Method, device and electronic equipment for determining output of generator set

Abstract

The present application discloses a method, device and electronic device for determining the output of a generator set. The method includes: obtaining historical output data of multiple generator sets including at least two generator sets that use different energy sources for power generation; clustering the historical output data to obtain multiple output data subsets; using Markov chains to model multiple output data subsets to obtain output data models, and sampling the output data models to obtain initial simulated output data; inputting the initial simulated output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model; using the pre-trained output model to output target simulated output data, so as to control the output of multiple generator sets based on the target simulated output data. The above method can improve the integrity, accuracy and reliability of the target simulated output data, thereby improving the accuracy and reliability of the prediction of the generator set output data.

Description

Technical Field

[0001] The present application relates to the field of renewable energy power generation technology, and in particular to a method, device and electronic equipment for determining the output of a generator set. Background Art

[0002] The introduction of new energy sources such as wind power and solar energy has changed the energy supply structure, and the large-scale access of new energy to the grid has also increased the uncertainty factors of the system, affecting the safety, stability and reliability of the system. Therefore, it is necessary to simulate the generation of new energy output data in order to evaluate the reliability of large-scale new energy integrated systems or control the output of generator sets.

[0003] At present, the generation of renewable energy output data mainly adopts the following two methods: (1) Modeling the power dependency structure of photovoltaic power stations based on the combination of mixed Gaussian model and Copula function. Specifically, the conditional sampling method based on Copula function is used to generate a probability power curve to identify error data, and the original data of the photovoltaic power station is cleaned. The data within the probability power curve is considered to be normal data, while the data outside the probability power curve is considered to be abnormal data. Then, the normal photovoltaic power data is analyzed based on the GMM and Copula function model, the marginal probability distribution of photovoltaic power is analyzed based on GMM, and the photovoltaic power dependency structure between power stations is analyzed based on the mixed Copula function. Finally, the Kendall rank correlation coefficient and tail correlation coefficient are used to describe the correlation between photovoltaic power. (2) A two-stage wind-solar joint output scenario generation method considering time and space correlation. Specifically, it includes two stages: in the first stage, the Copula model with HNAC structure is used to construct the output correlation between wind farms and photovoltaic power stations, as well as the output complementarity between wind power and photovoltaic power generation; in the second stage, the Copula model based on Markov chain is used to consider the time series correlation of wind power and photovoltaic output. At the same time, the fluctuation characteristics of wind power and photovoltaic power generation at different output levels are considered. Finally, a large number of wind power and photovoltaic output sequences are iteratively simulated using the Monte Carlo method, and typical scenarios of wind power and photovoltaic output are generated by clustering.

[0004] However, the above two types of data only focus on the correlation of individual wind and solar resources, ignoring the correlation between different energy sources, which will lead to insufficient and incomplete modeling. In addition, when studying the correlation patterns of wind power and photovoltaic output, there is a general lack of exploration of related dynamic changes, resulting in reduced model accuracy.

[0005] Currently, no effective solution has been proposed to the above-mentioned problem of being unable to generate complete, accurate and reliable new energy output data. Summary of the invention

[0006] The purpose of the embodiments of the present application is to provide a method, device and electronic equipment for determining the output of a generator set to solve the problem that new energy output data cannot be generated completely, accurately and reliably.

[0007] In order to solve the above technical problems, the first aspect of this specification provides a method for determining the output of a generator set, comprising:

[0008] Acquiring historical output data of a plurality of generator sets, wherein the plurality of generator sets include at least two generator sets that use different energy sources to generate electricity;

[0009] Clustering the historical output data to obtain multiple output data subsets;

[0010] Using Markov chains to model multiple output data subsets to obtain output data models, and sampling the output data models to obtain initial simulated output data;

[0011] Inputting the initial simulation output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model;

[0012] The pre-trained output model is used to output target simulated output data, so as to control the output of the plurality of generator sets based on the target simulated output data.

[0013] In some embodiments of this specification, clustering the historical output data to obtain multiple output data subsets includes:

[0014] Determine the similarity between any two pieces of data in the historical output data to form a similarity matrix;

[0015] Determining a bias parameter based on the similarity matrix;

[0016] Initialize the attraction matrix and the belonging matrix;

[0017] Based on the similarity matrix, the bias parameter and the preset damping coefficient, the attraction matrix and the attribution matrix are iteratively updated, and a plurality of cluster centers are determined based on the iteratively updated attraction matrix and attribution matrix;

[0018] Based on the multiple cluster centers, the historical output data are classified into each cluster center to obtain multiple output data subsets, each output data subset corresponding to a group of daily type output data.

[0019] In some embodiments of the present specification, after the historical output data is classified into each cluster center based on the multiple cluster centers to obtain multiple output data subsets, the method further includes:

[0020] Determining index values ​​of a plurality of preset clustering indexes based on the historical output data;

[0021] Optimizing the iteratively updated attraction matrix and attribution matrix based on the determined index values;

[0022] Based on the optimized attraction degree matrix and attribution degree matrix, multiple cluster centers are determined, and the historical output data are classified into each cluster center to obtain multiple output data subsets.

[0023] In some embodiments of the present specification, the method of using a Markov chain to model multiple output data subsets to obtain an output data model, and sampling the output data model to obtain initial simulated output data includes:

[0024] Based on the multiple output data subsets, construct an output data model represented as a state transition matrix, each element in the state transition matrix represents the probability that the day type corresponding to the output data at the previous moment is transferred to the day type corresponding to the output data at the current moment;

[0025] Divide the preset interval into equal proportions based on the state transfer matrix to obtain multiple sub-intervals and the probability of data in each sub-interval;

[0026] Based on the multiple subintervals and the probability of data in the subintervals, a plurality of pseudo-random numbers are generated, and the plurality of pseudo-random numbers are grouped to synthesize initial simulation output data of a target length.

[0027] In some embodiments of the present specification, the state transfer matrix is ​​represented by the following formula:

[0028] ;

[0029] ;

[0030] in, D Indicates the number of day types, P μv The output data of the generator set is represented by i Day type of the moment μ Transfer to i+1 Day type of the moment v The probability of H μv Indicates that the output data of the generator set is of daily type μ Japanese type v The frequency of transitions between H μd Indicates that the output data of the generator set is of daily type μ Japanese type d The frequency of transfer between .

