Source load prediction model processing method based on artificial neural network

Abstract

The invention discloses a source load prediction model processing method based on an artificial neural network, and the method comprises the steps: training a WGAN based on a historical operation scene, and enabling a generator to convert a random vector obtained through sampling from known distribution into any real scene with the statistical distribution consistent with the historical operation scene of a system; therefore, unsupervised representation of statistical distribution of the system operation scene is established to complete generation of the system operation scene; and obtaining an observation value and a point prediction value of the uncertain variable, constructing a scene prediction optimization problem based on the trained WGAN and the observation value and the point prediction value of the uncertain variable, thereby obtaining an optimal input vector of a generator, executing scene prediction, and obtaining a group of different system operation scenes at future moments. The method effectively solves the problem that a traditional prediction model is insufficient in prediction precision when processing high-dimensional and nonlinear data, and is suitable for accurate prediction of renewable energy power generation and electrical load in an electricity market environment.

Description

Technical Field

[0001] The invention relates to the technical field of power systems, and in particular to a source-load prediction model processing method based on an artificial neural network. Background Art

[0002] With the large-scale access of renewable energy and the deepening of power market reform, the power system is facing the problem of randomness on both the supply and demand sides. In the current power system, the power generation and power load of renewable energy (such as wind power and solar power) are highly volatile and uncertain.

[0003] Traditional forecasting methods, such as time series analysis and regression analysis, can reflect the changing trend of source load to a certain extent, but they have obvious shortcomings when dealing with large-scale, nonlinear, and high-dimensional data. Specifically, they are manifested in the following aspects: Nonlinear relationships are difficult to capture: Renewable energy generation and electricity load are affected by many factors, and there are often complex nonlinear relationships between these factors. Traditional methods are difficult to accurately capture these relationships. The data dimension is high and complex: Source load forecasting needs to consider multiple data dimensions such as meteorological factors (such as wind speed, light intensity, temperature, etc.), time factors (such as seasons, holidays, working days, etc.), and the operating status of the power system itself. Traditional methods are not effective when dealing with high-dimensional data. Limited prediction accuracy: Due to the above reasons, when faced with large-scale and highly volatile source load data, the prediction accuracy of traditional methods is often limited, and it is difficult to meet the power system's demand for high-precision predictions.

[0004] Therefore, it is particularly important to develop a source load prediction model that can effectively handle complex data relationships and improve prediction accuracy. Summary of the invention

[0005] In view of the problems existing in the prior art when dealing with large-scale, highly volatile renewable energy power generation and electricity load forecasting, such as insufficient prediction accuracy, difficulty in capturing nonlinear relationships, and poor high-dimensional data processing, the present invention provides a source-load prediction model processing method based on an artificial neural network, which predicts a group of high-quality system operation scenarios at each moment, thereby achieving an accurate characterization of the uncertainty of system operation in the future period.

[0006] The following is the technical solution of the present invention.

[0007] The source load prediction model processing method based on artificial neural network includes the following steps: Based on historical operation scenarios, WGAN is trained so that the generator can convert random vectors sampled from a known distribution into any real scenario with a statistical distribution consistent with the system's historical operation scenarios, thereby establishing an unsupervised representation of the statistical distribution of system operation scenarios to complete system operation scenario generation; Obtain the observed values ​​and point prediction values ​​of uncertain variables, construct a scenario prediction optimization problem based on the trained WGAN and the observed values ​​and point prediction values ​​of uncertain variables, so as to obtain the optimal input vector of the generator, perform scenario prediction, and obtain a set of system operation scenarios at different future moments.

[0008] Preferably, the WGAN is trained based on historical operation scenarios, so that the generator can convert random vectors sampled from a known distribution into any real scenario with a statistical distribution consistent with the historical operation scenarios of the system, thereby establishing an unsupervised representation of the statistical distribution of the system operation scenarios to complete the generation of system operation scenarios, including: Define the variables involved in system operation, and define the complete operation scenario for capturing the uncertainty of system operation by multi-dimensional uncertain variables; Construct a WGAN model, including two neural networks, the generator and the discriminator. Take the Wasserstein distance as the training target, and train the generator and the discriminator in alternating competition through the minimum-maximum game. During the training process, the parameters of the generator and the discriminator are continuously updated by optimizing the loss function. When the discriminator can no longer distinguish between real and generated scenarios, the WGAN training is considered to have reached an equilibrium state. The trained WGAN generator is used to sample random vectors from a simple known distribution as needed and convert them into system operation scenarios.

