Grid-connected power supply system of distributed PV grid-connected cabinet

Abstract

The present application discloses a grid-connected power supply system for a distributed photovoltaic grid-connected cabinet. By comprehensively utilizing the output power data of photovoltaic panels, the remaining capacity data of the energy storage system, and the grid load demand data, and combining deep learning algorithms to analyze the temporal correlation feature patterns of each sample data, and based on a posteriori inference-based fusion means, the temporal distribution patterns of the output power of photovoltaic panels, the temporal distribution patterns of the remaining capacity of the energy storage system, and the temporal distribution patterns of grid load demand are multi-source integrated to form a comprehensive representation of the system's power generation state characteristics, so as to guide the adjustment of the output power of photovoltaic panels. In this way, the output power of photovoltaic panels is adaptively adjusted according to the actual demand and the real-time power generation state characteristics of the system to ensure the stable operation of the power supply system and the grid.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to a grid-connected power supply system for a distributed photovoltaic grid-connected cabinet. Background Art

[0002] With the growth of global energy demand and the increasingly severe environmental problems, renewable energy, especially solar energy, as a clean and sustainable energy form, has received extensive attention in its development and utilization. The distributed photovoltaic power generation system, with its flexibility and small environmental impact, has become an important way to promote energy transformation and optimize the energy structure.

[0003] However, due to the intermittency and instability of solar energy, how to efficiently manage and control the power supply system of the distributed photovoltaic grid-connected cabinet to ensure its stable grid-connected operation with the power grid has become an urgent technical problem to be solved. Summary of the Invention

[0004] To solve the above technical problems, the present application is proposed. An embodiment of the present application provides a grid-connected power supply system for a distributed photovoltaic grid-connected cabinet, which comprehensively utilizes the output power data of photovoltaic panels, the remaining capacity data of energy storage systems, and the grid load demand data, combines deep learning algorithms to analyze the time-series correlation feature patterns of each sample data, and based on the posterior inference-based fusion means, multi-source integrates the time-series distribution patterns of the output power of photovoltaic panels, the time-series distribution patterns of the remaining capacity of energy storage systems, and the time-series distribution patterns of grid load demands to form a comprehensive representation of the system power generation state characteristics, so as to guide the adjustment of the output power of photovoltaic panels. In this way, the output power of photovoltaic panels is adaptively adjusted according to actual needs and the real-time power generation state characteristics of the system to ensure the stable operation of the power supply system and the power grid.

[0005] According to one aspect of the present application, there is provided a grid-connected power supply system for a distributed photovoltaic grid-connected cabinet, which includes:

[0006] A parameter real-time acquisition module for acquiring the time series of the output power of photovoltaic panels, the time series of the remaining capacity of energy storage systems, and the time series of grid load demands;

[0007] A parameter time-series correlation feature pattern extraction module for respectively extracting time-series correlation feature patterns from the time series of the output power of the photovoltaic panels, the time series of the remaining capacity of the energy storage systems, and the time series of grid load demands to obtain an enhanced photovoltaic panel output time-series correlation feature vector, an enhanced energy storage system remaining capacity time-series correlation feature vector, and an enhanced grid load demand time-series correlation feature vector;

[0008] The posterior fusion module is used to perform posterior inference fusion on the output timing correlation feature vector of the enhanced photovoltaic panel, the remaining capacity timing correlation feature vector of the enhanced energy storage system, and the load demand timing correlation feature vector of the enhanced power grid to obtain the posterior inference feature vector of the photovoltaic panel power output;

[0009] The control instruction generation module is used to determine a control instruction based on the posterior inference feature vector of the photovoltaic panel power output, and the control instruction is used to represent increasing the output power of the photovoltaic panel, decreasing the output power of the photovoltaic panel, or maintaining the output power of the photovoltaic panel;

[0010] The DC / AC conversion module is used to convert the direct current generated by the photovoltaic panel into alternating current with the same amplitude, frequency, and phase as the voltage of the power grid through an inverter.

[0011] Compared with the prior art, the grid-connected power supply system of a distributed photovoltaic grid-connected cabinet provided by the present application comprehensively utilizes the output power data of the photovoltaic panel, the remaining capacity data of the energy storage system, and the load demand data of the power grid, combines deep learning algorithms to analyze the timing correlation feature patterns of each sample data, and based on the posterior inference-based fusion method, multi-source integrates the timing distribution pattern of the photovoltaic panel output power, the timing distribution pattern of the remaining capacity of the energy storage system, and the timing distribution pattern of the power grid load demand to form a comprehensive representation of the system power generation state characteristics to guide the adjustment of the output power of the photovoltaic panel. In this way, the output power of the photovoltaic panel is adaptively adjusted according to the actual demand and the real-time power generation state characteristics of the system to ensure the stable operation of the power supply system and the power grid. Description of the Drawings

[0012] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. They are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 It is a block diagram of the grid-connected power supply system of a distributed photovoltaic grid-connected cabinet according to an embodiment of the present application;

[0014] Figure 2 It is a system architecture diagram of the grid-connected power supply system of a distributed photovoltaic grid-connected cabinet according to an embodiment of the present application;

[0015] Figure 3 It is a block diagram of the parameter timing correlation feature pattern extraction module in the grid-connected power supply system of a distributed photovoltaic grid-connected cabinet according to an embodiment of the present application;

[0016] Figure 4It is a block diagram of a feature vector enhancement unit in a grid-connected power supply system of a distributed photovoltaic grid-connected cabinet according to an embodiment of the present application. Detailed implementation manners

[0017] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0018] As shown in the present application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "including" and "comprising" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. A method or device may also include other steps or elements.