[0031] In some embodiments of the present specification, inputting the initial simulation output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model includes:

[0032] Based on the relationship between the changes of each data in the initial simulation output data over time, dividing the initial simulation output data into static data and dynamic data;

[0033] Constructing a time series generative adversarial network including an autoencoder subnetwork and a generative adversarial subnetwork, wherein the autoencoder subnetwork includes an embedding function and a recurrence function, and the generative adversarial subnetwork includes a generator and a discriminator;

[0034] Based on the static data, the dynamic data and the corresponding loss functions of each sub-network, the parameters of the autoencoder sub-network and the generative adversarial sub-network are updated respectively to obtain the pre-trained output model.

[0035] In some embodiments of the present specification, the updating of the parameters of the autoencoder subnetwork and the generative adversarial subnetwork respectively based on the static data, the dynamic data and the corresponding loss function of each subnetwork includes:

[0036] Inputting the static data and / or the dynamic data into the embedding function and the recurrence function, and updating the parameters of the embedding function and the recurrence function based on the output results of the embedding function and the recurrence function and the encoding loss function;

[0037] Inputting the static data and / or the dynamic data into the embedded function after parameter update to obtain a first output result;

[0038] Obtaining a second output result of the generator, and jointly encoding the first output result and the second output result to obtain a jointly encoded result;

[0039] The joint encoding result is input into the discriminator, and a third output result is output; wherein the third output result is a result output by the discriminator comparing the result obtained after the joint encoding with the static data and / or the dynamic data;

[0040] Based on the generation of the adversarial loss function and / or the supervised loss function, as well as the second output result and the third output result, the parameters of the generator and the discriminator are updated.

[0041] In some embodiments of the present specification, the coding loss function is expressed by the following formula:

[0042] ;

[0043] in L er represents the coding loss, s represents static data input to the embedded function, represents the static data output by the recurrence function, x t represents the dynamic data input to the embedded function, Describe the dynamic data output by the recurrence function.

[0044] In some embodiments of the present specification, the generative adversarial loss function and the supervised loss function are expressed by the following formula:

[0045] ;

[0046] ;

[0047] in, L gd represents the generative adversarial loss, L eg represents the supervision loss, y s , y t Respectively represent the real static data and dynamic data input to the discriminator; , They represent the third output results of the discriminator corresponding to static data and dynamic data respectively, h t represents the first output result corresponding to the updated embedding function, Represents the second output result corresponding to the generator.

[0048] A second aspect of the specification provides a device for determining the output of a generator set, comprising:

[0049] A data acquisition module, used to acquire historical output data of a plurality of generator sets, wherein the plurality of generator sets include at least two generator sets that use different energy sources for power generation;

[0050] A clustering module, used for clustering the historical output data to obtain multiple output data subsets;

[0051] A data simulation module is used to model multiple output data subsets using a Markov chain to obtain an output data model, and to sample the output data model to obtain initial simulated output data;

[0052] A model updating module, used for inputting the initial simulation output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model;

[0053] An output module is used to output target simulated output data using the pre-trained output model, so as to control the output of the multiple generator sets based on the target simulated output data.

[0054] The third aspect of this specification provides an electronic device, comprising: a memory and a processor, the processor and the memory are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the steps of the method described in the first aspect.

[0055] A fourth aspect of the present specification provides a computer storage medium, wherein the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the method described in the first aspect are implemented.

[0056] The fifth aspect of this specification provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0057] The method for determining the output of a generator set in the embodiment of the present specification obtains historical output data of multiple generator sets, wherein the multiple generator sets include at least two generator sets that use different energy sources for power generation; clusters the historical output data to obtain multiple output data subsets; uses Markov chains to model multiple output data subsets to obtain output data models, and samples the output data models to obtain initial simulated output data; inputs the initial simulated output data into a time series generative adversarial network, trains the time series generative adversarial network, and obtains a pre-trained output model; uses the pre-trained output model to output target simulated output data, so as to control the output of multiple generator sets based on the target simulated output data. In the embodiment of the present specification, the correlation of the output data of the generator sets that use different energy sources for power generation can be effectively combined by clustering the historical output data and the Markov chain, and the spatiotemporal correlation of various types of generator sets can be effectively simulated; and the dynamic time correlation existing in the generated initial simulated output data can be captured by the time series generative adversarial network, thereby improving the integrity, accuracy and reliability of the target simulated output data, so as to improve the accuracy and reliability of the prediction of the generator set output data. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some implementation methods recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0059] Figure 1 Shown is a schematic diagram of a method for determining the output of a generator set provided in an embodiment of this specification;

[0060] Figure 2 Shown is a schematic diagram of a time series generative adversarial network provided by an embodiment of this specification;

[0061] Figure 3 Shown is a schematic diagram of a training process of a time series generative adversarial network provided by an embodiment of this specification;

[0062] Figure 4 Shown is a schematic diagram of a device for determining the output of a generator set provided in an embodiment of this specification;

[0063] Figure 5 Shown is a schematic diagram of an electronic device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0064] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0065] It should be noted that the user-related information and data involved in the embodiments of this specification are all information and data authorized by the user or fully authorized by relevant parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users or relevant parties to choose to authorize or refuse.

[0066] It should also be noted that in the embodiments of this specification, certain software, components, models and other existing solutions in the industry may be mentioned. They should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0067] As mentioned above, the current simulation of output data of different types of generator sets only focuses on the correlation of a single energy source, ignoring the correlation between multiple energy sources. In addition, the correlation research lacks analysis of the dynamic changes of the correlation, resulting in low integrity, accuracy and reliability of the simulation data.

[0068] In order to solve the above problems, an embodiment of the present specification provides a method for determining the output of a generator set, by obtaining historical output data of multiple generator sets, wherein the multiple generator sets include at least two generator sets that use different energy sources to generate electricity; clustering the historical output data to obtain multiple output data subsets; using Markov chains to model the multiple output data subsets to obtain an output data model, and sampling the output data model to obtain initial simulated output data; inputting the initial simulated output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model; using the pre-trained output model to output target simulated output data, so as to control the output of the multiple generator sets based on the target simulated output data.

