CN114819337A - Multi-task learning-based comprehensive energy system multi-load prediction method - Google Patents
Multi-task learning-based comprehensive energy system multi-load prediction method Download PDFInfo
- Publication number
- CN114819337A CN114819337A CN202210437937.XA CN202210437937A CN114819337A CN 114819337 A CN114819337 A CN 114819337A CN 202210437937 A CN202210437937 A CN 202210437937A CN 114819337 A CN114819337 A CN 114819337A
- Authority
- CN
- China
- Prior art keywords
- value
- load prediction
- model
- energy system
- training
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 25
- 230000002159 abnormal effect Effects 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims abstract description 9
- 238000012795 verification Methods 0.000 claims abstract description 9
- 238000010606 normalization Methods 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000005611 electricity Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 2
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/08—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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Operations Research (AREA)
- Bioinformatics & Computational Biology (AREA)
- Pure & Applied Mathematics (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Optimization (AREA)
- Computational Linguistics (AREA)
- Mathematical Analysis (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computational Mathematics (AREA)
- Marketing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Algebra (AREA)
Abstract
The invention relates to the field of comprehensive energy system multi-load prediction, in particular to a comprehensive energy system multi-load prediction method based on multi-task learning and LSTM; determining the composition of various loads and necessary related information by setting an input/output characteristic set; then, preprocessing the determined characteristic data, including filling missing values, identifying abnormal values and carrying out normalization processing; dividing the sample data set into a training set and a verification set; and finally, constructing a comprehensive energy system multi-load prediction model based on multi-task learning and LSTM in a mode of matching offline training with online application, and evaluating the model by using mean absolute value errors MAE and R2.
Description
Technical Field
The invention relates to the field of comprehensive energy system multi-load prediction, in particular to a comprehensive energy system multi-load prediction method based on multi-task learning and LSTM.
Background
Under the current large background that global traditional energy sources are increasingly scarce and carbon emission causes environmental pollution, how to systematically save energy and reduce emission becomes the subject of joint research of global energy industry. The appearance of the comprehensive energy system provides a way for solving series energy problems such as renewable energy abandonment, and the IES is a multi-energy flow heterogeneous coupling system integrating the functions of energy conversion, storage, distribution, transportation and the like, and can meet the requirements of electricity, gas, heat and cold loads at the same time. And IES load prediction is taken as a primary consideration factor of IES operation management and optimized scheduling, so that the method has important significance and practical application value for accurately predicting the load of the integrated energy system.
The single traditional energy system design and planning method neglects the coupling relation among heterogeneous energy sources, so that the flexibility of the system is greatly limited, the current society of the society is not stopped, and the feasible way for the social development in the future can be provided only by converting the single energy supply mode of each energy source into the combined energy supply mode of multiple heterogeneous energy sources. At present, although the traditional load prediction method has the advantages of high calculation speed and the like and has certain effect, with the continuous development of the energy industry, the energy utilization requirement is influenced by various factors such as weather, economy, daily types and the like, so that an accurate mathematical model cannot be established, and the prediction result is not ideal. The artificial intelligence method can better fit the nonlinear relation between the influencing factors and the load without establishing an accurate model during prediction analysis, and therefore, the artificial intelligence method is used for energy load prediction. At present, the load prediction method of IES generally optimizes model parameters and structure, and has certain limitation in processing different forms of energy consumption coupling problems, so that the invention provides a comprehensive energy system multi-load prediction method based on multi-task learning.
Disclosure of Invention
Objects of the invention
The invention aims to solve the problem that the load prediction in the conventional comprehensive energy system is not accurate enough, and provides a multi-load prediction method of the comprehensive energy system based on multi-task learning. The method mainly comprises the steps of determining the composition of various loads and necessary related information by setting an input/output feature set; then, preprocessing the determined characteristic data, including filling missing values, identifying abnormal values and carrying out normalization processing; dividing the sample data set into a training set and a verification set; finally, a base is constructed in a mode of matching offline training with online applicationComprehensive energy system multivariate load prediction model based on multi-task learning and LSTM, and using mean absolute value errors MAE and R 2 The model is evaluated.
(II) technical scheme
In order to achieve the purpose, the method adopts the technical scheme that: through reasonable selection of the input/output characteristic set, the load prediction result is more accurate; then preprocessing the obtained data set to eliminate abnormal conditions of the data in the measuring, transmitting and storing processes; then dividing the data set into a training set and a verification set so as to facilitate the subsequent model training; finally, a model is constructed in a mode of matching offline training with online application, the model structure adopts multi-task learning and LSTM, and the evaluation indexes are mean absolute value errors MAE and R 2 。
The input/output feature set is the key to determining the model's performance capabilities. The method comprises the steps of selecting environment factors, day type information, multi-element load data and the like as input characteristics x, selecting output characteristics y as actual multi-element load data at a moment to be measured, and forming samples { x, y } of the multi-element load prediction problem by the input characteristics x and the output characteristics y together.