[0009] Preferably, the variables involved in the definition of the system operation are defined by multi-dimensional uncertain variables to capture a complete operation scenario of the system operation uncertainty, including: The variables involved in the operation of the system at time t are t It is expressed as: Where P t PV , P t WT , P t CHP , P t EB and P t load Respectively represent the electric power of PV, WT, CHP, EB and power load at the tth moment; P t dis and P t ch They represent the discharging and charging power of ESS at the tth moment respectively; P t grid is the electric power transmitted between the system and the main grid at the tth moment; and denote the thermal power of CHP, GB, EB and thermal load at the tth moment respectively; and They represent the heat release and heat charging power of TSS at the tth moment respectively; Ω t All variables except P t PV , P t WT , P t load and All other variables are controllable; The system operation scenario is represented as a multidimensional time series, where each dimension corresponds to a variable. The concept of channel is introduced, and the multidimensional time series is jointly represented as a whole with multiple channels. This results in a complete operation scenario for capturing the uncertainty of system operation, which is convenient for feature extraction.

[0010] Preferably, the WGAN model is constructed, including two neural networks, a generator and a discriminator, with Wasserstein distance as the training target, and the generator and the discriminator are alternately trained in competition through the minimum-maximum game. During the training process, the parameters of the generator and the discriminator are continuously updated by optimizing the loss function, including: The training objective of WGAN is as follows: Where G and D represent the generator and discriminator respectively; x and P x Represent the system's historical operating scenarios and corresponding statistical distributions respectively; z is the input implicit vector of the generator and z~P z ;P z Let P be a simple known distribution that is easy to sample. z Set to a uniform distribution of [-1,1]; The generator and the discriminator compete through a min-max game, and the generator is trained to deceive the discriminator into treating the generated scenes as real scenes, where the output of the generator is represented as G(z) and G(z)~P G , here P G Represents the statistical distribution of the generated scene, and the discriminator and the generator are trained alternately to distinguish P x and P G , where the output of the discriminator is an indication of whether the scene is real or fake, and the larger the output value, the more real the scene is from the perspective of the discriminator. The loss functions of the generator and the discriminator can be defined as: Where E represents the empirical mean during the training process; The alternating competitive training of the discriminator and the generator makes them evolve continuously until the discriminator can no longer distinguish the scenes generated by the generator from the real scenes, which also means that P x and P G The Wasserstein distance between them converges to a minimum value as the game between the generator and the discriminator reaches equilibrium. At this point, it can be considered that the generator has learned the true statistical distribution of the system operation scenario.

[0011] Preferably, when the discriminator can no longer distinguish between the real scene and the generated scene, it is considered that the WGAN training has reached an equilibrium state, and the trained WGAN generator is used to sample random vectors from a simple known distribution as needed and convert them into system operation scenes, including: When the parameters of the discriminator are given, P x and P G The Wasserstein distance between is expressed as: where θ (D) represents the parameters of the discriminator. When the discriminator can no longer distinguish between real scenes and generated scenes, its parameters θ (G) are fixed and used to generate sufficiently realistic system operation scenarios.

[0012] Preferably, the acquisition of the observed values ​​and point prediction values ​​of the uncertain variables, the construction of a scenario prediction optimization problem based on the trained WGAN and the observed values ​​and point prediction values ​​of the uncertain variables, thereby obtaining the optimal input vector of the generator, performing scenario prediction, and obtaining a set of system operation scenarios at different future moments, including: Obtain the observed values ​​and point prediction values ​​of uncertain variables, the domain of random latent vectors sampled from uniform distribution, and determine the input vector space of the generator; Construct the objective function of the optimization problem, which includes two parts: one part measures the closeness between the prediction scenario and the observed value, and the other part is used as a penalty term to ensure the authenticity of the prediction scenario; construct constraints to ensure that the prediction scenario does not conflict with the point prediction information, that is, the prediction scenario should fall within the prediction interval defined by the point prediction; By iteratively solving the optimization problem with different initial values, a set of optimal solutions is obtained; The optimal solution set is input into the generator to predict a set of different and realistic system operation scenarios at future moments.

[0013] The present invention also provides a source-load prediction model processing system based on an artificial neural network. When the system is running, the source-load prediction model processing method based on an artificial neural network is executed.