[0069] In the embodiments of this specification, the correlation of the output data of the units using different energy sources for power generation can be effectively combined through clustering and Markov chain of historical output data, and the spatiotemporal correlation of various units can be effectively simulated; and, by generating adversarial networks through time series, the dynamic time correlation existing in the generated initial simulation output data can be captured, thereby improving the integrity, accuracy and reliability of the target simulation output data, so as to improve the accuracy and reliability of the prediction of the output data of the generator set.

[0070] It can be understood that the above method provided in the present application can be applied to electronic devices, and the electronic devices can refer to electronic devices with data calculation, processing and storage capabilities. The electronic device can be a terminal such as a PC (Personal Computer), a tablet computer, a smart phone, a wearable device, an intelligent robot, etc.; it can also be a server. Among them, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0071] The following is an introduction to the method for determining the output of a generator set provided in an embodiment of the present application in conjunction with the accompanying drawings.

[0072] Figure 1Shown is a schematic diagram of a method for determining the output of a generator set provided in an embodiment of this specification. Although this specification provides method operation steps or device structures as shown in the following embodiments or drawings, the method or device may include more or fewer operation steps or module units after partial merger based on routine or no creative labor. In steps or structures that do not logically have a necessary causal relationship, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure shown in the embodiments or drawings of this specification. When the method or module structure described is applied to an actual device, server or terminal product, it can be executed sequentially or in parallel according to the method or module structure shown in the embodiments or drawings (for example, in an environment of parallel processors or multi-threaded processing, or even in an implementation environment of distributed processing and server clusters). As Figure 1 As shown, the method may include:

[0073] S101: Acquire historical output data of a plurality of generator sets, wherein the plurality of generator sets include at least two generator sets that use different energy sources to generate electricity.

[0074] It is understood that when generating electricity, the generator set often generates electricity by converting other energy sources into electrical energy. Different types of generator sets can convert different energy sources into electrical energy. The multiple generator sets may include but are not limited to wind generator sets, photovoltaic generator sets, hydroelectric generators, etc.

[0075] S102: Clustering the historical output data to obtain multiple output data subsets.

[0076] It can be understood that clustering the historical output data can divide the output data with the same characteristics into the same output data subset, and then analyze the relationship between multiple output data subsets to obtain the spatiotemporal correlation of the output data. Different output data subsets can correspond to different day types.

[0077] Among them, the historical output data can be clustered by using hierarchical clustering, K-maens clustering, grid-based clustering algorithm, AP (Affinity Propagation clustering) algorithm, model-based clustering algorithm, etc., and this specification does not limit this. Further, in order to analyze the similarity between historical output data more highly, the embodiment of this specification uses the AP clustering algorithm to cluster the historical output data. Of course, in order to better meet the needs of data analysis in different scenarios, other clustering algorithms can also be used to cluster the historical output data.

[0078] In some embodiments of the present specification, clustering the historical output data to obtain multiple output data subsets may include: determining the similarity between any two data in the historical output data to form a similarity matrix; determining a bias parameter based on the similarity matrix; initializing an attraction matrix and an attribution matrix; iteratively updating the attraction matrix and the attribution matrix based on the similarity matrix, the bias parameter, and a preset damping coefficient, and determining multiple clustering centers based on the iteratively updated attraction matrix and attribution matrix; based on the multiple clustering centers, classifying the historical output data into each clustering center to obtain multiple output data subsets, each output data subset corresponding to a group of daily type output data.

[0079] In some embodiments of this specification, the Euclidean distance can be used to calculate the similarity between two data points. i and data points k Similarity between s ( i , k ) can be expressed by the following formula:

[0080] Formula (1)

[0081] in, x i and x k Can represent data points i and data points k Corresponding values. Furthermore, multiple similarities can form a similarity matrix S .

[0082] In some embodiments of the present specification, the similarity matrix S The diagonal elements in s ( i , i ) can be used as a bias parameter, which is generally the median of all non-diagonal element values. The damping coefficient is preset to avoid introducing oscillation into the clustering algorithm to improve the convergence of the algorithm. The damping coefficient can be between 0 and 1, that is, the damping coefficient λ (0<λ<1).

[0083] In some embodiments of the present specification, initializing the attraction matrix and the attribution matrix may be to set the attraction matrix R and the attribution matrix A The initial value of is 0.

[0084] In some embodiments of the present specification, based on the similarity matrix, the bias parameter and the preset damping coefficient, the attraction matrix and the attribution matrix are iteratively updated, and a plurality of cluster centers are determined based on the iteratively updated attraction matrix and attribution matrix, which may specifically include the following steps:

[0085] a) Update the attraction matrix R . Attraction Matrix R Any element in r t+1 ( i , k ) can be used to describe data points k Suitable as a data point i The degree of cluster center, further, for any element in the attraction matrix r t+1 ( i , k ) can be expressed by the following formula:

[0086] Formula (2)

[0087] b) Update the attribution matrix A . Attribution matrix A Any element in a t+1 ( i , k ) can be used to describe data points i Select Data Points k As the fitness of its cluster center, further, for any element in the attribution matrix a t+1 ( i , k ) can be expressed by the following formula:

[0088] Formula (3)

[0089] c) The attraction matrix is ​​adjusted according to the preset damping coefficient R and the attribution matrix A Attenuation can be performed using the following formula:

[0090] Formula (4)

[0091] d) Determine the cluster center based on the attenuation result. The new judgment rule in clustering can be expressed by the following formula:

[0092] r ( i , i )+ a( i , i )>0 formula (5)

[0093] In the embodiment of the present specification, by cyclically executing the above steps a) to d) until a preset convergence condition or number of iterations is reached, the iterative update is stopped, and multiple cluster centers can be obtained.