Data is easy to have abnormal conditions in the processes of measurement, transmission and storage, and if the part of samples are directly abandoned, the available information for establishing a prediction model can be greatly reduced, and the model prediction performance is reduced. Therefore, missing value filling and abnormal value identification are adopted in the method to ensure the integrity and the goodness of the data set, and normalization processing is carried out on the final data set to prevent the model prediction accuracy from being influenced by larger order difference among variables.
In view of the fact that a large amount of computing resources are needed for building and training the deep learning model, a high-performance server is suitable to be adopted, and therefore the IES electric, thermal and cold multi-load prediction neural network model is built in a mode of matching offline training with online application. First through a high performance server pair modelAnd (4) performing off-line construction and training, and copying the trained model to a corresponding user side computer or terminal equipment for on-line application. The off-line training stage mainly learns the mapping relation between a plurality of input features and the multi-element load through LSTM and multi-task; the online application mainly comprises the steps of inputting the characteristics of the current moment into a trained LSTM and a multi-task learning model to quickly obtain the multi-element load prediction result of the next moment, and finally adopting average absolute value errors MAE and R 2 To evaluate the model accuracy.
(III) advantageous effects
The invention has the beneficial effects that: the relevance among the multiple loads of the comprehensive energy system is learned through multi-task learning, the multiple loads of the comprehensive energy system are predicted more accurately, and the operation management and the optimized scheduling of the comprehensive energy system can be facilitated. Meanwhile, an IES electrical, thermal and cold multi-load prediction neural network model is constructed in a mode of matching offline training with online application, so that the training time can be greatly saved; finally, average absolute value errors MAE and R are adopted 2 To evaluate whether the model accuracy verification method is successful.
Description of the drawings:
FIG. 1 shows a multi-load prediction method for an integrated energy system based on multi-task learning
And establishing a flow chart.
The specific implementation mode is as follows:
the input data set adopted by the invention comprises electricity, heat and cold historical load data, meteorological data and day type data.
Then, carrying out missing value processing and abnormal value identification on the data set, and finally carrying out normalization processing on the data set:
in the formula x std Is a normalized value, x scaled Is the value after the most value normalization, x,
σ、x max And x min Respectively, an original value, a sample mean, a sample standard deviation, a sample minimum value and a sample maximum value.
And dividing the preprocessed data set into a training set and a testing set.
And then, carrying out network modeling, determining partial hyper-parameters according to the established model characteristics, then adopting a random tracking method for the rest hyper-parameters, selecting different search ranges by utilizing different hyper-parameter subspaces to have different influence degrees on the network convergence speed, and accelerating the parameter selection efficiency.
And then training the network model, taking the multidimensional characteristic vector as input, taking the load prediction value as output, and training the network from bottom to top until iteration reaches preset times. And converting the low-dimensional features in the original data set into high-dimensional features layer by layer through a plurality of hidden layers, so that the model learns the hidden mapping relation.
And then adjusting and optimizing the network parameters, inputting the characteristic quantity of the verification set into the trained LSTM-MTL network by adopting an Adam optimization algorithm, comparing the output multivariate load prediction result with a real value, calculating a loss function, and adjusting the network parameters by generations according to the loss function.
Finally, network performance is evaluated, and average absolute value errors MAE and R are selected according to the condition that the constructed multi-element load prediction model needs to perform prediction analysis on a plurality of subtasks at the same time 2 To evaluate whether the model accuracy verification method is successful.
In the formula: n is the number of samples, y i Indicates the actual value of the ith sample,
representing the ith sample prediction value; and the Adam optimizer is selected for the weight updating algorithm.
Evaluation of network output accuracy using R 2 Index evaluation regression task accuracy:
in the formula:
represents the sample mean; if R is 2 And when the index meets the preset requirement, the prediction accuracy of the model meets the requirement, and the training is ended, otherwise, the parameters are adjusted to perform the training again.
So far, the technical solutions of the present invention have been described with reference to the accompanying drawings, but it is obvious to those skilled in the art that the scope of the present invention is not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention in the specification, and the technical scheme after the changes or substitutions is within the protection scope of the invention.
Claims (4)
1. A comprehensive energy system multi-load prediction method based on multitask learning and LSTM is characterized by comprising the following steps: determining the composition of various loads and necessary related information by setting an input/output characteristic set; then, preprocessing the determined characteristic data, including filling missing values, identifying abnormal values and carrying out normalization processing; dividing the sample data set into a training set and a verification set; and finally, constructing a comprehensive energy system multi-load prediction model based on multi-task learning and LSTM in a mode of matching offline training with online application, and evaluating the model by using mean absolute value errors MAE and R2.
2. The method for predicting the multivariate load of the integrated energy system based on the multitask learning and the LSTM according to the claim 1, wherein the composition of various types of loads and necessary related information are determined by setting an input/output characteristic set, and the main steps comprise:
step 1: determining a required input data set which comprises basic electricity, heat and cold historical load data and meteorological data such as rainfall, air temperature, humidity and the like, and increasing the data of the types of days for increasing prediction accuracy hackers;
step 2: determining a required output data set, namely multivariate load prediction data;
step 3: and after the data needing to be acquired is determined, acquiring the data.
