load prediction model to obtain an optimal space-time diagram neural network load prediction model;
s5: inputting the data in the test set into an optimal space-time diagram neural network prediction model to obtain normalized prediction results of electric, thermal and cold loads, and performing inverse normalization processing on the normalized prediction results of the electric, thermal and cold loads to obtain final prediction values of the electric, thermal and cold loads;
the space-time diagram neural network load prediction model in the S4 comprises a diagram learning module, a characteristic attention module, a diagram convolution module and a time sequence convolution module; the data in the training set is input into a drawing learning module for learning and training, and the drawing learning module extracts dynamic space-time relations between each meteorological feature and calendar feature and load from the training set to obtain an adjacency matrix; the time sequence convolution module extracts high-level time characteristics; the characteristic attention module calculates the weight of the mutual influence between each weather characteristic and calendar characteristic and the load, fully excavates the correlation between the weather characteristic and the calendar characteristic and the electric, thermal and cold loads respectively, and automatically selects weather and calendar characteristics with strong correlation with the final predicted electric, thermal and cold loads; the time sequence convolution module, the characteristic attention module and the graph convolution module are used in a staggered mode to capture the time and space dependency relationship respectively; residual connection is added between the input end of the time sequence convolution module and the output end of the graph convolution module, so that the problem of gradient disappearance is avoided;
the training of the space-time diagram neural network load prediction model in the S4 comprises the following steps:
s4.1, determining super parameters of a space-time diagram neural network load prediction model by adopting a grid search method, wherein the super parameters comprise the number of convolution channels, the learning rate, the number of layers of a graph convolution module and the number of layers of a time sequence convolution module;
s4.2, inputting data in the training set into a space-time diagram neural network load prediction model, outputting predicted values of electric, thermal and cold loads, and performing learning training on the space-time diagram neural network load prediction model;
s4.3, inputting data in the verification set into a space-time diagram neural network load prediction model, and evaluating the prediction capability of the space-time diagram neural network load prediction model while adjusting the super parameters of the model;
in step S4, the data input diagram learning module in the training set learns and trains, specifically: a super-parametric period P is set to divide the training MTS intoSegments, after obtaining the time series segments, are connected to obtain a four-dimensional tensor +.>Converting the four-dimensional tensor using the two-dimensional convolution and the two fully connected layers, the number of input channels being S and the number of output channels being the number of graphs to be obtained, the graphs constituting an atlas A;
capturing, by a feature attention module, dynamic associations between nodes in a spatial dimension:
firstly, converting X into two characteristic spaces Q and K, wherein the formula is as follows:
Q(X)＝(X·W Q1 )·W Q2
K(X)＝X·W K
then, the weight of the feature attention is calculated as follows:
G＝Q(X)·K(X) T +b s
finally, feature attention weighting alpha i,j Multiplying F to calculate the output of the feature attention module while multiplying alpha i,j Multiplying by the adjacency matrix further increases the weight of the important feature:
F′＝α i,j ·F
adj′＝α i,j ·adj
wherein F is the input of the feature attention module, W Q1 ∈R T ，W Q2 ∈R C×T ，W K ∈R C ，b s ∈R b×N×N Adj is the adjacency matrix.
2. The method for predicting the multi-energy load based on the space-time diagram neural network according to claim 1, wherein the step S2 comprises the following steps:
s201: according to the international measurement unit conversion method, the units of the acquired electric, thermal and cold load data are uniformly converted into KW serving as units; the conversion formula is as follows:
1KW＝3.4mBtu/h＝0.284ton-hr
wherein KW is electric load unit, mBtu/h is cold load unit, ton-hr is heat load unit;
s202: for unified electric, heat and cold load data and meteorological characteristic data, detecting time sequence abnormal values by using a box diagram quartile detection method and a first-order exponential smoothing method respectively; if the time sequence abnormal value is detected, setting the time sequence abnormal value as Nan, and filling Nan by adopting a linear interpolation method to obtain the sample data of electricity, heat, cold load and each meteorological characteristic;
s203: respectively carrying out normalization processing on the electric, thermal and cold load and sample data of each meteorological feature to obtain a time sequence data set of electric, thermal and cold load data, meteorological feature data and calendar feature data; the normalization formula is as follows:
wherein the method comprises the steps ofX is the characteristic of each load or weather or calendar, x min For minima of load or weather or calendar features, x max For maximum values of load-or weather-or calendar-characteristics, x nom Normalized data.
3. The method for predicting the multi-energy load based on the space-time diagram neural network according to claim 1, wherein the inverse normalization processing formula in S5 is as follows:
y m ＝y n *(y max -y min )+y min
wherein y is n Is multi-energy load prediction data based on space-time diagram neural network, y min Predicting a minimum value, y, of data for multiple energy loads max Predicting the maximum value of data, y, for the multi-energy load m Is the predicted data after inverse normalization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211609240.2A CN115983448B (en) | 2022-12-14 | 2022-12-14 | Multi-energy load prediction method based on space-time diagram neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211609240.2A CN115983448B (en) | 2022-12-14 | 2022-12-14 | Multi-energy load prediction method based on space-time diagram neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115983448A CN115983448A (en) | 2023-04-18 |
| CN115983448B true CN115983448B (en) | 2023-09-12 |
Family
ID=85971526
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211609240.2A Active CN115983448B (en) | 2022-12-14 | 2022-12-14 | Multi-energy load prediction method based on space-time diagram neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115983448B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117556369B (en) * | 2024-01-12 | 2024-04-19 | 南京信息工程大学 | Power theft detection method and system for dynamically generated residual error graph convolution neural network |
| CN117808175B (en) * | 2024-03-01 | 2024-05-17 | 南京信息工程大学 | DTformer-based short-term multi-energy load prediction method |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111815026A (en) * | 2020-06-09 | 2020-10-23 | 国网山东省电力公司经济技术研究院 | Multi-energy system load prediction method based on feature clustering |
| CN111950793A (en) * | 2020-08-17 | 2020-11-17 | 浙江工业大学 | Comprehensive energy system load prediction method considering multivariate load coupling characteristics |
| CN112508278A (en) * | 2020-12-07 | 2021-03-16 | 国网内蒙古东部电力有限公司电力科学研究院 | Multi-connected system load prediction method based on evidence regression multi-model |
| CN112613666A (en) * | 2020-12-26 | 2021-04-06 | 福建维力能源科技有限公司 | Power grid load prediction method based on graph convolution neural network and transfer learning |
| CN113591368A (en) * | 2021-06-29 | 2021-11-02 | 中国电力科学研究院有限公司 | Comprehensive energy system multi-energy load prediction method and system |
| CN113609762A (en) * | 2021-07-27 | 2021-11-05 | 山东大学 | Electric cooling and heating load combined prediction method and system based on GRU-MTL |
| CN113822481A (en) * | 2021-09-23 | 2021-12-21 | 国网天津市电力公司 | Comprehensive energy load prediction method based on multi-task learning strategy and deep learning |
| CN114021849A (en) * | 2021-11-26 | 2022-02-08 | 国网上海市电力公司 | Combined prediction method and device for regional cooling, heating and power loads |
| KR20220019376A (en) * | 2020-08-10 | 2022-02-17 | 세종대학교산학협력단 | Power consumption prediction system of residential and commercial building using hybrid convolution neural network and method thereof |
| CN114548509A (en) * | 2022-01-18 | 2022-05-27 | 湖南大学 | Multi-type load joint prediction method and system for multi-energy system |
| CN115062864A (en) * | 2022-07-14 | 2022-09-16 | 国网湖南省电力有限公司 | Method and system for decomposing-integrating prediction of medium-term load of power distribution network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416695B (en) * | 2018-02-24 | 2020-07-07 | 合肥工业大学 | Power load probability density prediction method, system and medium based on deep learning |
| CN109214948A (en) * | 2018-09-25 | 2019-01-15 | 新智数字科技有限公司 | A kind of method and apparatus of electric system heat load prediction |
| CN113128793A (en) * | 2021-05-19 | 2021-07-16 | 中国南方电网有限责任公司 | Photovoltaic power combination prediction method and system based on multi-source data fusion |
-
2022
- 2022-12-14 CN CN202211609240.2A patent/CN115983448B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111815026A (en) * | 2020-06-09 | 2020-10-23 | 国网山东省电力公司经济技术研究院 | Multi-energy system load prediction method based on feature clustering |
| KR20220019376A (en) * | 2020-08-10 | 2022-02-17 | 세종대학교산학협력단 | Power consumption prediction system of residential and commercial building using hybrid convolution neural network and method thereof |