[0014] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the steps of the above-mentioned source load prediction model processing method based on artificial neural network are implemented.

[0015] The present invention also provides a storage medium, in which computer executable instructions are stored. When the computer executable instructions are loaded and executed by a processor, the steps of the above-mentioned source load prediction model processing method based on artificial neural network are implemented.

[0016] The substantial effects of the present invention include: By using the generative adversarial network (WGAN) to extract features from historical operation scenarios, the present invention can generate diversified scenarios that are consistent with the statistical distribution of the system's historical operation scenarios, thereby improving the prediction accuracy of the system's operation scenarios at future moments. This unsupervised representation method effectively captures the inherent laws and uncertainties of system operation, making the prediction results closer to reality.

[0017] The WGAN model in the present invention uses a minimum-maximum game to alternately train the generator and the discriminator to optimize the loss function, so that it can handle the complex nonlinear relationship between the source and the load. Compared with traditional statistical models, this deep learning method performs well in processing high-dimensional and nonlinear data, significantly improving the accuracy and reliability of predictions.

[0018] The present invention represents the system operation scenario as a multidimensional time series and introduces the concept of channel for joint representation, which effectively reduces the mutual influence in the feature extraction process and improves the data processing efficiency. At the same time, the scenario prediction optimization problem construction method based on WGAN can flexibly handle scenario predictions within different prediction time ranges to meet the needs of different application scenarios.

[0019] By accurately describing the uncertainty of system operation, the present invention provides a rich and accurate source of information for the system's real-time energy control strategy, which helps system operators to formulate more scientific and reasonable operation plans and enhance the flexibility and resilience of system operation. When dealing with extreme weather, emergencies and other situations, it can respond more quickly and accurately to ensure the safe and stable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution will be clearly and completely described below in combination with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0023] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0024] It should be understood that in the present invention, "plurality" refers to two or more than two. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "Contains A, B and C", "Contains A, B, C" means that A, B, and C are all included, "Contains A, B or C" means that one of A, B, and C is included, and "Contains A, B and / or C" means that any one, any two, or any three of A, B, and C are included.

[0025] The technical solution of the present invention is described in detail with specific embodiments below. The embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0026] Example: Source load prediction model processing method based on artificial neural network, such as Figure 1 As shown, the following steps are included: S1: WGAN is trained based on historical operation scenarios, so that the generator can convert random vectors sampled from a known distribution into any real scenario with a statistical distribution consistent with the system's historical operation scenarios, thereby establishing an unsupervised representation of the statistical distribution of the system's operation scenarios to complete the generation of system operation scenarios.

[0027] S2: Obtain the observed values ​​and point prediction values ​​of uncertain variables, construct a scenario prediction optimization problem based on the trained WGAN and the observed values ​​and point prediction values ​​of uncertain variables, so as to obtain the optimal input vector of the generator, perform scenario prediction, and obtain a set of different system operation scenarios at future moments.

[0028] Specifically, S1 includes: Define the variables involved in system operation, and define the complete operation scenario for capturing the uncertainty of system operation by multi-dimensional uncertain variables; Construct a WGAN model, including two neural networks, the generator and the discriminator. Take the Wasserstein distance as the training target, and train the generator and the discriminator in alternating competition through the minimum-maximum game. During the training process, the parameters of the generator and the discriminator are continuously updated by optimizing the loss function. When the discriminator can no longer distinguish between real and generated scenarios, the WGAN training is considered to have reached an equilibrium state. The trained WGAN generator is used to sample random vectors from a simple known distribution as needed and convert them into system operation scenarios.

[0029] In order to achieve system operation scenario prediction, we must first train WGAN based on historical operation scenarios, so that the generator can convert random vectors sampled from simple known distributions into any real scenario with a statistical distribution consistent with the system's historical operation scenarios, thereby establishing an unsupervised representation of the statistical distribution of system operation scenarios. It is worth noting that the above process can also be called scenario generation. For the sake of clarity, this process is uniformly referred to as scenario generation in the following text.

[0030] The variables involved in the operation of the system at time t are t It is expressed as: Where P t PV , P t WT , P t CHP , P t EB and P t load Respectively represent the electric power of PV, WT, CHP, EB and power load at the tth moment; P t dis and P t ch They represent the discharging and charging power of ESS at the tth moment respectively; P t grid is the electric power transmitted between the system and the main grid at the tth moment; and denote the thermal power of CHP, GB, EB and thermal load at the tth moment respectively; and They represent the heat release and heat charging power of TSS at the tth moment respectively; Ω t All variables except P t PV , P t WT , P t load and All other variables are controllable; The system operation scenario is represented as a multidimensional time series, where each dimension corresponds to a variable. The concept of channel is introduced, and the multidimensional time series is jointly represented as a whole with multiple channels. This results in a complete operation scenario for capturing the uncertainty of system operation, which is convenient for feature extraction.

[0031] A complete operational scenario that captures the uncertainty of system operation can be defined as follows: Where T is the time range of the system operation scenario.

[0032] During the operation, the system operator has recorded a series of historical operation scenarios that can characterize the actual dynamic behavior of the system. Generally speaking, the four uncertain variables that constitute the system operation scenario are affected by different factors, such as PV power generation is mainly affected by solar radiation intensity, while WT power generation is mainly affected by wind speed. Therefore, it can be considered that these four variables are independent of each other. In order to efficiently extract features from historical scenarios, this embodiment migrates the channel concept (i.e., RGB channel) in the field of color images to the system operation scenario. In this way, the system operation scenario composed of multidimensional time series can be jointly represented as a whole with four channels and each channel is a time series of length T. Moreover, since each channel is relatively independent in the generative learning process, the mutual influence between different variables in the feature extraction process can be minimized. In addition, this embodiment also designs a corresponding WGAN based on the characteristics of the system operation scenario to adaptively construct an unsupervised representation of the statistical distribution of the historical operation scenario without any explicit statistical assumptions. The network uses Wasserstein distance as the training target and adopts a one-dimensional convolution layer suitable for extracting time series features from time series and generating time series in the network structure.

[0033] GAN consists of two neural networks called generator and discriminator. The generator is trained to convert random latent vectors sampled from simple known distributions (such as Gaussian distribution or uniform distribution) into sufficiently realistic data (i.e., consistent with the statistical distribution of historical data), while the discriminator is trained to distinguish whether the data is true (samples from historical data are true data) or false (samples converted by the generator are false data). When the alternating competitive training between the generator and the discriminator reaches equilibrium, the discriminator will no longer be able to distinguish whether the data comes from historical data or generated data, and the GAN is considered to have been trained.

[0034] The goal of system scenario generation is to generate diverse scenarios that can correctly capture the statistical distribution of historical operating scenarios. Wasserstein distance is an intuitive criterion for measuring the difference between two statistical distributions. Therefore, in order to perform scenario generation tasks in a reliable way, the training process of GAN can be prompted to establish an unsupervised representation of a distribution with the minimum Wasserstein distance from the statistical distribution of historical scenarios. Therefore, in this chapter, in the training of GAN, Wasserstein distance is used as the value function of the game between the generator and the discriminator (i.e., WGAN). In this way, when the game between them reaches equilibrium, a generator can be obtained so that the statistical distribution of the generated scenario is as close as possible to the statistical distribution of the historical scenario, thereby better completing the scenario generation task.

[0035] Among them, the training objective of WGAN is as follows: Where G and D represent the generator and discriminator respectively; x and P x Represent the system's historical operating scenarios and corresponding statistical distributions respectively; z is the input implicit vector of the generator and z~P z ;P z Let P be a simple known distribution that is easy to sample. z Set to a uniform distribution of [-1,1]; The generator and the discriminator compete through a min-max game, and the generator is trained to deceive the discriminator into treating the generated scenes as real scenes, where the output of the generator is represented as G(z) and G(z)~P G , here P G Represents the statistical distribution of the generated scene, and the discriminator and the generator are trained alternately to distinguish P x and P G , where the output of the discriminator is an indication of whether the scene is real or fake, and the larger the output value, the more real the scene is from the perspective of the discriminator. The loss functions of the generator and the discriminator can be defined as: Where E represents the empirical mean during the training process; The alternating competitive training of the discriminator and the generator makes them evolve continuously until the discriminator can no longer distinguish the scenes generated by the generator from the real scenes, which also means that P x and P G The Wasserstein distance between them converges to a minimum value as the game between the generator and the discriminator reaches equilibrium. At this point, it can be considered that the generator has learned the true statistical distribution of the system operation scenario.