[0019] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0020] In the technical solution of the present application, a grid-connected power supply system of a distributed photovoltaic grid-connected cabinet is proposed.

[0021] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0022] In the technical solution of the present application, a grid-connected power supply system of a distributed photovoltaic grid-connected cabinet is proposed. Figure 1 It is a block diagram of a grid-connected power supply system of a distributed photovoltaic grid-connected cabinet according to an embodiment of the present application. Figure 2 It is a system architecture diagram of a grid-connected power supply system of a distributed photovoltaic grid-connected cabinet according to an embodiment of the present application. As Figure 1 and Figure 2As shown, the grid-connected power supply system 300 of the distributed photovoltaic grid-connected cabinet according to an embodiment of the present application includes: a parameter real-time acquisition module 310, configured to acquire the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand; a parameter time series correlation feature pattern extraction module 320, configured to perform time series correlation feature pattern extraction on the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand respectively to obtain an enhanced photovoltaic panel output time series correlation feature vector, an enhanced energy storage system remaining capacity time series correlation feature vector, and an enhanced grid load demand time series correlation feature vector; a posterior fusion module 330, configured to perform posterior inference fusion on the enhanced photovoltaic panel output time series correlation feature vector, the enhanced energy storage system remaining capacity time series correlation feature vector, and the enhanced grid load demand time series correlation feature vector to obtain a photovoltaic panel power output posterior inference feature vector; a control instruction generation module 340, configured to determine a control instruction based on the photovoltaic panel power output posterior inference feature vector, where the control instruction is used to indicate increasing the output power of the photovoltaic panel, decreasing the output power of the photovoltaic panel, or maintaining the output power of the photovoltaic panel; a DC / AC conversion module 350, configured to convert the direct current generated by the photovoltaic panel into alternating current with the same amplitude, frequency, and phase as the voltage of the grid through an inverter.

[0023] Specifically, the parameter real-time acquisition module 310 is configured to acquire the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand. It should be understood that these data can provide important information such as the power generation status of the photovoltaic panel, the status of the energy storage system, and the real-time load demand of the grid, which are important bases for monitoring and analyzing the operation status of the power supply system. In the embodiment of the present application, the specific manner of acquiring the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand may be: first, install corresponding sensors and measurement devices at the connection points of the photovoltaic panel, the energy storage system, and the grid for real-time monitoring and recording of relevant parameters. Subsequently, use a data acquisition system (Data Acquisition System, DAS) to collect the readings of the sensors; and transmit the collected data to a central monitoring system or a cloud server through a wired or wireless communication network. This may include using Ethernet, Wi-Fi, cellular networks, or other industrial communication protocols.

[0024] Specifically, the parameter time-series correlation feature pattern extraction module 320 is configured to perform time-series correlation feature pattern extraction on the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand respectively, so as to obtain an enhanced photovoltaic panel output time-series correlation feature vector, an enhanced energy storage system remaining capacity time-series correlation feature vector, and an enhanced grid load demand time-series correlation feature vector. Specifically, in a specific example of the present application, as Figure 3 shown, the parameter time-series correlation feature pattern extraction module 320 includes: a time-series correlation encoding unit 321, configured to input the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand into a sequence encoder based on GRU units respectively, so as to obtain a photovoltaic panel output time-series correlation feature vector, an energy storage system remaining capacity time-series correlation feature vector, and a grid load demand time-series correlation feature vector; a vector segmentation unit 322, configured to perform vector segmentation on the photovoltaic panel output time-series correlation feature vector, the energy storage system remaining capacity time-series correlation feature vector, and the grid load demand time-series correlation feature vector, so as to obtain a sequence of photovoltaic panel output local time-series correlation feature vectors, a sequence of energy storage system remaining capacity local time-series correlation feature vectors, and a sequence of grid load demand local time-series correlation feature vectors; a feature vector enhancement unit 323, configured to input the sequence of the photovoltaic panel output local time-series correlation feature vectors, the sequence of the energy storage system remaining capacity local time-series correlation feature vectors, and the sequence of the grid load demand local time-series correlation feature vectors into a context feature vector enhancement module based on saliency-globalness respectively, so as to obtain the enhanced photovoltaic panel output time-series correlation feature vector, the enhanced energy storage system remaining capacity time-series correlation feature vector, and the enhanced grid load demand time-series correlation feature vector.

[0025] Specifically, the time-series correlation encoding unit 321 is configured to input the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand into a sequence encoder based on GRU units respectively, so as to obtain a photovoltaic panel output time-series correlation feature vector, an energy storage system remaining capacity time-series correlation feature vector, and a grid load demand time-series correlation feature vector. Among them, GRU is a special recurrent neural network (RNN), which is usually used to process sequence data. In the technical solution of the present application, GRU is used to construct a sequence encoder to capture the dynamic change laws and time dependencies of the output power of the photovoltaic panel, the remaining capacity of the energy storage system, and the grid load demand respectively.

