CN111950868A - Comprehensive energy system load scene generation method based on generation countermeasure network - Google Patents
Comprehensive energy system load scene generation method based on generation countermeasure network Download PDFInfo
- Publication number
- CN111950868A CN111950868A CN202010735071.1A CN202010735071A CN111950868A CN 111950868 A CN111950868 A CN 111950868A CN 202010735071 A CN202010735071 A CN 202010735071A CN 111950868 A CN111950868 A CN 111950868A
- Authority
- CN
- China
- Prior art keywords
- network
- scene
- load
- energy system
- model
- 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.)
- Granted
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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Feedback Control In General (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a comprehensive energy system load scene generation method based on a generation countermeasure network, which comprises the following steps: s1: acquiring load data of a comprehensive energy system, and establishing a cold, heat and power load sample data set; s2: constructing and generating a confrontation network model; s3: alternately training a generator network and a discriminator network D; s4: and generating a comprehensive energy system load scene. The invention captures the internal complex deep nonlinear structural characteristics by using the implicit relationship of the generated confrontation network learning sample data. And when the real data is lacked or the diversity of the real data is insufficient, generating a comprehensive energy system multi-load scene with the statistical characteristics similar to the real scene. The scene generated by the method of the invention can provide decision support for load prediction, abnormal detection and operation scheduling, and the flexibility and reliability of the system are improved.
Description
Technical Field
The invention relates to a comprehensive energy system load scene generation method based on a generation countermeasure network, and belongs to the technical field of operation and analysis of comprehensive energy systems.
Background
In recent years, as the degree of informatization and digitization has been increasing, mining techniques based on large data have become mature, and the value of data has been recognized and utilized. The comprehensive energy data with multi-source heterogeneity, complex relevance and real-time interactivity has huge information value, value information is extracted from the energy big data, a large number of scenes are calculated and analyzed, the change characteristics of loads are reflected, and decision support can be provided for key problems such as fault detection, load prediction, safety assessment and energy management. However, such data is often not readily available due to limitations in data privacy, data security, acquisition costs, etc., which prevents large-scale availability of data, posing challenges to further deployment of data-driven technologies in the energy field.
Therefore, it is necessary to provide a method for generating a comprehensive energy system scene, which finds and learns the potential association and the complex coupling in the training data when the real data is absent or the diversity of the real data is insufficient, mines the characteristics of the data, directly learns the probability distribution of the load time series data, and automatically generates a comprehensive energy system multi-load scene with the statistical characteristics similar to that of the real scene without explicitly modeling the probability distribution.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects of the prior art and provide a comprehensive energy system load scene generation method based on a generation countermeasure network, and when real data is lacked or the diversity of the real data is insufficient, a comprehensive energy system multi-element load scene with the similar statistical characteristics to the real scene is generated.
In order to achieve the above object, the present invention provides a method for generating a comprehensive energy system load scenario based on a generation countermeasure network, comprising the following steps:
s1: acquiring load data of a comprehensive energy system, and establishing a cold, heat and power load sample data set;
s2: constructing and generating a confrontation network model;
s3: alternately training a generator network and a discriminator network D;
s4: and generating a comprehensive energy system load scene.
Preferably, the step S1 includes the steps of:
s11: acquiring cold, heat and power load data of each day, acquiring N load values in one day, and acquiring the load values at the same interval;
s12: preprocessing the acquired load value, and completing lost data or correcting abnormal data;
s13: the load values are normalized according to the following equation, scaling the value of each datum to a value between 0 and 1:
in the formula: x denotes a sample data set, x ═ x1，x2，…，xn]N is the total number of samples; x is the number ofminRepresenting a minimum value in the sample data set; x is the number ofmaxRepresenting a maximum value in the sample data set; x is the number ofiIs the ith data in the sample data set, and the value range of i is [1, n]。
Preferably, the step S2 includes the steps of:
s21: defining a generator model G consisting of a full-connection layer and four deconvolution layers, performing batch normalization processing after each calculation layer, and finally outputting by using a Sigmoid activation function;
wherein, the Sigmoid activation function is:
s22: defining a discriminator model G consisting of three convolution layers and a full-link layer, wherein except the first convolution layer and the third convolution layer, the second convolution layer and the full-link layer are subjected to batch normalization processing, and finally LeakyRelu is used as an activation function to be output;
wherein, the LeakyRelu activation function is:
preferably, the input to the generator model G is a set of random arrays of size 96 × 1.