[0036] When the parameters of the discriminator are given, P x and P G The Wasserstein distance between is expressed as: where θ (D) represents the parameters of the discriminator. When the discriminator can no longer distinguish between real scenes and generated scenes, its parameters θ (G) are fixed and used to generate sufficiently realistic system operation scenarios.

[0037] Specifically, S2 includes: Obtain the observed values ​​and point prediction values ​​of uncertain variables, the domain of random latent vectors sampled from uniform distribution, and determine the input vector space of the generator; Construct the objective function of the optimization problem, which includes two parts: one part measures the closeness between the prediction scenario and the observed value, and the other part is used as a penalty term to ensure the authenticity of the prediction scenario; construct constraints to ensure that the prediction scenario does not conflict with the point prediction information, that is, the prediction scenario should fall within the prediction interval defined by the point prediction; By iteratively solving the optimization problem with different initial values, a set of optimal solutions is obtained; The optimal solution set is input into the generator to predict a set of different and realistic system operation scenarios at future moments.

[0038] For the prediction of system operation scenarios, the predicted scenarios must not only have the same statistical characteristics as the real scenarios, but also follow specific information. Assuming the current moment is k-1, the information that can be obtained includes the observed values ​​of the uncertain variables at the past k-1 moments. And point prediction information at future time k~T Therefore, the system's operating scenario prediction task is to predict a set of operating scenarios at future time k~T based on the above given information to accurately capture the possible dynamics and changes of the system's operating uncertainty.

[0039] The trained generator can generate any operation scenario that maintains the same mode as the system operation by converting the random implicit vector sampled from the uniform distribution of [-1,1]. By optimizing the input vector of the generator, a specific scenario can be generated according to specific information to accurately characterize the uncertainty of system operation in a specific period in the future. For this purpose, a scenario prediction optimization problem is constructed based on the trained WGAN and the observed values ​​and point prediction values ​​of uncertain variables to obtain the optimal input vector of the generator, thereby realizing the system operation scenario prediction. This method can flexibly perform scenario prediction without limiting the prediction range. The optimization problem for system operation scenario prediction is constructed as follows: The part of the scenario predicted at the k-1th moment in the past k-1 moments must be as close as possible to the observed value of the uncertain variable at the k-1th moment in the past, and the part in the future k~T moments must reflect the actual operating characteristics of the system and cannot conflict with the point prediction information. Based on the above ideas, the system operation scenario prediction optimization problem can be defined as: z∈Z; Where P obse (G(z)) and P pred (G(z)) represents the part of G(z) before k-1 moments and in the future k~T moments respectively; γ is a weight parameter to determine the influence of the discriminator output on the objective function; Z represents the domain of the vector z; and P pred Lower and upper bounds for (G(z)).

[0040] The meaning of the objective function is: the first term is a cost function to measure P obse (G(z)) and s obse The second term is a penalty term to help ensure that the predicted scenario is as realistic as possible. Since the input of the generator is a random implicit vector sampled from a uniform distribution of [-1,1], the domain of z is a hypercube bounded by [-1,1]. In addition, in order to prevent the predicted scenario from conflicting with the point prediction information, the predicted scenario should not fall outside the prediction interval defined by the point prediction. Based on this consideration, the third term is constructed, where and Can be calculated according to the following formula: Where α is a predefined point prediction confidence level, which plays a role in controlling the fluctuation range of the prediction scenario and α>1.

[0041] The optimization problem contains highly non-convex generators and discriminators, so there are a large number of local optimal solutions to this optimization problem. By iteratively solving the optimization problem with different initial values ​​z, a set of local optimal solutions can be obtained. Then, the optimal solution set is input into the generator to predict a set of different and realistic system operation scenarios at future k~T moments: where z * represents the set of local optimal solutions; represents the i-th local optimal solution; m is the number of times the optimization problem is iterated. By adjusting m, the number of prediction scenarios can be easily adjusted; S represents the prediction scenario set.

[0042] In order to solve the optimization problem efficiently, the inequality constraints can be incorporated into the objective function as follows: Where κ and ν are weight parameters and κ=ν.

[0043] In addition, the pre-optimization problem can be solved first to obtain a better initial value z for iteration, thereby improving the quality of the local optimal solution: z∈Z; where s ini For The samples are randomly sampled from , and the formula α is slightly smaller than the previous setting.

[0044] This embodiment also provides a source-load prediction model processing system based on an artificial neural network. When the system is running, the above-mentioned source-load prediction model processing method based on an artificial neural network is executed.