[0026] Specifically, the vector segmentation unit 322 is configured to segment the photovoltaic panel output time-series correlation feature vector, the energy storage system remaining capacity time-series correlation feature vector, and the grid load demand time-series correlation feature vector to obtain a sequence of photovoltaic panel output local time-series correlation feature vectors, a sequence of energy storage system remaining capacity local time-series correlation feature vectors, and a sequence of grid load demand local time-series correlation feature vectors. Considering that the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand not only contain long-term time dependencies but also include local short-term change trends. These local short-term change trends are also of great significance for grasping the complex time-series feature patterns of each sample data. At the same time, there are also complex correlation relationships between the local fluctuation patterns in the time dimension. Therefore, in the technical solution of this application, the photovoltaic panel output time-series correlation feature vector, the energy storage system remaining capacity time-series correlation feature vector, and the grid load demand time-series correlation feature vector are first segmented to highlight the short-term change rules within each local time span, so as to obtain a sequence of photovoltaic panel output local time-series correlation feature vectors, a sequence of energy storage system remaining capacity local time-series correlation feature vectors, and a sequence of grid load demand local time-series correlation feature vectors

[0027] Specifically, the feature vector enhancement unit 323 is configured to input the sequence of photovoltaic panel output local time-series correlation feature vectors, the sequence of energy storage system remaining capacity local time-series correlation feature vectors, and the sequence of grid load demand local time-series correlation feature vectors into the saliency-global context feature vector enhancement module respectively to obtain the enhanced photovoltaic panel output time-series correlation feature vector, the enhanced energy storage system remaining capacity time-series correlation feature vector, and the enhanced grid load demand time-series correlation feature vector. That is, in the technical solution of this application, the sequence of photovoltaic panel output local time-series correlation feature vectors, the sequence of energy storage system remaining capacity local time-series correlation feature vectors, and the sequence of grid load demand local time-series correlation feature vectors are input into the saliency-global context feature vector enhancement module respectively to further refine the context correlations between the local short-term change patterns in each sample data, so as to obtain the enhanced photovoltaic panel output time-series correlation feature vector, the enhanced energy storage system remaining capacity time-series correlation feature vector, and the enhanced grid load demand time-series correlation feature vector. Among them, the saliency-global context feature vector enhancement module uses global features and salient features (prominent features) to enhance the attention of the context correlation features of the local distribution patterns, thereby improving the utilization ability of key features. In particular, in a specific example of this application, such as Figure 4As shown, the feature vector enhancement unit 323 includes: a photovoltaic panel output prominent feature extraction subunit 3231, configured to extract a photovoltaic panel output prominent feature vector from a sequence of the photovoltaic panel output local temporal correlation feature vectors; a photovoltaic panel output global feature extraction subunit 3232, configured to extract a photovoltaic panel output global feature vector from the sequence of the photovoltaic panel output local temporal correlation feature vectors; a neighborhood encoding and activation subunit 3233, configured to perform neighborhood encoding and activation on the photovoltaic panel output prominent feature vector and the photovoltaic panel output global feature vector to obtain a photovoltaic panel output prominent feature context correlation activation feature vector and a photovoltaic panel output global feature context correlation activation feature vector; a fusion subunit 3234, configured to fuse the photovoltaic panel output prominent feature context correlation activation feature vector and the photovoltaic panel output global feature context correlation activation feature vector to obtain a photovoltaic panel output prominent-global feature context correlation activation feature vector, and then perform non-linear activation on the photovoltaic panel output prominent-global feature context correlation activation feature vector to obtain a photovoltaic panel output prominent-global feature context correlation weight feature vector; a reinforced photovoltaic panel output temporal correlation feature calculation subunit 3235, configured to use the photovoltaic panel output prominent-global feature context correlation weight feature vector as a weight to calculate a position-wise weighted sum of the sequence of the photovoltaic panel output local temporal correlation feature vectors to obtain the reinforced photovoltaic panel output temporal correlation feature vector.

[0028] More specifically, the photovoltaic panel output prominent feature extraction subunit 3231 is configured to extract a photovoltaic panel output prominent feature vector from a sequence of the photovoltaic panel output local temporal correlation feature vectors. In the technical solution of the present application, the maximum value of each photovoltaic panel output local temporal correlation feature vector in the sequence of the photovoltaic panel output local temporal correlation feature vectors is calculated to obtain a photovoltaic panel output prominent feature vector composed of each of the maximum values. Here, by calculating the maximum value, the most prominent feature information in each local time period can be identified.

[0029] More specifically, the photovoltaic panel output global feature extraction subunit 3232 is configured to extract a photovoltaic panel output global feature vector from a sequence of the photovoltaic panel output local temporal correlation feature vectors. In the technical solution of the present application, the average value of each photovoltaic panel output local temporal correlation feature vector in the sequence of the photovoltaic panel output local temporal correlation feature vectors is calculated to obtain a photovoltaic panel output global feature vector composed of each of the average values. Here, by calculating the average value, the overall trend and average distribution pattern in each local time period can be reflected.

[0030] More specifically, the neighborhood encoding and activation subunit 3233 is used to perform neighborhood encoding and activation on the photovoltaic panel output prominent feature vector and the photovoltaic panel output global feature vector to obtain a photovoltaic panel output prominent feature context-associated activation feature vector and a photovoltaic panel output global feature context-associated activation feature vector. In particular, in a specific example of the present application, one-dimensional convolutional encoding, non-linear activation, and linear transformation are performed on the photovoltaic panel output prominent feature vector to obtain a photovoltaic panel output prominent feature context-associated activation feature vector; at the same time, one-dimensional convolutional encoding, non-linear activation, and linear transformation are performed on the photovoltaic panel output global feature vector to obtain a photovoltaic panel output global feature context-associated activation feature vector. That is, one-dimensional convolutional encoding is used to capture the implicit association between the prominent feature distribution pattern and the global feature distribution pattern, and non-linear activation is introduced to learn and simulate complex feature expressions. Further, linear transformation is used to adjust and optimize the feature representation to provide a more suitable feature space for subsequent fusion.