[0094] In some embodiments of the present specification, after the historical output data is classified into each cluster center based on the multiple cluster centers to obtain multiple output data subsets, it can also include: determining the index values ​​of multiple preset clustering indicators based on the historical output data; optimizing the iteratively updated attraction matrix and attribution matrix based on the determined index values; determining multiple cluster centers based on the optimized attraction matrix and attribution matrix, and classifying the historical output data into each cluster center to obtain multiple output data subsets.

[0095] In some embodiments of the present specification, the preset clustering index may include the average output power and output standard deviation of each type of generator set. Taking wind turbine generator sets and photovoltaic generator sets as examples, the preset clustering index may include the wind power output mean, wind power output standard deviation, photovoltaic output mean, photovoltaic output standard deviation, and the above preset clustering index may be expressed by the following formula:

[0096] Formula (6)

[0097] Formula (7)

[0098] Formula (8)

[0099] Formula (9)

[0100] in, It can represent the mean value of wind power output, P WT ( t ) can be expressed as t The wind power output at the time, It can represent the standard deviation of wind power output, It can represent the mean photovoltaic output, P PV ( t ) can be expressed as t The photovoltaic output power at the moment, It can represent the standard deviation of photovoltaic output.

[0101] Furthermore, when the index value of the above-mentioned preset clustering index is determined based on the historical output data, the above-mentioned index value can be used to optimize the attraction matrix and the attribution matrix, or the index value of the above-mentioned preset clustering index of each output data subset is calculated, and then the calculated index value is used to optimize the attraction matrix and the attribution matrix.

[0102] S103: Modeling multiple output data subsets using a Markov chain to obtain an output data model, and sampling the output data model to obtain initial simulated output data.

[0103] It can be understood that the Markov chain can be used to analyze the probability of data state transfer between different output data subsets, and then the spatiotemporal correlation between the output data of adjacent units (including adjacent wind farms and photovoltaic power stations) can be effectively combined, rather than focusing only on a single energy source, which can provide strong support for the stable operation of the power system. The output data model can be understood as a data model that characterizes the probability of data state transfer between different output data subsets, and can be expressed in the form of matrices, arrays, data tables, etc.

[0104] In some embodiments of the present specification, the output data model may be a state transfer matrix, and the establishment of a Markov chain between multiple output data subsets is achieved through the state transfer matrix.

[0105] In some embodiments of the present specification, the method of modeling multiple output data subsets using a Markov chain to obtain an output data model, and sampling the output data model to obtain initial simulated output data, may include: constructing an output data model represented by a state transfer matrix based on the multiple output data subsets, wherein each element in the state transfer matrix represents the probability of the day type corresponding to the output data at the previous moment transferring to the day type corresponding to the output data at the current moment; dividing a preset interval in equal proportion based on the state transfer matrix to obtain multiple sub-intervals and the probability of data in each sub-interval; generating multiple pseudo-random numbers based on the multiple sub-intervals and the probability of data in each sub-interval, and synthesizing the multiple pseudo-random numbers into initial simulated output data of a target length.

[0106] In some embodiments of the present specification, the state transfer matrix can be expressed by the following formula:

[0107] Formula (10)

[0108] Formula (11)

[0109] Among them, the state transfer matrix P Can be expressed as a D * D The matrix ofD Indicates the number of day types, P μv The output data of the generator set is represented by i Day type of the moment μ Transfer to i+1 Day type of the moment v The probability of i The unit output day type of the day is μ , then i+ The unit output day type for one day is v The probability of P μv . H μv Indicates that the output data of the generator set is of daily type μ Japanese type v The frequency of transitions between H μd Indicates that the output data of the generator set is of daily type μ Japanese type d The frequency of transfer between .

[0110] In some embodiments of the present specification, after determining the state transfer matrix, the initial simulation output data of the target length can be obtained by continuous sampling. Specifically, the sample space and probability distribution, that is, the state transfer matrix obtained by the above calculation, can be determined first. Then, the [0,1] interval is divided in equal proportion according to the probability of the value represented by each element in the state transfer matrix, and then a pseudo-random number between [0,1] is generated by a computer based on the division result, and a sampling can be completed according to the landing point of the pseudo-random number. After that, the generation of random numbers is repeated until the data length reaches the target length, and the multiple random numbers obtained are combined to obtain the initial simulation output data of the target length.

[0111] S104: Inputting the initial simulation output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model.

[0112] It can be understood that in addition to the spatiotemporal correlation between the output data of different types of units, there is also dynamic time correlation in the time series. The dynamic time correlation in the time series can be captured by the time series generative adversarial network (Time GAN, Time Generative Adversarial Networks), so as to realize the analysis and capture of the dynamic correlation between the output data of different types of units.

[0113] S105: Outputting target simulation output data using the pre-trained output model, so as to control the output of the plurality of generator sets based on the target simulation output data.

[0114] It is understood that the target simulation output data may be a combined output sequence of multiple generator sets, which may include simulation output data of each generator set. The generator sets may be controlled based on the target simulation output data to achieve optimal scheduling or optimal planning of the power system.

[0115] The following describes the model training process and simulation output process in the above step S103 and step S104 in conjunction with the accompanying drawings. It can be understood that in order to fully analyze the dynamic correlation of the data in time, and considering that some data will change with time and some data will not change with time, the initial simulation output data can be divided into dynamic data (i.e., dynamic features) and static data (i.e., static features).

[0116] refer to Figure 2 The structure of the time series generative adversarial network shown in Figure 3 The training process of the network shown in , before training the time series generative adversarial network, the vector space and the training sample set can be defined: Definition S For the vector space representing static features, we have S ∈ S ;definition χ For the vector space representing dynamic features, we have X ∈ χ Assume ( S , X 1:T ) obeys some joint distribution P , time series length T is a random variable. Sample X contains n eigenvectors x , n ∈{1,…, N}. The training sample set is represented as .

[0117] In some embodiments of the present specification, the inputting the initial simulation output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model may include: dividing the initial simulation output data into static data and dynamic data based on the relationship between the changes of each data in the initial simulation output data over time; constructing a time series generative adversarial network including an autoencoder subnetwork and a generative adversarial subnetwork, wherein the autoencoder subnetwork includes an embedding function and a reproduction function, and the generative adversarial subnetwork includes a generator and a discriminator; based on the static data, the dynamic data, and the corresponding loss functions of each subnetwork, respectively updating the parameters of the autoencoder subnetwork and the generative adversarial subnetwork to obtain the pre-trained output model.

[0118] In some embodiments of this specification, the structure of the time series generation adversarial network is as follows: Figure 2 As shown, it can include two networks: an autoencoder network (embedding function and reproduction function) and an adversarial network (generator and discriminator). A joint training method can be adopted to train the autoencoder network (i.e., the autoencoder subnetwork mentioned above) and the adversarial network (i.e., the generative adversarial network mentioned above) by relying on at least two different loss functions.

[0119] Specifically, for the autoencoder network, the embedding function can be used to obtain the mapping between data features and latent space and reduce the dimension of data features. When mapping dynamic features, assuming that the dynamic features at the next moment are related to the dynamic features at the previous moment, the embedding function can be expressed as:

[0120] Formula (12)

[0121] in, s It can represent high-dimensional static feature vectors, that is, static data in the initial simulated output data of the input value time series generation adversarial network. h s It can represent the low-dimensional static feature vector after being mapped by the embedding function; h t It can be expressed as t The low-dimensional static feature vector at each moment, x t Can be expressed t The high-dimensional dynamic feature vector at the moment, that is, the first in the initial simulated output data of the input value time series generation adversarial network t The dynamic data at each moment can be realized through the recursive network. The function in the above formula (12) e .

[0122] The recurrence function can be understood as the inverse process corresponding to the encoding process of the embedding function. Through the recurrence function, we can try to reconstruct the low-dimensional feature vector back to the original high-dimensional dynamic or static features. The recurrence function can be expressed as:

[0123] Formula (13)

[0124] in, It can represent the restored high-dimensional static feature vector, that is, the static feature vector reconstructed and output by the recurrence function. It can be expressed as the first t The high-dimensional dynamic feature vector at the moment, that is, the first reconstructed output after the complex function output t The dynamic feature vector at the moment. The function can be implemented through a bidirectional recurrent network with a feedforward output layer r .

[0125] It can be understood that the latent space is a high-dimensional matrix that constructs the relationship between external variables (input data) and simulated variables (simulated output data) in the autoencoder network. The latent space is used to describe how various external variables affect the simulated variables. For example, wind power output is affected by various meteorological factors. Since it is not clear how the various factors specifically interact with the output results, the latent space can be constructed by dimensionality reduction to compress all variables as much as possible to screen out factors related to wind power output, thereby establishing a mapping relationship between the latent space and the reproduced results in the reproduction function. The reproduction function can be used to determine the error between the real input sequence and the sequence reconstructed by the reproduction function, so that the generator and discriminator in the subsequent generative adversarial network can be trained based on the error.

[0126] In some embodiments of this specification, in order to enable the embedding function and the recurrence function to accurately construct the low-dimensional feature space of the wind and solar output data and reconstruct the high-dimensional feature space, the encoding loss function needs to consider the loss of both parts at the same time, and the network parameters of the recurrence function are adjusted during network training. θ r and embedded network parameters θ e Furthermore, since the data features are divided into static feature vectors and dynamic feature vectors during self-encoding, the loss corresponding to the static feature vector and the loss corresponding to the dynamic feature vector need to be considered in the encoding loss. Specifically, the encoding loss function can be expressed by the following formula:

[0127] Formula (14)

[0128] in L er represents the coding loss, srepresents static data input to the embedded function, represents the static data output by the recurrence function, x t represents the dynamic data input to the embedded function, Describe the dynamic data output by the recurrence function.

[0129] refer to Figure 2 and Figure 3 As shown, a generative adversarial network may include a generator and a discriminator, wherein the generator is used to simulate generated data, and the discriminator is used to determine whether the data generated by the generator is true data or false data based on the data input to the time series generative adversarial network. True data can be understood as data that conforms to the data distribution input to the time series generative adversarial network, and false data can be understood as data that does not conform to the data distribution input to the time series generative adversarial network. In the embodiments of this specification, it is necessary to jointly encode the output result of the generator and the output result of the embedding function to obtain a synthesized output (i.e., the joint encoding result below), and the synthesized output is input to the discriminator. The discriminator determines whether the synthesized output is true data or false data based on the input data input to the time series generative adversarial network to obtain a classification result. Further, it can be defined Z S and Z χ Respectively represent static and dynamic random vector spaces with known distributions, z S and z 1:T Random sampling belongs to two vector spaces respectively. The generator can be expressed by the following formula:

[0130] Formula (15)

[0131] in, can represent the static feature vector vector generated by the generator, , can be respectively expressed as t Moment and t-1 The dynamic feature vector at the moment. The function can be implemented through a recursive network g .

[0132] The output of the embedding function and the output of the generator are encoded in the latent space to obtain a synthetic output. The discriminator can compare the synthetic output with the real wind and solar power generation time series (the data input to the time series generative adversarial network) to determine whether the synthetic output is real power generation data. The discriminator classification result of False indicates false data, and the output of True indicates true data. The discriminator can be expressed by the following formula:

[0133] Formula) (16)

[0134] in, , can represent the static feature vector and the first t The classification result of the dynamic feature vector at the moment, It can represent synthetic output, and can be implemented as a function through a bidirectional recurrent network with a feedforward output layer d .

[0135] It can be understood that the generator is part of the generative adversarial network. The output of the generator is essentially a noise sequence, which is used to represent the error. The mapping from the latent space to the real sequence cannot be perfect, and there will be some errors. If the error can be simulated by the generator, the output time series can be close to the real data. Therefore, by synthesizing and then inputting it into the discriminator, it can be tested whether the generated noise sequence can fit the error.

[0136] In some embodiments of this specification, a joint loss function of the generator and the discriminator of the generative adversarial network (hereinafter referred to as the generative adversarial loss function) can be set to achieve joint optimization of the generator and the discriminator, thereby improving the accuracy of the discriminator classification result. The generative adversarial loss function can be expressed by the following formula:

[0137] Formula (17)

[0138] in, L gd can be expressed as generating adversarial loss, L eg can represent the supervision loss, y s , y t can represent the real static data and dynamic data input to the discriminator respectively; , The output results of the discriminator corresponding to static data and dynamic data can be represented respectively.