[0045] This embodiment also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the steps of the above-mentioned source load prediction model processing method based on artificial neural network are implemented.

[0046] This embodiment also provides a storage medium, in which computer executable instructions are stored. When the computer executable instructions are loaded and executed by a processor, the steps of the above-mentioned source load prediction model processing method based on artificial neural network are implemented.

[0047] The substantial effects of this embodiment include: By using the generative adversarial network (WGAN) to extract features from historical operation scenarios, this embodiment can generate diversified scenarios that are consistent with the statistical distribution of the system's historical operation scenarios, thereby improving the prediction accuracy of the system's operation scenarios at future moments. This unsupervised representation method effectively captures the inherent laws and uncertainties of system operation, making the prediction results closer to reality.

[0048] The WGAN model in this embodiment uses the minimum-maximum game to alternately train the generator and the discriminator to optimize the loss function so that it can handle the complex nonlinear relationship between the source and the load. Compared with traditional statistical models, this deep learning method performs well in processing high-dimensional and nonlinear data, significantly improving the accuracy and reliability of predictions.

[0049] This embodiment represents the system operation scenario as a multi-dimensional time series and introduces the concept of channels for joint representation, which effectively reduces the mutual influence in the feature extraction process and improves data processing efficiency. At the same time, the scenario prediction optimization problem construction method based on WGAN can flexibly handle scenario predictions within different prediction time ranges to meet the needs of different application scenarios.

[0050] By accurately describing the uncertainty of system operation, this embodiment provides a rich and accurate source of information for the system's real-time energy control strategy, which helps system operators to formulate more scientific and reasonable operation plans and enhance the flexibility and resilience of system operation. When dealing with extreme weather, emergencies, etc., it can respond more quickly and accurately to ensure the safe and stable operation of the system.

[0051] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the specific device can be divided into different functional modules to complete all or part of the functions described above.

[0052] In the embodiments provided in the present application, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the embodiments of the structure described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another structure, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, structures or units, which can be electrical, mechanical or other forms.

[0053] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0054] In addition, each functional unit in the embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0055] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0056] The above contents are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A source load prediction model processing method based on an artificial neural network, characterized in that: The following steps are involved: Based on historical operation scenarios, WGAN is trained so that the generator can convert random vectors sampled from a known distribution into any real scenario with a statistical distribution consistent with the system's historical operation scenarios, thereby establishing an unsupervised representation of the statistical distribution of system operation scenarios to complete system operation scenario generation; Obtain the observed values ​​and point prediction values ​​of uncertain variables, build a scenario prediction optimization problem based on the trained WGAN and the observed values ​​and point prediction values ​​of uncertain variables, so as to obtain the optimal input vector of the generator, perform scenario prediction, and obtain a set of system operation scenarios at different future moments.

2. The source load prediction model processing method based on artificial neural network according to claim 1 is characterized in that: The WGAN is trained based on historical operation scenarios, so that the generator can convert random vectors sampled from a known distribution into any real scenario with a statistical distribution consistent with the historical operation scenarios of the system, thereby establishing an unsupervised representation of the statistical distribution of the system operation scenarios to complete the generation of system operation scenarios, including: Define the variables involved in system operation, and define the complete operation scenario for capturing the uncertainty of system operation by multi-dimensional uncertain variables; Construct a WGAN model, including two neural networks, the generator and the discriminator. Take the Wasserstein distance as the training target, and train the generator and the discriminator in alternating competition through the minimum-maximum game. During the training process, the parameters of the generator and the discriminator are continuously updated by optimizing the loss function. When the discriminator can no longer distinguish between real and generated scenarios, the WGAN training is considered to have reached an equilibrium state. The trained WGAN generator is used to sample random vectors from a simple known distribution as needed and convert them into system operation scenarios.

3. The source load prediction model processing method based on artificial neural network according to claim 2 is characterized in that: The variables involved in the definition of system operation are defined by multi-dimensional uncertain variables to capture the complete operation scenario of the system operation uncertainty, including: The variables involved in the operation of the system at time t are t It is expressed as: Where P t PV , P t WT , P t CHP , P t EB and P t load Respectively represent the electric power of PV, WT, CHP, EB and power load at the tth moment; P t dis and P t ch They represent the discharging and charging power of ESS at the tth moment respectively; P t grid is the electric power transmitted between the system and the main grid at the tth moment; and denote the thermal power of CHP, GB, EB and thermal load at the tth moment respectively; and They represent the heat release and heat charging power of TSS at the tth moment respectively; Ω t All variables except P t PV , P t WT , P t load and All other variables are controllable; The system operation scenario is represented as a multidimensional time series, where each dimension corresponds to a variable. The concept of channel is introduced, and the multidimensional time series is jointly represented as a whole with multiple channels. This results in a complete operation scenario for capturing the uncertainty of system operation, which is convenient for feature extraction.

4. The source load prediction model processing method based on artificial neural network according to claim 2 is characterized in that: The WGAN model is constructed, including two neural networks, a generator and a discriminator. The Wasserstein distance is used as the training target, and the generator and the discriminator are alternately trained in competition through the minimum-maximum game. During the training process, the parameters of the generator and the discriminator are continuously updated by optimizing the loss function, including: The training objective of WGAN is as follows: Where G and D represent the generator and discriminator respectively; x and P x Represent the system's historical operating scenarios and corresponding statistical distributions respectively; z is the input implicit vector of the generator and z~P z ;P z Let P be a simple known distribution that is easy to sample. z Set to a uniform distribution of [-1,1]; The generator and the discriminator compete through a min-max game, and the generator is trained to deceive the discriminator into treating the generated scenes as real scenes, where the output of the generator is represented as G(z) and G(z)~P G , here P G Represents the statistical distribution of the generated scene, and the discriminator and the generator are trained alternately to distinguish P x and P G , where the output of the discriminator is an indication of whether the scene is real or fake, and the larger the output value, the more real the scene is from the perspective of the discriminator. The loss functions of the generator and the discriminator can be defined as: Where E represents the empirical mean during the training process; The alternating competitive training of the discriminator and the generator makes them evolve continuously until the discriminator can no longer distinguish the scenes generated by the generator from the real scenes, which also means that P x and P G The Wasserstein distance between them converges to a minimum value as the game between the generator and the discriminator reaches equilibrium. At this point, it can be considered that the generator has learned the true statistical distribution of the system operation scenario.

5. The source load prediction model processing method based on artificial neural network according to claim 4 is characterized in that: When the discriminator can no longer distinguish between the real scene and the generated scene, it is considered that the WGAN training has reached an equilibrium state. The trained WGAN generator is used to sample random vectors from a simple known distribution as needed and convert them into system operation scenes, including: when the parameters of the discriminator are given, P x and P G The Wasserstein distance between is expressed as: where θ (D) represents the parameters of the discriminator. When the discriminator can no longer distinguish between real scenes and generated scenes, its parameters θ (G) are fixed and used to generate sufficiently realistic system operation scenarios.

6. The source load prediction model processing method based on artificial neural network according to claim 1 is characterized in that: The method obtains the observed values ​​and point prediction values ​​of the uncertain variables, constructs a scenario prediction optimization problem based on the trained WGAN and the observed values ​​and point prediction values ​​of the uncertain variables, thereby obtaining the optimal input vector of the generator, performing scenario prediction, and obtaining a set of system operation scenarios at different future moments, including: Obtain the observed values ​​and point prediction values ​​of uncertain variables, the domain of random latent vectors sampled from uniform distribution, and determine the input vector space of the generator; Construct the objective function of the optimization problem, which includes two parts: one part measures the closeness between the prediction scenario and the observed value, and the other part is used as a penalty term to ensure the authenticity of the prediction scenario; construct constraints to ensure that the prediction scenario does not conflict with the point prediction information, that is, the prediction scenario should fall within the prediction interval defined by the point prediction; By iteratively solving the optimization problem with different initial values, a set of optimal solutions is obtained; The optimal solution set is input into the generator to predict a set of different and realistic system operation scenarios at future moments.

7. The source load prediction model processing system based on artificial neural network is characterized by: When the system is running, the source load prediction model processing method based on artificial neural network as described in any one of claims 1 to 6 is executed.

8. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the steps of the source load prediction model processing method based on an artificial neural network as claimed in any one of claims 1 to 6 are implemented.

9. A storage medium, characterized in that: The storage medium stores computer executable instructions, and when the computer executable instructions are loaded and executed by the processor, the steps of the source load prediction model processing method based on an artificial neural network as described in any one of claims 1 to 6 are implemented.

Images