[0031] More specifically, the fusion subunit 3234 is used to fuse the photovoltaic panel output prominent feature context-associated activation feature vector and the photovoltaic panel output global feature context-associated activation feature vector to obtain a photovoltaic panel output prominent-global feature context-associated activation feature vector, and then perform non-linear activation on the photovoltaic panel output prominent-global feature context-associated activation feature vector to obtain a photovoltaic panel output prominent-global feature context-associated weight feature vector. That is, in the technical solution of the present application, the photovoltaic panel output prominent feature context-associated activation feature vector and the photovoltaic panel output global feature context-associated activation feature vector are fused to obtain a photovoltaic panel output prominent-global feature context-associated activation feature vector; and non-linear activation based on the tanh activation function and non-linear activation based on the sigmoid activation function are performed on the photovoltaic panel output prominent-global feature context-associated activation feature vector to obtain the photovoltaic panel output prominent-global feature context-associated weight feature vector. Among them, the photovoltaic panel output prominent-global feature context-associated weight feature vector is a comprehensive representation obtained by integrating local saliency and overall global information.

[0032] More specifically, the enhanced photovoltaic panel output temporal correlation feature calculation subunit 3235 is used to calculate the position-wise weighted sum of the sequence of the photovoltaic panel output local temporal correlation feature vectors with the photovoltaic panel output prominent-global feature context-associated weight feature vector as the weight to obtain the enhanced photovoltaic panel output temporal correlation feature vector. That is, by weighting the local temporal correlation feature vectors of the photovoltaic panel output with the global feature context-associated weight feature vector, an enhanced photovoltaic panel output temporal correlation feature vector is obtained, so as to more comprehensively characterize the features and change trends of the photovoltaic panel output. In this way, the enhanced expression of the context-associated features is realized.

[0033] In summary, in the above embodiments, the sequences of the local temporal correlation feature vectors of the photovoltaic panel output, the local temporal correlation feature vectors of the remaining capacity of the energy storage system, and the local temporal correlation feature vectors of the grid load demand are respectively input into the saliency-global context feature vector enhancement module to obtain the enhanced temporal correlation feature vectors of the photovoltaic panel output, the enhanced temporal correlation feature vectors of the remaining capacity of the energy storage system, and the enhanced temporal correlation feature vectors of the grid load demand, including: using the saliency-global context feature vector enhancement module to perform feature vector enhancement on the sequences of the local temporal correlation feature vectors of the photovoltaic panel output, the local temporal correlation feature vectors of the remaining capacity of the energy storage system, and the local temporal correlation feature vectors of the grid load demand according to the following formula to obtain the enhanced temporal correlation feature vectors of the photovoltaic panel output, the enhanced temporal correlation feature vectors of the remaining capacity of the energy storage system, and the enhanced temporal correlation feature vectors of the grid load demand; where the formula is:

[0034] H' = Avg(H)

[0035] H" = Max(H)

[0036]

[0037]

[0038] C = ∑H ⊙ (tanh(C1 + C2) ⊙ σ(C1 + C2)) where H is the sequence of the local temporal correlation feature vectors of the photovoltaic panel output, H′ is the global feature vector of the photovoltaic panel output, H″ is the prominent feature vector of the photovoltaic panel output, Avg represents taking the average value, Max represents taking the maximum value, Conv represents one-dimensional convolution operation, ReLU represents the ReLU activation function, and are the first weight matrix and the second weight matrix for linear transformation respectively, C1 and C2 are the global feature context correlation activation feature vectors of the photovoltaic panel output and the prominent feature context correlation activation feature vectors of the photovoltaic panel output respectively, σ represents the Sigmoid function, tanh represents the tanh activation function, ⊙ represents dot product, and C is the enhanced temporal correlation feature vector of the photovoltaic panel output.

[0039] It is worth mentioning that in other specific examples of this application, the sequences of the locally time - series - correlated feature vectors of the photovoltaic panel output, the sequences of the locally time - series - correlated feature vectors of the remaining capacity of the energy storage system, and the sequences of the locally time - series - correlated feature vectors of the grid load demand can also be input into the significance - global context feature vector enhancement module in other ways to obtain the enhanced time - series - correlated feature vectors of the photovoltaic panel output, the enhanced time - series - correlated feature vectors of the remaining capacity of the energy storage system, and the enhanced time - series - correlated feature vectors of the grid load demand. For example: input the sequences of the locally time - series - correlated feature vectors of the photovoltaic panel output, the sequences of the locally time - series - correlated feature vectors of the remaining capacity of the energy storage system, and the sequences of the locally time - series - correlated feature vectors of the grid load demand; take the sequences of the locally time - series - correlated feature vectors of the photovoltaic panel output, the remaining capacity of the energy storage system, and the grid load demand as inputs and input them into the enhancement module respectively; the enhancement module will process these three input sequences separately, perform weighted sum processing on each time - series - correlated feature vector according to the significance and global features to obtain the enhanced feature vectors; obtain the enhanced time - series - correlated feature vectors of the photovoltaic panel output, the enhanced time - series - correlated feature vectors of the remaining capacity of the energy storage system, and the enhanced time - series - correlated feature vectors of the grid load demand.