[0139] In some embodiments of the present specification, the generative adversarial loss is an unsupervised loss including binary adversarial feedback. Relying solely on the generative adversarial loss is not enough to motivate the generator to capture the step-by-step dynamic conditional distribution in the data, such as the step-by-step dynamic conditional distribution in the building air conditioning load data, so an additional supervised loss function can be introduced, and the generator can accept alternating training of the generative adversarial loss function and the supervised loss function. The supervised loss function can be expressed by the following formula:

[0140] Formula (18)

[0141] in, h t It can represent the output result corresponding to the updated embedding function, It can represent the second output result corresponding to the generator.

[0142] Reference Figure 3 As shown, in some embodiments of the present specification, the updating of the parameters of the self-encoding subnetwork and the generative adversarial subnetwork based on the static data, the dynamic data and the corresponding loss function of each subnetwork may include: inputting the static data and / or the dynamic data into the embedding function and the recurrence function, and updating the parameters of the embedding function and the recurrence function based on the output results of the embedding function and the recurrence function and the encoding loss function; inputting the static data and / or the dynamic data into the embedding function after parameter update to obtain a first output result; obtaining the second output result of the generator, and jointly encoding the first output result and the second output result to obtain a joint encoding result; inputting the joint encoding result into the discriminator to output a third output result; wherein the third output result is the result of the discriminator comparing the result obtained after the joint encoding with the static data and / or the dynamic data; based on the generative adversarial loss function and / or the supervision loss function, and the second output result and the third output result, the parameters of the generator and the discriminator are updated. It can be understood that the solid line in the figure can represent the forward transmission of data, and the dotted line can represent the feedback of the gradient. in, θ e , θ r , θ g , θ d They can represent the parameters of the embedding function, recurrent function, generator, and discriminator respectively.

[0143] The method for determining the output of the generator set in the embodiment of this specification effectively combines the temporal and spatial correlation between the power generation of adjacent generator sets through clustering methods and Markov chains, which can better predict and respond to the power generation of the generator sets in practical applications, and provide strong support for the stable operation of the power system. In addition, in response to the dynamic changes in temporal and spatial correlation, the Time GAN model that combines unsupervised training with supervised training can more accurately capture and simulate the dynamic correlation in the time series, thereby greatly improving the accuracy and reliability of predicting the power generation of multiple generator sets.

[0144] Based on the above-mentioned method for determining the output of a generator set, one or more embodiments of this specification also provide a device for determining the output of a generator set. The device may include a device (including a distributed system), software (application), module, plug-in, server, client, etc. that uses the method described in the embodiments of this specification and is combined with necessary implementation hardware. Based on the same innovative concept, the device in one or more embodiments provided in the embodiments of this specification is as described in the following embodiments. Since the implementation scheme and method for solving the problem of the device are similar, the implementation of the specific device in the embodiments of this specification can refer to the implementation of the aforementioned method, and the repetitions will not be repeated. As used below, the term "unit" or "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived. Figure 4 FIG. 1 is a schematic diagram of a device for determining the output of a generator set provided in an embodiment of this specification. Figure 4 As shown, the generator set output determination device 400 may include:

[0145] The data acquisition module 401 is used to acquire historical output data of a plurality of generator sets, wherein the plurality of generator sets include at least two generator sets that use different energy sources for power generation.

[0146] The clustering module 402 is used to cluster the historical output data to obtain multiple output data subsets.

[0147] The data simulation module 403 is used to use a Markov chain to model multiple output data subsets to obtain an output data model, and to sample the output data model to obtain initial simulated output data.

[0148] The model updating module 404 is used to input the initial simulation output data into the time series generative adversarial network, train the time series generative adversarial network, and obtain a pre-trained output model.

[0149] The output module 405 is used to output target simulation output data using the pre-trained output model, so as to control the output of the multiple generator sets based on the target simulation output data.

[0150] In some embodiments of the present specification, the clustering module 402 can be specifically used to: determine the similarity between each pair of data in the historical output data to form a similarity matrix; determine the bias parameter based on the similarity matrix; initialize the attraction matrix and the attribution matrix; iteratively update the attraction matrix and the attribution matrix based on the similarity matrix, the bias parameter and the preset damping coefficient, and determine multiple clustering centers based on the iteratively updated attraction matrix and attribution matrix; based on the multiple clustering centers, classify the historical output data into each clustering center to obtain multiple output data subsets, each output data subset corresponding to a group of daily type output data.

[0151] In some embodiments of the present specification, the clustering module 402 can also be used to: determine the index values ​​of multiple preset clustering indicators based on the historical output data; optimize the iteratively updated attraction matrix and attribution matrix based on the determined index values; determine multiple clustering centers based on the optimized attraction matrix and attribution matrix, and classify the historical output data into each clustering center to obtain multiple output data subsets.

[0152] In some embodiments of the present specification, the data simulation module 403 can be specifically used to: construct an output data model represented by a state transfer matrix based on the multiple output data subsets, each element in the state transfer matrix represents the probability of the day type corresponding to the output data at the previous moment transferring to the day type corresponding to the output data at the current moment; divide the preset interval into equal proportions based on the state transfer matrix to obtain multiple sub-intervals and the probability of data in each sub-interval; generate multiple pseudo-random numbers based on the multiple sub-intervals and the probability of data in each sub-interval, and synthesize the multiple pseudo-random numbers into initial simulation output data of the target length.

[0153] In some embodiments of the present specification, the state transfer matrix can be expressed by the following formula:

[0154] ;

[0155] ;

[0156] in, D Indicates the number of day types, P μv The output data of the generator set is represented by i Day type of the moment μ Transfer to i+1 Day type of the moment v The probability of H μv Indicates that the output data of the generator set is of daily type μ Japanese typev The frequency of transitions between H μd Indicates that the output data of the generator set is of daily type μ Japanese type d The frequency of transfer between .

[0157] In some embodiments of the present specification, the model update module 404 can be specifically used to: divide the initial simulation output data into static data and dynamic data based on the relationship between the changes of each data in the initial simulation output data over time; construct a time series generative adversarial network including an autoencoder subnetwork and a generative adversarial subnetwork, wherein the autoencoder subnetwork includes an embedding function and a reproduction function, and the generative adversarial subnetwork includes a generator and a discriminator; based on the static data, the dynamic data and the corresponding loss functions of each subnetwork, the parameters of the autoencoder subnetwork and the generative adversarial subnetwork are updated respectively to obtain the pre-trained output model.

[0158] In some embodiments of the present specification, the model updating module 404 updates the parameters of the autoencoder subnetwork and the generative adversarial subnetwork respectively based on the static data, the dynamic data and the corresponding loss functions of each subnetwork, and can be specifically used to: input the static data and / or the dynamic data into the embedding function and the reproduction function, and update the parameters of the embedding function and the reproduction function based on the output results of the embedding function and the reproduction function and the encoding loss function; input the static data and / or the dynamic data into the embedding function after parameter update to obtain a first output result; obtain the second output result of the generator, and jointly encode the first output result and the second output result to obtain a joint encoding result; input the joint encoding result into the discriminator to output a third output result; wherein the third output result is the result of the discriminator comparing the result obtained after the joint encoding with the static data and / or the dynamic data; based on the generative adversarial loss function and / or the supervision loss function, as well as the second output result and the third output result, update the parameters of the generator and the discriminator.

[0159] In some embodiments of the present specification, the encoding loss function can be expressed by the following formula:

[0160] ;

[0161] in L er represents the coding loss, s represents static data input to the embedded function, represents the static data output by the recurrence function, xt represents the dynamic data input to the embedded function, Describe the dynamic data output by the recurrence function.

[0162] In some embodiments of the present specification, the generative adversarial loss function and the supervised loss function may be expressed by the following formula:

[0163] ;

[0164] ;

[0165] in, L gd represents the generative adversarial loss, L eg represents the supervision loss, y s , y t Respectively represent the real static data and dynamic data input to the discriminator; , They represent the third output results of the discriminator corresponding to static data and dynamic data respectively, h t represents the first output result corresponding to the updated embedding function, Represents the second output result corresponding to the generator.

[0166] The description and functions of the above modules can be understood by referring to the method for determining the output of the generator set, which will not be elaborated here.

[0167] The present application also provides an electronic device, such as Figure 5 As shown, the electronic device may include a processor 501 and a memory 502, wherein the processor 501 and the memory 502 may be connected via a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0168] The processor 501 may be a central processing unit (CPU). The processor 501 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0169] The memory 502 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the method for determining the output of a generator set in an embodiment of the present invention (for example, Figure 4 The processor 501 executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory 502, that is, the method for determining the output of the generator set in the above method embodiment is implemented.

[0170] The memory 502 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required by at least one function; the data storage area may store data created by the processor 501, etc. In addition, the memory 502 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 502 may optionally include a memory remotely arranged relative to the processor 501, and these remote memories may be connected to the processor 501 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0171] The one or more modules are stored in the memory 502, and when executed by the processor 501, the following method for determining the output of a generator set is performed:

[0172] The method comprises the steps of obtaining historical output data of a plurality of generator sets, wherein the plurality of generator sets include at least two generator sets that generate electricity using different energy sources; clustering the historical output data to obtain a plurality of output data subsets; modeling the plurality of output data subsets using a Markov chain to obtain an output data model, and sampling the output data model to obtain initial simulated output data; inputting the initial simulated output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model; and outputting target simulated output data using the pre-trained output model to control the output of the plurality of generator sets based on the target simulated output data.

[0173] The specific details of the above electronic device can be understood by referring to the corresponding descriptions and effects in the above method embodiments, and will not be repeated here.

[0174] The present specification also provides a computer storage medium, wherein the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the method for determining the output of the above-mentioned generator set are implemented.

[0175] Those skilled in the art can understand that the implementation of all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0176] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0177] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions.

[0178] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0179] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute certain parts of the methods of each implementation method of the present application.

[0180] The present application can be used in many general or special computer system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0181] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0182] Although the present application has been described through embodiments, those skilled in the art will appreciate that there are many modifications and variations to the present application without departing from the spirit of the present application, and it is intended that the appended claims include these modifications and variations without departing from the spirit of the present application.

Claims

1. A method for determining the output of a generator set, characterized in that: Acquiring historical output data of a plurality of generator sets, wherein the plurality of generator sets include at least two generator sets that use different energy sources to generate electricity; Clustering the historical output data to obtain multiple output data subsets; Using Markov chains to model multiple output data subsets to obtain output data models, and sampling the output data models to obtain initial simulated output data; Inputting the initial simulation output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model; Outputting target simulated output data using the pre-trained output model, so as to control the output of the plurality of generator sets based on the target simulated output data; The step of inputting the initial simulation output data into a time series generative adversarial network and training the time series generative adversarial network to obtain a pre-trained output model comprises: Based on the relationship between the changes of each data in the initial simulation output data over time, the initial simulation output data is divided into static data and dynamic data; a time series generative adversarial network including an autoencoder subnetwork and a generative adversarial subnetwork is constructed, wherein the autoencoder subnetwork includes an embedding function and a recurrence function, and the generative adversarial subnetwork includes a generator and a discriminator; the static data and / or the dynamic data are input into the embedding function and the recurrence function, and the parameters of the embedding function and the recurrence function are updated based on the output results of the embedding function and the recurrence function and the encoding loss function; the static data and / or the dynamic data are input into the embedding function and the recurrence function The embedding function after parameter update obtains a first output result; obtains a second output result of the generator, and jointly encodes the first output result and the second output result to obtain a joint encoding result; inputs the joint encoding result into the discriminator, and outputs a third output result; wherein the third output result is the result of the discriminator comparing the result obtained after joint encoding with the static data and / or the dynamic data; based on the generation of adversarial loss function and / or supervision loss function, as well as the second output result and the third output result, the parameters of the generator and the discriminator are updated to obtain the pre-trained output model; The encoding loss function is expressed by the following formula: ； in L er represents the coding loss, s represents static data input to the embedded function, represents the static data output by the recurrence function, x t represents the dynamic data input to the embedded function, Expressing dynamic data output by the recurrence function; The generative adversarial loss function and the supervised loss function are expressed by the following formula: ； ； in, L gd represents the generative adversarial loss, L eg represents the supervision loss, y s , y t Respectively represent the real static data and dynamic data input to the discriminator; , They represent the third output results of the discriminator corresponding to static data and dynamic data respectively, h t represents the first output result corresponding to the updated embedding function, Represents the second output result corresponding to the generator.

2. The method for determining the output of a generator set according to claim 1, characterized in that: The clustering of the historical output data to obtain multiple output data subsets includes: Determine the similarity between any two pieces of data in the historical output data to form a similarity matrix; Determining a bias parameter based on the similarity matrix; Initialize the attraction matrix and the belonging matrix; Based on the similarity matrix, the bias parameter and the preset damping coefficient, the attraction matrix and the attribution matrix are iteratively updated, and a plurality of cluster centers are determined based on the iteratively updated attraction matrix and attribution matrix; Based on the multiple cluster centers, the historical output data are classified into each cluster center to obtain multiple output data subsets, each output data subset corresponding to a group of daily type output data.

3. The method for determining the output of a generator set according to claim 2, characterized in that: After the historical output data is classified into each cluster center based on the multiple cluster centers to obtain multiple output data subsets, the method further includes: Determining index values ​​of a plurality of preset clustering indexes based on the historical output data; Optimizing the iteratively updated attraction matrix and attribution matrix based on the determined index values; Based on the optimized attraction degree matrix and attribution degree matrix, multiple cluster centers are determined, and the historical output data are classified into each cluster center to obtain multiple output data subsets.

4. The method for determining the output of a generator set according to claim 1, characterized in that: The method of using the Markov chain to model multiple output data subsets to obtain an output data model, and sampling the output data model to obtain initial simulated output data includes: Based on the multiple output data subsets, construct an output data model represented as a state transition matrix, each element in the state transition matrix represents the probability that the day type corresponding to the output data at the previous moment is transferred to the day type corresponding to the output data at the current moment; Divide the preset interval into equal proportions based on the state transfer matrix to obtain multiple sub-intervals and the probability of data in each sub-interval; Based on the multiple subintervals and the probability of data in the subintervals, a plurality of pseudo-random numbers are generated, and the plurality of pseudo-random numbers are grouped to synthesize initial simulation output data of a target length.

5. The method for determining the output of a generator set according to claim 4, characterized in that: The state transfer matrix is ​​expressed by the following formula: ； ； in, D Indicates the number of day types, P μv The output data of the generator set is represented by i Day type of the moment μ Transfer to i+1 Day type of the moment v The probability of H μv Indicates that the output data of the generator set is of daily type μ Japanese Type v The frequency of transitions between H μd Indicates that the output data of the generator set is of daily type μ Japanese type d The frequency of transfer between .

6. A device for determining the output of a generator set, characterized in that: include: A data acquisition module, used to acquire historical output data of a plurality of generator sets, wherein the plurality of generator sets include at least two generator sets that use different energy sources for power generation; A clustering module, used for clustering the historical output data to obtain multiple output data subsets; A data simulation module is used to model multiple output data subsets using a Markov chain to obtain an output data model, and to sample the output data model to obtain initial simulated output data; A model updating module, used for inputting the initial simulation output data into a time series generative adversarial network, training the time series generative adversarial network, and obtaining a pre-trained output model; an output module, configured to output target simulated output data using the pre-trained output model, so as to control the output of the plurality of generator sets based on the target simulated output data; The model updating module is specifically used for: dividing the initial simulation output data into static data and dynamic data based on the relationship between the changes of each data in the initial simulation output data over time; constructing a time series generative adversarial network including an autoencoder subnetwork and a generative adversarial subnetwork, wherein the autoencoder subnetwork includes an embedding function and a recurrence function, and the generative adversarial subnetwork includes a generator and a discriminator; inputting the static data and / or the dynamic data into the embedding function and the recurrence function, and updating the parameters of the embedding function and the recurrence function based on the output results of the embedding function and the recurrence function and the encoding loss function; inputting the static data and / or the dynamic data into the embedding function and the recurrence function, and updating the parameters of the embedding function and the recurrence function based on the output results of the embedding function and the recurrence function and the encoding loss function; The embedding function after the dynamic data input parameter update is input to obtain a first output result; the second output result of the generator is obtained, and the first output result and the second output result are jointly encoded to obtain a joint encoding result; the joint encoding result is input to the discriminator, and a third output result is output; wherein the third output result is the result of the discriminator comparing the result obtained after the joint encoding with the static data and / or the dynamic data; based on the generation of adversarial loss function and / or supervision loss function, as well as the second output result and the third output result, the parameters of the generator and the discriminator are updated to obtain the pre-trained output model; The encoding loss function is expressed by the following formula: ； in L er represents the coding loss, s represents static data input to the embedded function, represents the static data output by the recurrence function, x t represents the dynamic data input to the embedded function, Expressing dynamic data output by the recurrence function; The generative adversarial loss function and the supervised loss function are expressed by the following formula: ； ； in, L gd represents the generative adversarial loss, L eg represents the supervision loss, y s , y t Respectively represent the real static data and dynamic data input to the discriminator; , They represent the third output results of the discriminator corresponding to static data and dynamic data respectively, h t represents the first output result corresponding to the updated embedding function, Represents the second output result corresponding to the generator.

7. An electronic device, characterized in that: include: A memory and a processor, wherein the processor and the memory are communicatively connected to each other, the memory stores computer instructions, and the processor implements the steps of the method according to any one of claims 1 to 5 by executing the computer instructions.

8. A computer storage medium, characterized in that The computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer program product, characterized in that The invention comprises a computer program, which implements the steps of the method according to any one of claims 1 to 5 when being executed by a processor.

Images