Preferably, the filter sizes of the first, second and third convolutional layers are all 5 × 5.
Preferably, the step S3 includes the steps of:
s31: the noise generation unit N generates a random noise sequence z of a certain lengthi；
S32: random noise sequence ziInputting a generator model G, and mapping the generator model G into a scene G (z) through a network of the generator model G;
s33: scene G (z) generated by generator model G and sampled training sample scene xiInputting the values into a discriminator network D, wherein the discriminator network D respectively gives values D (x) and D (G (z)) representing the scene as true and false;
s34: respectively calculating loss functions of the discriminator model G and the generator model G according to the following formula;
LG＝-EZ[D(G(z))]，
LD＝-EX[D(x)+EZ[D(G(z))]]，
wherein L isGRepresenting the loss function of the generator, LDRepresenting the loss function of the arbiter, EZExpected value, E, representing the corresponding random noise distributionXAn expected value representing a corresponding true sample distribution;
s35: and reversely propagating the result of the S34 to the generator model G and the discriminator model, and updating and adjusting the network parameter theta according to the following formulagAnd a network parameter thetad；
Wherein, thetag,t-1And thetad,t-1The parameters of the generator network and the parameters of the discriminator network D are generated before the t round of training begins;
s36: repeating the steps S31-S35 until the network parameter thetagAnd a network parameter thetadUntil convergence.
Preferably, in step S4, the trained generator model G and the discriminator model D are fixed, and the generator model G is used for the random noise sequence z input from the inputiAnd simulating to generate and output phase cooling heat and power load data.
The invention achieves the following beneficial effects:
the invention discloses a comprehensive energy system load scene generation method based on generation of a countermeasure network, which utilizes the implicit relation of the generated countermeasure network learning sample data to capture the internal complex deep nonlinear structure characteristics and generate new data with the statistical distribution characteristics similar to the real data. When the real data are lacked or the diversity of the real data is insufficient, the comprehensive energy system multi-load scene with the statistical characteristics similar to the real scene is automatically generated, explicit modeling on probability distribution is not needed, decision support is provided for abnormal detection and operation scheduling, the flexibility and the reliability of the system are improved, and the problem that the diversity of the real data is lacked or the diversity of the real data is insufficient is solved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a graph of loss function variation for the generator and arbiter training process of the present invention;
FIG. 3 is a comparison of generated scene and real scene sample data of the present invention;
fig. 4 is a comparison graph of probability distribution characteristics of generated scene and real scene data of the present invention.
Detailed Description
The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
A comprehensive energy system load scene generation method based on generation of a countermeasure network comprises the following steps:
s1: acquiring load data of a comprehensive energy system, and establishing a cold, heat and power load sample data set;
s2: constructing and generating a confrontation network model;
s3: alternately training a generator network and a discriminator network D;
s4: and generating a comprehensive energy system load scene.
Further, the step S1 includes the following steps:
s11: acquiring cold, heat and power load data of each day, acquiring N load values in one day, and acquiring the load values at the same interval;
s12: preprocessing the acquired load value, and completing lost data or correcting abnormal data;
s13: the load values are normalized according to the following equation, scaling the value of each datum to a value between 0 and 1:
in the formula: x denotes a sample data set, x ═ x1，x2，…，xn]N is the total number of samples; x is the number ofminRepresenting a minimum value in the sample data set; x is the number ofmaxRepresenting a maximum value in the sample data set; x is the number ofiIs the ith data in the sample data set, and the value range of i is [1, n]。
Further, the step S2 includes the following steps:
s21: defining a generator model G consisting of a full-connection layer and four deconvolution layers, performing batch normalization processing after each calculation layer, and finally outputting by using a Sigmoid activation function;
wherein, the Sigmoid activation function is:
s22: defining a discriminator model G consisting of three convolution layers and a full-link layer, wherein except the first convolution layer and the third convolution layer, the second convolution layer and the full-link layer are subjected to batch normalization processing, and finally LeakyRelu is used as an activation function to be output;
wherein, the LeakyRelu activation function is:
further, the input to the generator model G is a set of random arrays of size 96 × 1.