[0040] It is worth mentioning that in other specific examples of this application, the time - series of the output power of the photovoltaic panel, the time - series of the remaining capacity of the energy storage system, and the time - series of the grid load demand can also be subjected to time - series - correlated feature pattern extraction respectively to obtain the enhanced time - series - correlated feature vectors of the photovoltaic panel output, the enhanced time - series - correlated feature vectors of the remaining capacity of the energy storage system, and the enhanced time - series - correlated feature vectors of the grid load demand. For example: input the time - series of the output power of the photovoltaic panel, the time - series of the remaining capacity of the energy storage system, and the time - series of the grid load demand; extract the time - series - correlated feature patterns of the time - series of the output power of the photovoltaic panel, the time - series of the remaining capacity of the energy storage system, and the time - series of the grid load demand to capture the associations and variation rules between them; convert the extracted time - series - correlated feature patterns into the form of feature vectors, with each time - series corresponding to a feature vector; to obtain the enhanced time - series - correlated feature vectors of the photovoltaic panel output, the enhanced time - series - correlated feature vectors of the remaining capacity of the energy storage system, and the enhanced time - series - correlated feature vectors of the grid load demand.

[0041] Specifically, the posterior fusion module 330 is configured to perform posterior inference fusion on the enhanced photovoltaic panel output timing correlation feature vector, the enhanced energy storage system remaining capacity timing correlation feature vector, and the enhanced power grid load demand timing correlation feature vector to obtain the photovoltaic panel power output posterior inference feature vector. Specifically, in a specific example of the present application, the enhanced photovoltaic panel output timing correlation feature vector, the enhanced energy storage system remaining capacity timing correlation feature vector, and the enhanced power grid load demand timing correlation feature vector are input into a power output posterior inference device based on a Bayesian probability network-like to obtain the photovoltaic panel power output posterior inference feature vector. Among them, the Bayesian probability network is an inference method based on conditional probability, which uses prior knowledge and observed data to estimate the probability distribution of unknown parameters or hypotheses. The core of the Bayesian probability network is Bayes' theorem, which describes how to calculate the posterior probability based on the prior probability and the likelihood function (the conditional probability of the observed data). That is, by utilizing the characteristic that the Bayesian probability network can effectively model and represent the complex relationships between multiple variables, the complex non-linear interaction relationships between the photovoltaic panel output timing correlation feature pattern, the energy storage system remaining capacity timing correlation feature pattern, and the power grid load demand timing correlation feature pattern are described, so as to more comprehensively characterize and infer the operating state of the power grid.

[0042] In summary, in the above embodiment, inputting the enhanced photovoltaic panel output timing correlation feature vector, the enhanced energy storage system remaining capacity timing correlation feature vector, and the enhanced power grid load demand timing correlation feature vector into a power output posterior inference device based on a Bayesian probability network-like to obtain the photovoltaic panel power output posterior inference feature vector includes: using the power output posterior inference device based on the Bayesian probability network-like to perform posterior inference fusion on the enhanced photovoltaic panel output timing correlation feature vector, the enhanced energy storage system remaining capacity timing correlation feature vector, and the enhanced power grid load demand timing correlation feature vector according to the following formula to obtain the photovoltaic panel power output posterior inference feature vector; where the formula is:

[0043]

[0044] where q i represents the i-th eigenvalue of the photovoltaic panel power output posterior inference feature vector, p i represents the i-th eigenvalue of the enhanced photovoltaic panel output timing correlation feature vector, a i represents the i-th eigenvalue of the enhanced power grid load demand timing correlation feature vector, and b i represents the i-th eigenvalue of the enhanced energy storage system remaining capacity timing correlation feature vector.

[0045] Specifically, the control instruction generation module 340 and the DC / AC conversion module 350 are configured to determine a control instruction based on the posterior inference feature vector of the photovoltaic panel power output, where the control instruction is used to indicate increasing the output power of the photovoltaic panel, decreasing the output power of the photovoltaic panel, or maintaining the output power of the photovoltaic panel; and convert the direct current generated by the photovoltaic panel into alternating current with the same amplitude, frequency, and phase as the voltage of the power grid through an inverter. Specifically, in a specific example of the present application, the posterior inference feature vector of the photovoltaic panel power output is input into a classifier-based power output controller to obtain a control instruction, where the control instruction is used to indicate increasing the output power of the photovoltaic panel, decreasing the output power of the photovoltaic panel, or maintaining the output power of the photovoltaic panel; and convert the direct current generated by the photovoltaic panel into alternating current with the same amplitude, frequency, and phase as the voltage of the power grid through an inverter.

[0046] In particular, in a specific example of the present application, the grid-connected power supply system of the distributed photovoltaic grid-connected cabinet further includes: a training module for training the model; wherein, the training module includes: a training data acquisition unit for acquiring the time series of the training output power of the photovoltaic panel, the time series of the training remaining capacity of the energy storage system, and the time series of the training grid load demand; a training time series correlation encoding unit for respectively inputting the time series of the training output power of the photovoltaic panel, the time series of the training remaining capacity of the energy storage system, and the time series of the training grid load demand into a sequence encoder based on GRU units to obtain a training photovoltaic panel output time series correlation feature vector, a training energy storage system remaining capacity time series correlation feature vector, and a training grid load demand time series correlation feature vector; a training vector splitting unit for splitting the training photovoltaic panel output time series correlation feature vector, the training energy storage system remaining capacity time series correlation feature vector, and the training grid load demand time series correlation feature vector to obtain a sequence of training photovoltaic panel output local time series correlation feature vectors, a sequence of training energy storage system remaining capacity local time series correlation feature vectors, and a sequence of training grid load demand local time series correlation feature vectors; a training feature vector enhancement unit for respectively inputting the sequence of training photovoltaic panel output local time series correlation feature vectors, the sequence of training energy storage system remaining capacity local time series correlation feature vectors, and the sequence of training grid load demand local time series correlation feature vectors into a context feature vector enhancement module based on significance-globalness to obtain a training enhanced photovoltaic panel output time series correlation feature vector, a training enhanced energy storage system remaining capacity time series correlation feature vector, and a training enhanced grid load demand time series correlation feature vector; a training posterior fusion unit for inputting the training enhanced photovoltaic panel output time series correlation feature vector, the training enhanced energy storage system remaining capacity time series correlation feature vector, and the training enhanced grid load demand time series correlation feature vector into a power output posterior inference engine based on a class Bayesian probability network to obtain a training photovoltaic panel power output posterior inference feature vector; a classification loss function acquisition unit for inputting the training photovoltaic panel power output posterior inference feature vector into a power output controller based on a classifier to obtain a classification loss function; a backpropagation training unit for training the model by gradient backpropagation based on the loss function.

[0047] In an embodiment of the present application, the backpropagation training unit includes: training the model by gradient backpropagation based on the classification loss function.

[0048] In the technical solution of this application, the training photovoltaic panel output timing correlation feature vector, the training energy storage system remaining capacity timing correlation feature vector, and the training power grid load demand timing correlation feature vector respectively represent the local timing context correlation features of the training output power of the photovoltaic panel, the training remaining capacity of the energy storage system, and the training power grid load demand. When inputting the training photovoltaic panel output timing correlation feature vector, the training energy storage system remaining capacity timing correlation feature vector, and the training power grid load demand timing correlation feature vector into the saliency-global context feature vector enhancement module, the saliency-global context semantic enhancement module uses global features and prominent features to enhance context attention to improve the feature utilization ability of key features. However, this also causes a more significant local structure correspondence deviation in the feature distributions among the training enhanced photovoltaic panel output timing correlation feature vector, the training enhanced energy storage system remaining capacity timing correlation feature vector, and the training enhanced power grid load demand timing correlation feature vector. Thus, when inputting the training enhanced photovoltaic panel output timing correlation feature vector, the training enhanced energy storage system remaining capacity timing correlation feature vector, and the training enhanced power grid load demand timing correlation feature vector into the power output posterior inference engine based on the Bayesian probability network, there is a technical problem that the overall feature distribution of the training photovoltaic panel power output posterior inference feature vector obtained through position-by-position posterior probability calculation has a complex fine-grained structure distribution, which affects its classification convergence efficiency through the classifier, that is, it affects the efficiency of classification training and the accuracy of control instructions.

[0049] Therefore, during the model training process of this application, a predetermined loss function other than the classification loss function is further introduced. That is, in a preferred embodiment, training the model through gradient backpropagation based on the loss function includes the following steps: calculating a first photovoltaic panel power output posterior inference weight matrix and a second photovoltaic panel power output posterior inference weight matrix based on the training photovoltaic panel power output posterior inference feature vector, where the eigenvalues at the (i, j) positions of the first photovoltaic panel power output posterior inference weight matrix and the second photovoltaic panel power output posterior inference weight matrix are respectively half of the mean and the absolute value of the difference between the i-th eigenvalue and the j-th eigenvalue of the training photovoltaic panel power output posterior inference feature vector;

[0050] Performing query-style matrix multiplications of the training photovoltaic panel power output posterior inference feature vector with the first photovoltaic panel power output posterior inference weight matrix and the second photovoltaic panel power output posterior inference weight matrix respectively to obtain a first photovoltaic panel power output posterior inference intermediate vector and a second photovoltaic panel power output posterior inference intermediate vector;

[0051] Calculate the vector inner product of the posterior inference intermediate vector of the first PV panel power output and the posterior inference intermediate vector of the second PV panel power output to obtain the posterior inference loss term of the first PV panel power output;

[0052] Perform matrix multiplication on the posterior inference weight matrix of the first PV panel power output and the posterior inference weight matrix of the second PV panel power output, and calculate the Frobenius norm of the resulting matrix to obtain the posterior inference loss term of the second PV panel power output;

[0053] Subtract the product of the predetermined weight hyperparameter and the posterior inference loss term of the second PV panel power output from the posterior inference loss term of the first PV panel power output to obtain the posterior inference loss function of the PV panel power output; and

[0054] Optimize the model parameters through gradient backpropagation based on the weighted sum of the posterior inference loss function of the PV panel power output and the classification loss function.

[0055] Wherein, the posterior inference loss function of the PV panel power output is specifically expressed as:

[0056]

[0057]

[0058] Wherein, V represents the posterior inference feature vector of the training PV panel power output, v i and v j respectively represent the i-th eigenvalue and the j-th eigenvalue of the posterior inference feature vector of the training PV panel power output, M μ represents the posterior inference weight matrix of the first PV panel power output, M σ represents the posterior inference weight matrix of the second PV panel power output, M μ (i, j) represents the eigenvalue at the (i, j) position of the posterior inference weight matrix of the first PV panel power output, M σ (i, j) represents the eigenvalue at the (i, j) position of the posterior inference weight matrix of the second PV panel power output, represents matrix multiplication, T represents transpose, α represents the predetermined weight hyperparameter, ||·|| F represents the norm of the matrix, and Loss represents the posterior inference loss function of the PV panel power output.