| CN111950793A (en) * | 2020-08-17 | 2020-11-17 | 浙江工业大学 | Comprehensive energy system load prediction method considering multivariate load coupling characteristics |
| CN112508278A (en) * | 2020-12-07 | 2021-03-16 | 国网内蒙古东部电力有限公司电力科学研究院 | Multi-connected system load prediction method based on evidence regression multi-model |
| CN112613666A (en) * | 2020-12-26 | 2021-04-06 | 福建维力能源科技有限公司 | Power grid load prediction method based on graph convolution neural network and transfer learning |
| CN113591368A (en) * | 2021-06-29 | 2021-11-02 | 中国电力科学研究院有限公司 | Comprehensive energy system multi-energy load prediction method and system |
| CN113609762A (en) * | 2021-07-27 | 2021-11-05 | 山东大学 | Electric cooling and heating load combined prediction method and system based on GRU-MTL |
| CN113822481A (en) * | 2021-09-23 | 2021-12-21 | 国网天津市电力公司 | Comprehensive energy load prediction method based on multi-task learning strategy and deep learning |
| CN114021849A (en) * | 2021-11-26 | 2022-02-08 | 国网上海市电力公司 | Combined prediction method and device for regional cooling, heating and power loads |
| CN114548509A (en) * | 2022-01-18 | 2022-05-27 | 湖南大学 | Multi-type load joint prediction method and system for multi-energy system |
| CN115062864A (en) * | 2022-07-14 | 2022-09-16 | 国网湖南省电力有限公司 | Method and system for decomposing-integrating prediction of medium-term load of power distribution network |
Non-Patent Citations (1)
| Title |
|---|
| 基于实时出行需求和交通路况的电动汽车充电负荷预测;吴钉捷;李晓露;;电力建设(第08期);第57-67页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115983448A (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115983448B (en) | Multi-energy load prediction method based on space-time diagram neural network | |
| CN107591844B (en) | Active power distribution network robust reconstruction method considering node injection power uncertainty | |
| Yang et al. | Deep learning‐based SCUC decision‐making: An intelligent data‐driven approach with self‐learning capabilities | |
| CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
| CN110796307A (en) | Distributed load prediction method and system for comprehensive energy system | |
| CN112149890A (en) | Comprehensive energy load prediction method and system based on user energy label | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN113591368A (en) | Comprehensive energy system multi-energy load prediction method and system | |
| CN114239991A (en) | Building heat supply load prediction method, device and equipment based on data driving | |
| He et al. | Forecasting power grid states for regional energy markets with deep neural networks | |
| CN115081269A (en) | Bridge thermal boundary condition prediction method, device and medium based on hybrid input neural network model | |
| CN115238854A (en) | Short-term load prediction method based on TCN-LSTM-AM | |
| CN111582588B (en) | Building energy consumption prediction method based on triple convolution fusion GRU | |
| CN115545503B (en) | Power load medium-short term prediction method and system based on parallel time sequence convolutional neural network | |
| CN115310355A (en) | Multi-energy coupling-considered multi-load prediction method and system for comprehensive energy system | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| CN115293406A (en) | Photovoltaic power generation power prediction method based on Catboost and Radam-LSTM | |
| CN115936236A (en) | Method, system, equipment and medium for predicting energy consumption of cigarette factory | |
| CN115238967A (en) | Photovoltaic power prediction method and device combining cloud picture and adjacent power station cluster | |
| Huang et al. | A Machine Learning Based Method for Short-term Urban Rail Transit Passenger Flow Prediction | |
| CN117808175B (en) | DTformer-based short-term multi-energy load prediction method | |
| Kaur et al. | Residential Electricity Demand Prediction using Machine Learning. | |
| Huang et al. | Short-term load forecasting method based on heterogeneous big data of power | |
| Zhang et al. | A study on Short-Term Electricity Load Forecasting for Industrial Parks method using QPSO-TCN Based on Autoencoder | |
| Zhao | Comprehensive Review of Machine Learning-Based Methods for Electricity Load Forecasting |
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 |