Further, the filter sizes of the first, second, and third convolutional layers are all 5 × 5.
Further, the step S3 includes the following steps:
s31: the noise generation unit N generates a random noise sequence z of a certain lengthi；
S32: random noise sequence ziInputting a generator model G, and mapping the generator model G into a scene G (z) through a network of the generator model G;
s33: scene G (z) generated by generator model G and sampled training sample scene xiInputting the values into a discriminator network D, wherein the discriminator network D respectively gives values D (x) and D (G (z)) representing the scene as true and false;
s34: respectively calculating loss functions of the discriminator model G and the generator model G according to the following formula;
LG＝-EZ[D(G(z))]，
LD＝-EX[D(x)+EZ[D(G(z))]]，
wherein L isGRepresenting the loss function of the generator, LDRepresenting the loss function of the arbiter, EZExpected value, E, representing the corresponding random noise distributionXAn expected value representing a corresponding true sample distribution;
s35: and reversely propagating the result of the S34 to the generator model G and the discriminator model, and updating and adjusting the network parameter theta according to the following formulagAnd a network parameter thetad；
Wherein, thetag,t-1And thetad,t-1The parameters of the generator network and the parameters of the discriminator network D are generated before the t round of training begins;
s36: repeating the steps S31-S35 until the network parameter thetagAnd a network parameter thetadUntil convergence.
Further, in step S4, the generator model G and the discriminator model D after training are fixed, and the generator model G is used for the random noise sequence z input from the inputiAnd simulating to generate and output phase cooling heat and power load data.
Example (b):
the embodiment uses the load data of residents in the certain area 2013 and 2017 to generate a typical scene.
The generator and arbiter training process loss function variation graph of this embodiment is shown in fig. 2. After 2000 times of iterative training, the loss functions of the generator and the discriminator are basically converged, which shows that the generated countermeasure network reaches the balance at the moment. The loss function of the generator steadily fluctuates around the value 0 after continuously decreasing, which indicates that the artificial sample distribution synthesized by the generator is very close to the real sample distribution.
And selecting partial curves from the test set to verify the generated scene. A comparison graph of sample data of a generated scene and a real scene is shown in fig. 3, a curve generated by the model has an output trend consistent with that of a real sample, and is in accordance with an actual load characteristic but not completely consistent with an original sample, and a consistent fluctuation rule is maintained between the two. This means that the model does not over-fit the data in the training set, but learns the inherent relevance of the sample data, and then generates completely new data, which indicates that the model has strong generalization capability.
And analyzing the probability distribution characteristics of the generated scene by adopting a Cumulative Distribution Function (CDF). The comparison graph of the probability distribution characteristics of the generated scene data and the real scene data of the invention is shown in FIG. 4. The CDF curve of the generated load scene set is highly fitted, the probability distribution characteristics of the generated scene are very similar to those of the historical scene, and the GAN model is proved to have the learning capability of learning the potential probability distribution of given data.
According to the comprehensive energy system load scene generation method, the comprehensive energy system load scene is generated by adopting the generation countermeasure network, the fluctuation and probability distribution characteristics of the load under different conditions can be effectively represented, and the load scene can be directly generated, so that the model has strong generalization capability. . The generated data is ensured to have high similarity with the actual scene, and meanwhile, the sample shows the diversity of mode changes.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (7)
1. A comprehensive energy system load scene generation method based on generation countermeasure network is characterized by comprising the following steps:
s1: acquiring load data of a comprehensive energy system, and establishing a cold, heat and power load sample data set;
s2: constructing and generating a confrontation network model;
s3: alternately training a generator network and a discriminator network D;
s4: and generating a comprehensive energy system load scene.
2. The method for generating the integrated energy system load scenario based on generation of countermeasure network as claimed in claim 1, wherein said step S1 comprises the following steps:
s11: acquiring cold, heat and power load data of each day, acquiring N load values in one day, and acquiring the load values at the same interval;
s12: preprocessing the acquired load value, and completing lost data or correcting abnormal data;
s13: the load values are normalized according to the following equation, scaling the value of each datum to a value between 0 and 1:
in the formula: x denotes a sample data set, x ═ x1，x2，…，xn]N is the total number of samples; x is the number ofminRepresenting a sample data setMinimum value of (1); x is the number ofmaxRepresenting a maximum value in the sample data set; x is the number ofiIs the ith data in the sample data set, and the value range of i is [1, n]。
3. The method for generating the integrated energy system load scenario based on generation of countermeasure network as claimed in claim 1, wherein said step S2 comprises the following steps:
s21: defining a generator model G consisting of a full-connection layer and four deconvolution layers, performing batch normalization processing after each calculation layer, and finally outputting by using a Sigmoid activation function;
wherein, the Sigmoid activation function is:
s22: defining a discriminator model G consisting of three convolution layers and a full-link layer, wherein except the first convolution layer and the third convolution layer, the second convolution layer and the full-link layer are subjected to batch normalization processing, and finally LeakyRelu is used as an activation function to be output;
wherein, the LeakyRelu activation function is:
4. the method as claimed in claim 3, wherein the generator model G is a set of random arrays with a size of 96 x 1.
5. The method for generating the comprehensive energy system load scene based on the generation countermeasure network of claim 3, wherein the filter sizes of the first layer of convolutional layer, the second layer of convolutional layer and the third layer of convolutional layer are all 5 x 5.
6. The method for generating the integrated energy system load scenario based on generation of countermeasure network as claimed in claim 1, wherein said step S3 comprises the following steps:
s31: the noise generation unit N generates a random noise sequence z of a certain lengthi；
S32: random noise sequence ziInputting a generator model G, and mapping the generator model G into a scene G (z) through a network of the generator model G;
s33: scene G (z) generated by generator model G and sampled training sample scene xiInputting the values into a discriminator network D, wherein the discriminator network D respectively gives values D (x) and D (G (z)) representing the scene as true and false;
s34: respectively calculating loss functions of the discriminator model G and the generator model G according to the following formula;
LG＝-EZ[D(G(z))]，
LD＝-EX[D(x)+EZ[D(G(z))]]，
LD＝-EX[D(x)+EZ[D(G(z))]]，
wherein L isGRepresenting the loss function of the generator, LDRepresenting the loss function of the arbiter, EZExpected value, E, representing the corresponding random noise distributionXAn expected value representing a corresponding true sample distribution;
s35: and reversely propagating the result of the S34 to the generator model G and the discriminator model, and updating and adjusting the network parameter theta according to the following formulagAnd a network parameter thetad；
Wherein, thetag,t-1And thetad,t-1For when the t round trainsGenerating parameters of a generator network and parameters of a discriminator network D before training;
s36: repeating the steps S31-S35 until the network parameter thetagAnd a network parameter thetadUntil convergence.
7. The method for generating the comprehensive energy system load scene based on the generation countermeasure network of claim 1, wherein in step S4, the generator model G and the discriminator model D are fixed, and the generator model G is used for the random noise sequence z from the inputiAnd simulating to generate and output phase cooling heat and power load data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010735071.1A CN111950868B (en) | 2020-07-28 | 2020-07-28 | Comprehensive energy system load scene generation method based on generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010735071.1A CN111950868B (en) | 2020-07-28 | 2020-07-28 | Comprehensive energy system load scene generation method based on generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111950868A true CN111950868A (en) | 2020-11-17 |
| CN111950868B CN111950868B (en) | 2022-11-15 |
Family
ID=73339673
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010735071.1A Active CN111950868B (en) | 2020-07-28 | 2020-07-28 | Comprehensive energy system load scene generation method based on generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111950868B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113256017A (en) * | 2021-06-07 | 2021-08-13 | 国网山东省电力公司电力科学研究院 | Short-term load prediction method and system |
| CN113269314A (en) * | 2021-06-16 | 2021-08-17 | 福州大学 | New energy power generation scene data migration method based on generation countermeasure network |
| CN113673159A (en) * | 2021-08-19 | 2021-11-19 | 东北电力大学 | Renewable energy space-time scene generation method based on federal deep-generation learning |
| WO2022143021A1 (en) * | 2020-12-29 | 2022-07-07 | 沃太能源股份有限公司 | Neural network generating method and apparatus, power forecasting method and apparatus, and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490659A (en) * | 2019-08-21 | 2019-11-22 | 广东电网有限责任公司 | A kind of customer charge curve generation method based on GAN |
-
2020
- 2020-07-28 CN CN202010735071.1A patent/CN111950868B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490659A (en) * | 2019-08-21 | 2019-11-22 | 广东电网有限责任公司 | A kind of customer charge curve generation method based on GAN |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022143021A1 (en) * | 2020-12-29 | 2022-07-07 | 沃太能源股份有限公司 | Neural network generating method and apparatus, power forecasting method and apparatus, and storage medium |
| CN113256017A (en) * | 2021-06-07 | 2021-08-13 | 国网山东省电力公司电力科学研究院 | Short-term load prediction method and system |
| CN113269314A (en) * | 2021-06-16 | 2021-08-17 | 福州大学 | New energy power generation scene data migration method based on generation countermeasure network |
| CN113269314B (en) * | 2021-06-16 | 2022-10-28 | 福州大学 | New energy power generation scene data migration method based on generation countermeasure network |
| CN113673159A (en) * | 2021-08-19 | 2021-11-19 | 东北电力大学 | Renewable energy space-time scene generation method based on federal deep-generation learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111950868B (en) | 2022-11-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111950868B (en) | Comprehensive energy system load scene generation method based on generation countermeasure network | |
| CN108665460B (en) | Image quality evaluation method based on combined neural network and classified neural network | |
| CN111210024A (en) | Model training method and device, computer equipment and storage medium | |
| CN109033513B (en) | Power transformer fault diagnosis method and power transformer fault diagnosis device | |
| CN112367273B (en) | Flow classification method and device of deep neural network model based on knowledge distillation | |
| Moayyed et al. | A Cyber-Secure generalized supermodel for wind power forecasting based on deep federated learning and image processing | |
| CN104869126A (en) | Network intrusion anomaly detection method | |
| CN114219147A (en) | Power distribution station fault prediction method based on federal learning | |
| CN113486078A (en) | Distributed power distribution network operation monitoring method and system | |
| CN117113262B (en) | Network traffic identification method and system | |
| Costilla-Enriquez et al. | Attack power system state estimation by implicitly learning the underlying models | |
| Fujita et al. | An approach for intelligent evaluation of the state of complex autonomous objects based on the wavelet analysis | |
| Li et al. | A LSTM-based method for comprehension and evaluation of network security situation | |
| CN117218425A (en) | Power generation loss analysis method and system for photovoltaic power station | |
| CN116523001A (en) | Method, device and computer equipment for constructing weak line identification model of power grid | |
| CN116545733A (en) | Power grid intrusion detection method and system | |
| CN113988189B (en) | Migration fault diagnosis method for cross-wind turbine generator | |
| CN110705638A (en) | Credit rating prediction classification method using deep network learning fuzzy information feature technology | |
| CN115578325A (en) | Image anomaly detection method based on channel attention registration network | |
| Xiang et al. | An Improved Multiple Imputation Method Based on Chained Equations for Distributed Photovoltaic Systems | |
| CN114760128A (en) | Network abnormal flow detection method based on resampling | |
| CN114154415A (en) | Equipment life prediction method and device | |
| CN114298300A (en) | Uncertainty prediction method, uncertainty prediction device, electronic device, and storage medium | |
| CN113536508A (en) | Method and system for classifying manufacturing network nodes | |
| CN113328986A (en) | Network flow abnormity detection method based on combination of convolutional neural network and LSTM |
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 |