[0059] That is, the posterior inference loss function of the photovoltaic panel power output performs query-based composition of the detail inner product space within the training photovoltaic panel power output posterior inference feature vector through the structured feature representation of the short-range-long-range cross-scale detail link of the training photovoltaic panel power output posterior inference feature vector, so as to approximate the low-rank independent observable composition formed by the link details provided by the structured detail interaction of the training photovoltaic panel power output posterior inference feature vector. In this way, by training with the posterior inference loss function of the photovoltaic panel power output, the distributed detail group of the training photovoltaic panel power output posterior inference feature vector can be used to perform detail group decomposition based on detail complexity, so as to promote the class regression decomposition recognition of the complex feature structure of the training photovoltaic panel power output posterior inference feature vector and improve the classification training efficiency.

[0060] As described above, the grid-connected power supply system 300 of the distributed photovoltaic grid-connected cabinet according to the embodiments of the present application can be implemented in various wireless terminals, such as a server with a grid-connected power supply algorithm for the distributed photovoltaic grid-connected cabinet. In a possible implementation manner, the grid-connected power supply system 300 of the distributed photovoltaic grid-connected cabinet according to the embodiments of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the grid-connected power supply system 300 of the distributed photovoltaic grid-connected cabinet can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the grid-connected power supply system 300 of the distributed photovoltaic grid-connected cabinet can also be one of the many hardware modules of the wireless terminal.

[0061] Alternatively, in another example, the grid-connected power supply system 300 of the distributed photovoltaic grid-connected cabinet and the wireless terminal can also be separate devices, and the grid-connected power supply system 300 of the distributed photovoltaic grid-connected cabinet can be connected to the wireless terminal through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0062] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technologies in the market, or to enable other ordinary skill in the art in the technical field to understand the embodiments disclosed herein.

Claims

1. A distributed photovoltaic grid-connected cabinet grid-connected power supply system, characterized in that: include: The parameter real-time acquisition module is used to obtain the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system and the time series of the grid load demand; A parameter timing correlation feature pattern extraction module is used to extract timing correlation feature patterns from the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand to obtain an enhanced photovoltaic panel output timing correlation feature vector, an enhanced energy storage system remaining capacity timing correlation feature vector, and an enhanced grid load demand timing correlation feature vector; A posteriori fusion module is used to perform a posteriori reasoning fusion on the enhanced photovoltaic panel output timing correlation feature vector, the enhanced energy storage system remaining capacity timing correlation feature vector and the enhanced power grid load demand timing correlation feature vector to obtain a posteriori reasoning feature vector for photovoltaic panel power output; A control instruction generation module, used to determine a control instruction based on the a posteriori reasoning feature vector of the photovoltaic panel power output, wherein the control instruction is used to indicate increasing the output power of the photovoltaic panel, decreasing the output power of the photovoltaic panel, or maintaining the output power of the photovoltaic panel; A DC / AC conversion module, used to convert the DC power generated by the photovoltaic panel into AC power with the same amplitude, frequency and phase as the voltage of the power grid through an inverter; The posterior fusion module is used to: Input the enhanced photovoltaic panel output timing correlation feature vector, the enhanced energy storage system remaining capacity timing correlation feature vector and the enhanced power grid load demand timing correlation feature vector into a power output posterior reasoner based on a Bayesian probability network to obtain the photovoltaic panel power output posterior reasoning feature vector; The power output a posteriori reasoning device based on the Bayesian probability network is used to perform a posteriori reasoning fusion on the enhanced photovoltaic panel output timing correlation feature vector, the enhanced energy storage system remaining capacity timing correlation feature vector and the enhanced power grid load demand timing correlation feature vector using the following formula to obtain the photovoltaic panel power output a posteriori reasoning feature vector; wherein the formula is: Among them, q i represents the ith eigenvalue of the posterior inference eigenvector of the photovoltaic panel power output, p i represents the i-th eigenvalue of the eigenvector of the output timing correlation of the enhanced photovoltaic panel, a i represents the i-th eigenvalue of the time series correlation eigenvector of the enhanced power grid load demand, b i Represents the i-th eigenvalue of the time series correlation eigenvector of the remaining capacity of the enhanced energy storage system.

2. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 1 is characterized in that: The parameter time series correlation characteristic pattern extraction module includes: A timing association encoding unit, used to input the time series of the output power of the photovoltaic panel, the time series of the remaining capacity of the energy storage system, and the time series of the grid load demand into a sequence encoder based on a GRU unit to obtain a photovoltaic panel output timing association feature vector, an energy storage system remaining capacity timing association feature vector, and a grid load demand timing association feature vector; A vector segmentation unit, used for performing vector segmentation on the photovoltaic panel output timing correlation feature vector, the energy storage system remaining capacity timing correlation feature vector and the power grid load demand timing correlation feature vector to obtain a sequence of photovoltaic panel output local timing correlation feature vectors, a sequence of energy storage system remaining capacity local timing correlation feature vectors and a sequence of power grid load demand local timing correlation feature vectors; The feature vector enhancement unit is used to input the sequence of the photovoltaic panel output local time series association feature vectors, the sequence of the energy storage system remaining capacity local time series association feature vectors and the sequence of the grid load demand local time series association feature vectors into the context feature vector enhancement module based on significance-globality to obtain the enhanced photovoltaic panel output time series association feature vector, the enhanced energy storage system remaining capacity time series association feature vector and the enhanced grid load demand time series association feature vector.

3. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 2 is characterized in that: The feature vector enhancement unit comprises: A photovoltaic panel output salient feature extraction subunit, used to extract a photovoltaic panel output salient feature vector of a sequence of local temporal correlation feature vectors output by the photovoltaic panel; A photovoltaic panel output global feature extraction subunit, used to extract a photovoltaic panel output global feature vector of a sequence of local temporal correlation feature vectors output by the photovoltaic panel; A neighborhood encoding and activation subunit, used for performing neighborhood encoding and activation on the photovoltaic panel output salient feature vector and the photovoltaic panel output global feature vector to obtain a photovoltaic panel output salient feature context-associated activation feature vector and a photovoltaic panel output global feature context-associated activation feature vector; a fusion subunit, configured to fuse the photovoltaic panel output salient feature context-associated activation feature vector and the photovoltaic panel output global feature context-associated activation feature vector to obtain a photovoltaic panel output salient-global feature context-associated activation feature vector, and then perform nonlinear activation on the photovoltaic panel output salient-global feature context-associated activation feature vector to obtain a photovoltaic panel output salient-global feature context-associated weighted feature vector; The enhanced photovoltaic panel output timing association feature calculation subunit is used to calculate the position-weighted sum of the sequence of the photovoltaic panel output local timing association feature vectors using the photovoltaic panel output salient-global feature context association weight feature vector as a weight to obtain the enhanced photovoltaic panel output timing association feature vector.

4. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 3 is characterized in that: The photovoltaic panel output salient feature extraction subunit is further used for: The maximum value of each photovoltaic panel output local time series correlation feature vector in the sequence of the photovoltaic panel output local time series correlation feature vector is calculated to obtain the photovoltaic panel output prominent feature vector composed of each of the maximum values.

5. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 4 is characterized in that: The photovoltaic panel output global feature extraction subunit is further used for: The average value of each photovoltaic panel output local time series correlation feature vector in the sequence of the photovoltaic panel output local time series correlation feature vector is calculated to obtain the photovoltaic panel output global feature vector composed of each of the average values.

6. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 5, characterized in that: The neighborhood encoding and activation subunit comprises: A salient feature context-related activation secondary subunit is used to perform one-dimensional convolution encoding, nonlinear activation and linear transformation on the salient feature vector output by the photovoltaic panel to obtain the salient feature context-related activation feature vector output by the photovoltaic panel; The global feature context associated activation secondary subunit is used to perform one-dimensional convolution encoding, nonlinear activation and linear transformation on the photovoltaic panel output global feature vector to obtain the photovoltaic panel output global feature context associated activation feature vector.

7. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 6, characterized in that: The fusion subunit is further used for: The photovoltaic panel output salient-global feature context associated activation feature vector is subjected to nonlinear activation based on a tanh activation function and nonlinear activation based on a sigmoid activation function to obtain the photovoltaic panel output salient-global feature context associated weight feature vector.

8. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 7, characterized in that: The control instruction generating module is used for: The photovoltaic panel power output a posteriori reasoning feature vector is input into a classifier-based power output controller to obtain the control instruction.

9. The grid-connected power supply system of the distributed photovoltaic grid-connected cabinet according to claim 8, characterized in that: Also includes: Training module, used to train the model; Wherein, the training module includes: A training data acquisition unit, used to acquire a time series of a training output power of a photovoltaic panel, a time series of a training remaining capacity of an energy storage system, and a time series of a training grid load demand; A training timing association coding unit, used to input the time series of the training output power of the photovoltaic panel, the time series of the training remaining capacity of the energy storage system, and the time series of the training grid load demand into a sequence encoder based on a GRU unit to obtain a training photovoltaic panel output timing association feature vector, a training energy storage system remaining capacity timing association feature vector, and a training grid load demand timing association feature vector; A training vector segmentation unit, used for performing vector segmentation on the training photovoltaic panel output timing correlation feature vector, the training energy storage system remaining capacity timing correlation feature vector and the training power grid load demand timing correlation feature vector to obtain a sequence of training photovoltaic panel output local timing correlation feature vectors, a sequence of training energy storage system remaining capacity local timing correlation feature vectors and a sequence of training power grid load demand local timing correlation feature vectors; A training feature vector reinforcement unit is used to input the sequence of the training photovoltaic panel output local time series association feature vectors, the sequence of the training energy storage system remaining capacity local time series association feature vectors and the sequence of the training power grid load demand local time series association feature vectors into a context feature vector reinforcement module based on significance-globality to obtain a training reinforcement photovoltaic panel output time series association feature vector, a training reinforcement energy storage system remaining capacity time series association feature vector and a training reinforcement power grid load demand time series association feature vector; A training a posteriori fusion unit is used to input the training enhanced photovoltaic panel output timing correlation feature vector, the training enhanced energy storage system remaining capacity timing correlation feature vector and the training enhanced power grid load demand timing correlation feature vector into a power output a posteriori reasoner based on a Bayesian probability network to obtain a training photovoltaic panel power output a posteriori reasoning feature vector; A classification loss function acquisition unit, used for inputting the training photovoltaic panel power output posterior inference feature vector into a classifier-based power output controller to obtain a classification loss function; The back-propagation training unit is used to train the model by back-propagating gradients based on the loss function.

Images