3. The method for predicting the multivariate load of the integrated energy system based on the multitask learning and the LSTM according to the claim 1, wherein the pre-processing is carried out on the determined characteristic data, the pre-processing comprises the steps of filling missing values, identifying abnormal values and normalizing, and the main steps comprise:
step 1: carrying out missing value processing on the acquired data set;
step 2: carrying out abnormal value identification on the acquired data set;
step 3: and (3) carrying out normalization processing on the acquired data set:
in the formula x std Is a normalized value, x scaled Is the value after the most value normalization, x,
σ、x max And x min Respectively, the original value, the sample mean, the sample standard deviation, the sample minimum value and the sample maximumA large value.
4. The multitask learning and LSTM based comprehensive energy system multi-load prediction method according to claim 1, wherein the multitask learning and LSTM based comprehensive energy system multi-load prediction model is constructed by means of off-line training and on-line application, and the model is evaluated by using mean absolute value errors MAE and R2, and the method mainly comprises the following steps:
step 1: modeling a network, determining partial hyper-parameters according to the established model characteristics, then adopting a random tracking method for the rest hyper-parameters, selecting different search ranges by utilizing different hyper-parameter subspaces to have different influence degrees on the network convergence speed so as to accelerate the parameter selection efficiency;
step 2: and training the network model, taking the multi-dimensional characteristic vector as input, taking the load prediction value as output, and training the network from bottom to top until iteration reaches a preset number. Converting the low-dimensional features in the original data set into high-dimensional features layer by layer through a plurality of hidden layers, and enabling the model to learn hidden mapping relations;
step 3: optimizing the network parameters, inputting the characteristic quantity of the verification set into the trained LSTM-MTL network by adopting an Adam optimization algorithm, comparing the output multivariate load prediction result with a real value, calculating a loss function, and adjusting the network parameters by generations according to the loss function;
step 4: finally, evaluating the network performance, and considering that the constructed multivariate load prediction model needs to carry out prediction analysis on a plurality of subtasks at the same time, selecting average absolute value errors MAE and R2 to evaluate whether the model accuracy verification method is successful;
in the formula: n is the number of samples, y i Indicates the actual value of the ith sample,
representing the ith sample prediction value; selecting an Adam optimizer for the weight updating algorithm;
in the formula:
represents the sample mean; if R is 2 And when the index meets the preset requirement, the prediction accuracy of the model meets the requirement, and the training is ended, otherwise, the parameters are adjusted to perform the training again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210437937.XA CN114819337A (en) | 2022-04-25 | 2022-04-25 | Multi-task learning-based comprehensive energy system multi-load prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210437937.XA CN114819337A (en) | 2022-04-25 | 2022-04-25 | Multi-task learning-based comprehensive energy system multi-load prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114819337A true CN114819337A (en) | 2022-07-29 |
Family
ID=82506979
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210437937.XA Pending CN114819337A (en) | 2022-04-25 | 2022-04-25 | Multi-task learning-based comprehensive energy system multi-load prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114819337A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116307136A (en) * | 2023-02-24 | 2023-06-23 | 国网安徽省电力有限公司营销服务中心 | Deep reinforcement learning-based energy system parameter optimization method, system, device and storage medium |
-
2022
- 2022-04-25 CN CN202210437937.XA patent/CN114819337A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116307136A (en) * | 2023-02-24 | 2023-06-23 | 国网安徽省电力有限公司营销服务中心 | Deep reinforcement learning-based energy system parameter optimization method, system, device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| CN109461025B (en) | Electric energy substitution potential customer prediction method based on machine learning | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| CN110796307B (en) | Distributed load prediction method and system for comprehensive energy system | |
| CN113205207A (en) | XGboost algorithm-based short-term power consumption load fluctuation prediction method and system | |
| CN111814956B (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| CN114676941B (en) | Electric-thermal load combined self-adaptive prediction method and device for park comprehensive energy system | |
| CN112149890A (en) | Comprehensive energy load prediction method and system based on user energy label | |
| CN115099511A (en) | Photovoltaic power probability estimation method and system based on optimized copula | |
| CN111815026A (en) | Multi-energy system load prediction method based on feature clustering | |
| CN112949207A (en) | Short-term load prediction method based on improved least square support vector machine | |
| CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
| CN115358437A (en) | Power supply load prediction method based on convolutional neural network | |
| CN114119273A (en) | Park comprehensive energy system non-invasive load decomposition method and system | |
| CN114819395A (en) | Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model | |
| CN111815039A (en) | Weekly scale wind power probability prediction method and system based on weather classification | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| Jayashankara et al. | A novel approach for short-term energy forecasting in smart buildings | |
| CN114819337A (en) | Multi-task learning-based comprehensive energy system multi-load prediction method | |
| Zhang et al. | The power big data-based energy analysis for intelligent community in smart grid | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN117113086A (en) | Energy storage unit load prediction method, system, electronic equipment and medium | |
| CN112288187A (en) | Big data-based electricity sales amount prediction method | |
| CN115481788A (en) | Load prediction method and system for phase change energy storage system | |
| CN113780686A (en) | Distributed power supply-oriented virtual power plant operation scheme optimization method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |