CN114862123A - Comprehensive energy system scene generation method and device - Google Patents
Comprehensive energy system scene generation method and device Download PDFInfo
- Publication number
- CN114862123A CN114862123A CN202210376573.9A CN202210376573A CN114862123A CN 114862123 A CN114862123 A CN 114862123A CN 202210376573 A CN202210376573 A CN 202210376573A CN 114862123 A CN114862123 A CN 114862123A
- Authority
- CN
- China
- Prior art keywords
- data
- discriminator
- scene
- generator
- scene data
- 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 43
- 238000012549 training Methods 0.000 claims abstract description 24
- 230000006870 function Effects 0.000 claims description 62
- 230000004913 activation Effects 0.000 claims description 9
- 230000015654 memory Effects 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 abstract description 6
- 208000009119 Giant Axonal Neuropathy Diseases 0.000 description 29
- 201000003382 giant axonal neuropathy 1 Diseases 0.000 description 29
- 238000010586 diagram Methods 0.000 description 19
- 238000009826 distribution Methods 0.000 description 16
- 238000004590 computer program Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 238000005315 distribution function Methods 0.000 description 6
- 230000005611 electricity Effects 0.000 description 6
- 230000001186 cumulative effect Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 239000008186 active pharmaceutical agent Substances 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- 239000003921 oil Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001172 regenerating effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000000007 visual effect Effects 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
-
- 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
- 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)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Educational Administration (AREA)
- Computational Linguistics (AREA)
- Game Theory and Decision Science (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a method and a device for generating a comprehensive energy system scene, which effectively generate scene data conforming to the characteristics of a state variable data set by utilizing a training process of a GAN network according to the state variable data set and random noise in a sample set, generate the comprehensive energy system scene based on all the scene data and provide a foundation for IES planning and regulation.
Description
Technical Field
The invention relates to a scene generation method and device of an integrated energy system, and belongs to the field of integrated energy systems.
Background
An Integrated Energy System (IES) is a system that integrates various energy sources such as coal, oil, natural gas, heat energy, wind energy and the like in a certain area by using a rapid and accurate information transmission technology and an innovative intelligent regulation and control strategy and taking electric energy as a core, so as to realize unified planning, perfect operation, mutual-assistance management and economic complementation among various energy systems.
How to plan and regulate the IES is an important issue of research, and the first problem facing the IES research is: for the state variable of the IES, when the collected variable sample is a small sample (i.e. there is less data in the sample), a corresponding scenario cannot be generated based on the small sample.
Disclosure of Invention
The invention provides a method and a device for generating a comprehensive energy system scene, which solve the problems disclosed in the background technology.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
an integrated energy system scenario generation method, comprising:
collecting a sample set of state variables of the comprehensive energy system;
traversing a state variable data group in a sample set, inputting a group of state variable data and random noise into a preset GAN network, training the GAN network, and generating scene data according with the characteristics of the group of state variable data; in the GAN network, a generator generates scene data according to random noise, a discriminator calculates a loss function according to state variable data and the scene data, if the scene data is judged to not meet preset requirements based on the loss function, parameters of the discriminator and the generator are modified, the generator regenerates the scene data, and the discriminator recalculates the loss function until the scene data meets the preset requirements, and the discriminator outputs the scene data meeting the preset requirements;
and generating a comprehensive energy system scene according to the scene data.
In the GAN network, all the fully connected layers except in the generator and the discriminator are replaced with convolutional layers, and no pooling layer is employed.
In the GAN network, the data output by the generator and the discriminator are normalized.
The discriminator adopts a LeakRelu activation function, and the generator adopts a Relu activation function.
And calculating a loss function by adopting a gradient penalty method.
The loss function comprises a loss function of the generator and a loss function of the discriminator;
the loss function of the generator is:
gen_cost＝D(G(Z))
the penalty function for the discriminator is:
disc_cost＝-D(G(Z))+D(X)+γ*d(D(X′))
wherein Z is random noise, D is a discriminator, g (Z) is scene data, D (g (Z)) is output data of D after g (Z) is input, X is state variable data, and a parameter X' ═ X θ + (1- θ) × X fake θ is a generated random number, X fake For each batch of scene data, D (D (X ')) is the discriminator gradient, D (X ') is the output data of D after inputting X ', and gamma is the penalty parameter of gradient.
Parameters of the discriminator and the generator are modified by an error back propagation algorithm.
An integrated energy system scenario generation apparatus, comprising:
an acquisition module: collecting a sample set of state variables of the comprehensive energy system;
a scene data generation module: traversing a state variable data group in a sample set, inputting a group of state variable data and random noise into a preset GAN network, training the GAN network, and generating scene data according with the characteristics of the group of state variable data; in the GAN network, a generator generates scene data according to random noise, a discriminator calculates a loss function according to state variable data and the scene data, if the scene data is judged to not meet preset requirements based on the loss function, parameters of the discriminator and the generator are modified, the generator regenerates the scene data, and the discriminator recalculates the loss function until the scene data meets the preset requirements, and the discriminator outputs the scene data meeting the preset requirements;
a scene generation module: and generating a comprehensive energy system scene according to the scene data.
A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform an integrated energy system scenario generation method.
A computing device comprising one or more processors, one or more memories, and one or more programs stored in the one or more memories and configured to be executed by the one or more processors, the one or more programs including instructions for performing an integrated energy system scenario generation method.
The invention achieves the following beneficial effects: according to the invention, the scene data which accords with the characteristics of the state variable data set is effectively generated by utilizing the training process of the GAN network according to the state variable data set and the random noise in the sample set, and the comprehensive energy system scene is generated based on all the scene data, thereby providing a foundation for IES planning and regulation.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
fig. 2 is a diagram of a GAN network architecture;
FIG. 3 is a diagram of a structure of the discriminator;
FIG. 4 is a diagram of a generator configuration;
FIG. 5(a) is a graph of the time series of the original single day electricity prices;
FIG. 5(b) is a single day electricity price time series diagram for round 5;
FIG. 5(c) is a single day price time series chart for the 20 th round;
FIG. 5(d) is a diagram of a 40 th round of single day price time series;
FIG. 6(a) is a graph of the original single-day load power time series;
FIG. 6(b) is a power time series diagram of the single day load of the 5 th round;
FIG. 6(c) is a power time series diagram of the single day load for the 20 th round;
FIG. 6(d) is a 40 th round single day load power time series diagram;
FIG. 7(a) is a graph of the original single day temperature time series;
FIG. 7(b) is a temperature-time sequence chart of a single day for the 5 th round;
FIG. 7(c) is a diagram of a 20 th round of a single day temperature-time series;
FIG. 7(d) is a 40 th round single day temperature time series chart;
FIG. 8(a) is a graph of the time series of the original single day photovoltaic output;
FIG. 8(b) is a time series diagram of photovoltaic output of the 5 th round of single day;
FIG. 8(c) is a graph of photovoltaic output time series for the 20 th round of single day;
FIG. 8(d) is a diagram of a 40 th round of single-day photovoltaic output time series;
FIG. 9(a) is a graph of the original 16 day fan out time series;
FIG. 9(b) is a graph showing the time series of the fan output on the 16 th day of the 5 th round;
FIG. 9(c) is a graph of fan out time series on day 16 of the 20 th round;
FIG. 9(d) is a graph of fan out time series on day 16 of the 40 th round;
FIG. 10(a) is a plot of the original 16 day stroke rate time series;
FIG. 10(b) is a graph of the time series of the stroke rate on day 16 of round 5;
FIG. 10(c) is a graph of the time series of the stroke rate on day 16 of the 20 th round;
FIG. 10(d) is a plot of the time series of stroke rates on day 16 of round 40;
FIG. 11 is a graph of a cumulative fan output distribution function;
FIG. 12 is a graph of a photovoltaic contribution cumulative distribution function;
FIG. 13 is a fan output autocorrelation coefficient chart;
FIG. 14 is a photovoltaic output autocorrelation coefficient plot;
FIG. 15(a) is a raw load distribution histogram;
FIG. 15(b) is a 5 th round load distribution histogram;
FIG. 15(c) is a 20 th round load distribution histogram;
FIG. 15(d) is a 40 th round load distribution histogram;
FIG. 16(a) is a histogram of raw fan output distribution;
FIG. 16(b) is a 5 th-wheel fan output distribution histogram;
FIG. 16(c) is a 20 th wheel fan output distribution histogram;
FIG. 16(d) is a 40 th wheel fan output distribution histogram.
Detailed Description
The invention is further described below with reference to the accompanying drawings. 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.
As shown in fig. 1, a method for generating an integrated energy system scenario includes the following steps:
and 3, generating a comprehensive energy system scene according to the scene data.
According to the method, the scene data which accord with the characteristics of the state variable data set is effectively generated by utilizing the training process of the GAN network according to the state variable data set and the random noise in the sample set, and the comprehensive energy system scene is generated based on all the scene data, so that a foundation is provided for IES planning and regulation.
Before the implementation of the method, the selection of an experimental platform is carried out: the model (namely the GAN network) is built and operated on the Spyder, the Python language is adopted, the Spyder is an integrated development environment of Python, and compared with other Python development environments, the model has the greatest advantage that the values of an array and external variables in a program can be viewed through a working space like Matlab.
Besides the experiment platform, the experiment also depends on a Tensorflow function library in an important way. Tensorflow is an open source framework introduced by Google, which contains many APIs well-suited for building and training neural networks, and many of the latest research results are rapidly submitted to Tensorflow libraries, well-suited for deep-learning researchers.
The name of TensorFlow contains the two most important concepts of Tensor and Flow, Tensor is the meaning of Tensor, the existence form in a program is similar to a multidimensional array, Flow is the meaning of Flow, and the TensorFlow defines the characteristic that the TensorFlow variables are mutually converted through calculation, namely, a value in a single modification Tensor cannot be the same as that of a common array, namely, except an initialization operation, each change of TensorFlow needs to be an integral change.
Because the Tensorflow is mainly used for deep learning with a neural network as a core, if a large number of variables used in the neural network are operated, the running resources of a computer are greatly wasted, and the running speed of a program is slowed down. Therefore, the Tensorflow executes well-defined operation in a Session (Session) mode, namely, the Session owns and manages all resources of the runtime of the Tensorflow program, and the Tensorflow helps a system to recycle the resources after the computation is completed so as to avoid resource leakage. The context manager, which is often used with Python, uses a session to write the Tensorflow's computation into a with statement.
After the structural design is finished, a sample set of state variables of the comprehensive energy system is collected, the sample set comprises a plurality of state variable data groups, the types of the state variables of each group are consistent, the data groups generally exist in a time sequence form, for example, load, wind energy, solar energy, electricity price and weather data of France which are obtained by sorting after being uploaded and downloaded from a website https:// data.open-power-system-data.org/with 1 hour as a time scale are randomly captured from 2016, 1 month and 1 day to 2016, 12 month and 31 days as samples, and 40 days of the data are obtained from 2016, 1 month and 1 day to 2016, 12 month and 31 days; the 24 hours a day contains 8 time sequences of unit electricity price, total load, total wind power generation, wind speed, temperature, diffusion altitude, vertical altitude, total photovoltaic power generation and the like, and the sampling interval is 1 hour, so that the size of the sample set is 24 x 8.
The overall structure of the GAN network is shown in fig. 2, and adopts a feedforward neural network structure, including an arbiter and a generator. In the GAN network, a generator generates scene data according to random noise, a discriminator calculates a loss function according to state variable data and the scene data, if the scene data is judged to not meet preset requirements based on the loss function, parameters of the discriminator and the generator are modified by adopting an error Back Propagation (BP) algorithm, the generator regenerates the scene data, the discriminator recalculates the loss function until the scene data meets the preset requirements, and the discriminator outputs the scene data meeting the preset requirements.
As shown in fig. 3, the discriminator includes three convolution layers and one full-link layer; the filter sizes of the first convolution layer, the second convolution layer and the third convolution layer are all 5 x 5, the convolution step length is 2, and the Output1 with the depth of 48 and the size of 12 x 4 is obtained after the input quantity with the size of 24 x 8 is subjected to the convolution of the first layer; output2 with the depth of 96 and the size of 6 multiplied by 2 is obtained after the second layer convolution; output3 with depth of 192 and size of 3 × 1 is obtained after the third layer of convolution; output3 obtains a 1 x 1 Output through the fully connected layer that is used to determine whether the input data is actually present (i.e., the collected data) or the Output generated by the generator.
As shown in fig. 4, the generator includes one fully connected layer and four deconvolution layers; wherein the filter sizes of the four deconvolution layers are all 5 × 5; the input of the generator is a group of random arrays with the size of 96 multiplied by 1, and Output1 with the size of 768 multiplied by 1 is obtained through the full connection layer; in order to facilitate the subsequent deconvolution operation, Output1 is rearranged to obtain Output2 with the depth of 368 and the size of 2 × 1; obtaining Output3 with depth of 192 and size of 3 multiplied by 1 through the first deconvolution operation; obtaining Output4 with the depth of 96 and the size of 6 multiplied by 2 through the second deconvolution operation; obtaining Output5 with the depth of 48 and the size of 12 multiplied by 4 through the deconvolution operation for the third time; and after the last deconvolution operation, obtaining an output with the depth of 1 and the size of 24 multiplied by 8, wherein the output is the scene data generated by the generator. It can be seen that the generator is similar to the inverse of the arbiter, aiming to trade off the depth of the input data for the size of the output data.
All known GANs have similar and different overall structures, but the structures inside the arbiter and generator, including the selection of the activation function, and even the loss function of the whole GAN and the details of the training method, affect the learning effect of GAN. In the invention, the concept of WGAN-GP model is adopted to slightly modify GAN, and the specific modification comprises the following steps:
1) compared with the original GAN, the WGAN-GP almost completely uses the convolution layer to replace the full connection layer, and does not adopt a pooling layer, and the convolution step length is 2, so that the stability of training is increased;
in the GAN network, all the fully connected layers except in the generator and the discriminator are replaced with convolutional layers, and no pooling layer is employed.
2) No matter the input of the discriminator or the output of the generator, the normalization processing is adopted, the training stability is improved, and meanwhile, the training is accelerated, namely, in the GAN network, the normalization processing is carried out on the data output by the generator and the discriminator.
3) A LeakRelu activation function is adopted in the discriminator, and a Relu activation function is adopted in the generator;
the LeakRelu activation function is used in the discriminator instead of the Relu activation function, and the reason is to prevent gradient sparsity, namely, when the partial derivative of the model parameters is operated by a back propagation algorithm, the result is 0, the model parameters of the previous layer cannot be changed, and the Relu function is used in the generator.
4) The cross entropy (JS divergence) is no longer used to calculate the loss function, because the objective of the present invention is not to generate a scene that is just the same as the sample data (state variable data set), but rather to generate a scene in which the data distribution is close to the sample data distribution, which are essentially different, whereas the cross entropy (JS divergence) to calculate the loss function is not suitable for measuring the distance between the generated data distribution and the true data distribution.
The general structure of the GAN can know that the loss function comprises a loss function gen _ cost of a generator and a loss function disc _ cost of a discriminator, the generator and the discriminator are trained simultaneously in a single cycle, and how to balance the training strength of the generator and the discriminator depends on whether the training can be stable or not so as to obtain a better result; the present invention cites the core contents in the WGAN-GP model method: a Gradient penalty (Gradient penalty) method is used for calculating the loss function, and the specific process is as follows:
s1), D (G (Z)) is calculated, namely scene data generated by the generator is input into the discriminator, the output of the discriminator is obtained as a loss function gen _ cost of the generator, gen _ cost is used as the input of a BP algorithm to modify generator model parameters, namely gen _ cost is D (G (Z));
s2) calculating a loss function disc _ cost of the discriminator;
in order to balance the training degree of the generator and the arbiter and prevent gradient disappearance and gradient explosion, a gradient punishment method is adopted, namely, the gradient d (D (X)) of the arbiter is firstly solved, and then the relation between the gradient punishment method and the gradient limiting value K is established, so that simple loss function calculation can be realized; the gradient disappearance means that the parameters cannot obtain updated information, and the gradient explosion means that the training is unstable due to overlarge parameter updating amplitude;
however, the data dimension input by the discriminator is very high, so the WGAN-GP model only needs to sample from each batch of samples, for example, a random number θ is generated, and an interpolation algorithm is used to obtain a parameter X' ═ X θ + (1- θ) × X fake Thereby calculating D (D (X'));
the expression of the final disc _ cost is therefore:
disc_cost＝-D(G(Z))+D(X)+γ*d(D(X′))
wherein Z is random noise, D is discriminator, G (Z) is scene data, D (G (Z)) is output data of D after G (Z) is input, X is state variable data, theta is generated random number, X is output data of D after G (Z) is input, D (G (Z)) is output data of D after B (G (Z)) is input, X is output data of D after B (G (Z)) is input, theta is output data of D after B (G (Z)) is input, X is output data of D after B (G (Z)) is input, theta is output data of D after B (G) (D) is input, X is output data of D) is output data of D, where X is output data of D (G (Z) is output data of D) and is output data of D (D) which is output data of D, where D is output data of D, where Z and where D is output data of D and where Z and where X is output data of D fake For each batch of scene data, D (D (X ')) is the discriminator gradient, D (X ') is the output data of D after inputting X ', and gamma is the penalty parameter of gradient.
Inputting random noise into a generator to generate scene data, inputting the scene data and a group of state variable data into a discriminator to calculate a loss function, if the scene data is judged to not meet the preset requirement based on the loss function, modifying parameters of the discriminator and the generator by adopting a BP algorithm, regenerating the scene data by the generator, recalculating the loss function by the discriminator until the scene data meets the preset requirement, outputting the scene data meeting the preset requirement by the discriminator, and enabling the output scene data to accord with the characteristics of the group of state variable data, particularly to have the same visual characteristic and statistical characteristic.
The sample data and generated scenes are not a perfect match, as the goal of the invention is to generate new, unique scenes that capture the intrinsic features of the state variable data, rather than simple memory training data.
In order to verify the method, a time sequence of 4 physical quantities such as electricity price, total load, temperature, photovoltaic output and the like is selected for display, wherein the original data (namely, the collected data) are data of each physical quantity in a day randomly selected from a real scene, and the scene data generated by the generator is displayed according to results in the 5 th round, the 20 th round and the 40 th round.
As can be seen from fig. 5(a) - (d) -8 (a) - (d), although the data generated by the scene is not exactly the same as the original data in terms of extreme values and average values, for the physical quantities having specific variation rules in a day, such as electricity price, load power, temperature, and photovoltaic output, the variation rules of the physical quantities are closer to the original data as the number of times of training increases. Namely, the scene generation model of the experiment can better capture the daily circulation rule of physical quantities such as load, photovoltaic output and the like.
Because physical quantities such as fan output, wind speed and the like do not have strong daily cycle, the method is more suitable for observing a time sequence of a period of time, and original data of the fan output and the wind speed of 16 days and results in the 5 th, 20 th and 40 th rounds are selected for displaying. As can be seen from fig. 9(a) to (d) to 10(a) to (d), the scenes generated by the generator are relatively single in the initial training stage, and the scenes generated by the generator are more abundant as the number of times of training increases.
Comparing the original data with the 40 th round of generated data, and comparing the Cumulative Distribution Function (CDF) on the fan output and the Cumulative Distribution Function (CDF) on the photovoltaic output; as can be seen from fig. 11 and 12, the quality of the sequence generated by the generator on the fan output and the photovoltaic output is observed through the CDF function, the CDFs of the original data and the 40 th generation data are almost the same on the fan output and the photovoltaic output, and the fan distribution function similarity of the fan output is calculated to be (100-16.57)% -83.43%, and the photovoltaic output cumulative distribution function similarity is calculated to be (100-1.32)% -98.68%.
Comparing the original data with the 40 th round of generated data, and comparing the autocorrelation coefficients on the output of the fan and the autocorrelation coefficients on the output of the photovoltaic power; as can be seen from fig. 13 and 14, comparing the autocorrelation coefficients of the 40 th round of generated data with the autocorrelation coefficients of the original data, the two portions of data have very similar autocorrelation characteristics, and according to the calculation, the similarity of the fan output autocorrelation coefficients is (100-11.28)% -88.72%, and the similarity of the photovoltaic output autocorrelation coefficients is (100-19.13)% -80.87%.
Fig. 15(a) to 16(a) to (d) compare the distribution histograms of the raw data, the 5 th generation data, the 20 th generation data, and the 40 th generation data in terms of load and fan output, and indicate that the data generated by the generator is distributed closer to the raw data as the number of training times increases.
Therefore, the method avoids explicit modeling of distribution by using the training process of the GAN network, can directly generate scenes conforming to the distribution of the original data, and has stronger capabilities of capturing time behaviors, data correlation and distribution probability.
Based on the same technical scheme, the invention also discloses corresponding software of the method, and the comprehensive energy system scene generation device comprises:
an acquisition module: collecting small sample data of state variables of the comprehensive energy system;
a scene generation module: inputting small sample data and random noise into a preset GAN network, training the GAN network, and generating scene data according with the characteristics of the small sample data; in the GAN network, a generator generates scene data according to random noise, a discriminator calculates a loss function according to small sample data and the scene data, if the scene data does not meet preset requirements based on the loss function, parameters of the discriminator and the generator are modified, the generator regenerates the scene data, the discriminator recalculates the loss function until the scene data meets the preset requirements, and the discriminator outputs the scene data meeting the preset requirements.
The data processing flow of each module in the device is consistent with that of the method, and the description is not repeated here.
Based on the same technical solution, the present invention also discloses a computer readable storage medium storing one or more programs, the one or more programs including instructions, which when executed by a computing device, cause the computing device to execute an integrated energy system scene generating method.
Based on the same technical solution, the present invention also discloses a computing device comprising one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs comprise instructions for executing the integrated energy system scenario generation method.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The present invention is not limited to the above embodiments, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are included in the scope of the claims of the present invention which are filed as the application.
Claims (10)
1. An integrated energy system scenario generation method, comprising:
collecting a sample set of state variables of the comprehensive energy system;
traversing a state variable data group in a sample set, inputting a group of state variable data and random noise into a preset GAN network, training the GAN network, and generating scene data according with the characteristics of the group of state variable data; in the GAN network, a generator generates scene data according to random noise, a discriminator calculates a loss function according to state variable data and the scene data, if the scene data is judged to not meet preset requirements based on the loss function, parameters of the discriminator and the generator are modified, the generator regenerates the scene data, and the discriminator recalculates the loss function until the scene data meets the preset requirements, and the discriminator outputs the scene data meeting the preset requirements;
and generating a comprehensive energy system scene according to the scene data.
2. The method of claim 1, wherein in the GAN network, all the fully connected layers except the fully connected layers in the generator and the discriminator are replaced by convolutional layers, and no pooling layer is used.
3. The method as claimed in claim 1, wherein the generator and the discriminator output data are normalized in the GAN network.
4. The method according to claim 1, wherein a LeakRelu activation function is used in the discriminator, and a Relu activation function is used in the generator.
5. The method according to claim 1, wherein a gradient penalty method is used to calculate the loss function.
6. The method according to claim 5, wherein the loss function comprises a generator loss function and a discriminator loss function;
the loss function of the generator is:
gen_cost＝D(G(Z))
the penalty function for the discriminator is:
disc_cost＝-D(G(Z))+D(X)+γ*d(D(X′))
wherein Z is random noise, D is a discriminator, g (Z) is scene data, D (g (Z)) is output data of D after g (Z) is input, X is state variable data, and a parameter X' ═ X θ + (1- θ) × X fake θ is a generated random number, X fake For each batch of scene data, D (D (X ')) is the discriminator gradient, D (X ') is the output data of D after inputting X ', and gamma is the penalty parameter of gradient.
7. The method of claim 1, wherein the parameters of the discriminator and the generator are modified using an error back propagation algorithm.
8. An integrated energy system scenario generation apparatus, comprising:
an acquisition module: collecting a sample set of state variables of the comprehensive energy system;
a scene data generation module: traversing a state variable data group in a sample set, inputting a group of state variable data and random noise into a preset GAN network, training the GAN network, and generating scene data according with the characteristics of the group of state variable data; in the GAN network, a generator generates scene data according to random noise, a discriminator calculates a loss function according to state variable data and the scene data, if the scene data is judged to not meet preset requirements based on the loss function, parameters of the discriminator and the generator are modified, the generator regenerates the scene data, and the discriminator recalculates the loss function until the scene data meets the preset requirements, and the discriminator outputs the scene data meeting the preset requirements;
a scene generation module: and generating a comprehensive energy system scene according to the scene data.
9. A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-7.
10. A computing device, comprising:
one or more processors, one or more memories, and one or more programs stored in the one or more memories and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210376573.9A CN114862123A (en) | 2022-04-12 | 2022-04-12 | Comprehensive energy system scene generation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210376573.9A CN114862123A (en) | 2022-04-12 | 2022-04-12 | Comprehensive energy system scene generation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114862123A true CN114862123A (en) | 2022-08-05 |
Family
ID=82629477
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210376573.9A Pending CN114862123A (en) | 2022-04-12 | 2022-04-12 | Comprehensive energy system scene generation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114862123A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115310361A (en) * | 2022-08-16 | 2022-11-08 | 中国矿业大学 | Method and system for predicting underground dust concentration of coal mine based on WGAN-CNN |
-
2022
- 2022-04-12 CN CN202210376573.9A patent/CN114862123A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115310361A (en) * | 2022-08-16 | 2022-11-08 | 中国矿业大学 | Method and system for predicting underground dust concentration of coal mine based on WGAN-CNN |
| CN115310361B (en) * | 2022-08-16 | 2023-09-15 | 中国矿业大学 | Underground coal mine dust concentration prediction method and system based on WGAN-CNN |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111553587A (en) | New energy scene generation method and system based on confrontation learning model | |
| CN111079351B (en) | Power distribution network probability power flow acquisition method and device considering wind power uncertainty | |
| CN111027732B (en) | Method and system for generating multi-wind power plant output scene | |
| CN107133695A (en) | A kind of wind power forecasting method and system | |
| CN114282646B (en) | Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement | |
| CN113688869B (en) | Photovoltaic data missing reconstruction method based on generation countermeasure network | |
| CN111401755A (en) | Multi-new-energy output scene generation method, device and system based on Markov chain | |
| CN109412161A (en) | A kind of Probabilistic Load calculation method and system | |
| CN114862123A (en) | Comprehensive energy system scene generation method and device | |
| CN116702831A (en) | Hybrid short-term wind power prediction method considering massive loss of data | |
| CN111612160A (en) | Incremental Bayesian network learning method based on particle swarm optimization algorithm | |
| CN113673788A (en) | Photovoltaic power generation power prediction method based on decomposition error correction and deep learning | |
| CN112069639B (en) | Grid planning method and system for electric power system | |
| CN109840308B (en) | Regional wind power probability forecasting method and system | |
| CN111950752A (en) | Photovoltaic power station generating capacity prediction method, device and system and storage medium thereof | |
| CN115528684A (en) | Ultra-short-term load prediction method and device and electronic equipment | |
| CN114118611A (en) | Wind power prediction method and system based on prediction algorithm pool | |
| CN113947237A (en) | AMS-TCN-based wind power error correction method | |
| CN113190973A (en) | Bidirectional optimization method, device, equipment and storage medium for wind, light and load multi-stage typical scene | |
| CN111753252A (en) | Nataf transformation-based random variable sample generation method and system | |
| CN109858799B (en) | Active power distribution network reconstruction measure and line heavy load rate relevance mining method and device | |
| CN116935102B (en) | Lightweight model training method, device, equipment and medium | |
| CN117937587A (en) | Distribution optimization method and system for distributed energy | |
| CN117435308B (en) | Modelica model simulation method and system based on parallel computing algorithm | |
| CN114352485A (en) | Wind speed prediction method and device for wind driven generator and computer equipment |
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 |