Data detection method for distributed photovoltaic system and energy storage system thereof

Abstract

The invention relates to a data detection method for a distributed photovoltaic system and an energy storage system thereof, and relates to the technical field of photovoltaic energy storage data detection. According to the method, a high-precision distributed photovoltaic and energy storage system data detection model is constructed by combining the powerful nonlinear feature extraction and state reconstruction capability of the deep belief network with the precise setting of key parameters by a second-order oscillation firefly optimization algorithm. Therefore, the system can predict the key operation state of equipment or positions where sensor nodes are not directly installed according to the actually measured data of a limited number of intelligent fusion terminals in the area, so that the deployment number of physical sensors and communication modules is greatly reduced on the hardware level, and the equipment investment and the later maintenance cost are directly reduced.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic energy storage data detection technology, and in particular to a data detection method for distributed photovoltaic and its energy storage systems. Background Technology

[0002] With the widespread deployment of distributed photovoltaic and various energy storage systems such as electrochemical and hydrogen energy storage, the power distribution network is evolving into a complex system containing massive, heterogeneous, and distributed source and storage units. The foundation and key to efficient management and optimized scheduling of this system lies in achieving comprehensive, accurate, and timely collection of operational status data from various devices. However, the traditional direct data collection model using physical sensors faces severe challenges: deploying complete sensing and communication modules in each unit would result in high hardware costs, communication loads, and maintenance pressures, leading to poor economic efficiency; reducing the number of points would result in data blind spots, making it impossible to accurately depict the overall operational status of the system. This problem constitutes the primary bottleneck in the current field of distributed energy data acquisition.

[0003] Furthermore, distributed photovoltaic output and various types of energy storage status data are characterized by high dimensionality, nonlinearity, and strong coupling. Traditional threshold alarms or simple statistical analyses are insufficient to deeply extract crucial information such as the true health status and operating modes of the system from limited and potentially noisy data. Especially for energy storage systems, their internal state cannot be directly measured, and methods relying on physical models are difficult to widely apply in practice at low cost and high accuracy due to model complexity and parameter drift issues, thus failing to provide deep and reliable state support for subsequent optimization control.

[0004] Existing technological solutions struggle to balance economic efficiency, real-time performance, and depth of state recognition. While some data-driven virtual sensing technologies can infer variables not directly measured through soft measurement methods, they typically target single devices or types, lacking the ability to coordinate the overall state of the coupled "photovoltaic-multi-energy storage" system. Their models are often computationally complex and cannot meet system-level real-time requirements. Therefore, there is an urgent need in this field for an innovative method that can efficiently process massive amounts of heterogeneous data at low cost and accurately reconstruct and predict the internal and external operating states of the system, thereby overcoming the current bottlenecks in the refined operation and management of distributed photovoltaic and multi-type energy storage systems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a data detection method for distributed photovoltaic and its energy storage systems.

[0006] A data detection method for distributed photovoltaic and its energy storage system includes the following steps: S1: Obtain historical data from distributed photovoltaic and its energy storage systems and divide it into training set, test set and verification set. The historical data includes input data and target power data. The input data includes photovoltaic current data. The target power data includes photovoltaic active power data, energy storage charging power and energy storage discharging power. S2: Construct a data detection model for distributed photovoltaic and energy storage systems based on deep belief networks, and train the data detection model for distributed photovoltaic and energy storage systems using the training set to obtain a fine-tuned and optimized data detection model for distributed photovoltaic and energy storage systems. S3: The output weights of the finely tuned and optimized data detection model of the distributed photovoltaic and its energy storage system are optimized using the test set and the second-order oscillating firefly algorithm to obtain the optimized data detection model of the distributed photovoltaic and its energy storage system. S4: Using the validation set, verify the detection accuracy of the optimized data detection model for distributed photovoltaic and its energy storage system; S5: Input the real-time collected photovoltaic current data into the optimized data detection model of the distributed photovoltaic and its energy storage system to obtain the target power data.

[0007] Preferably, the input data also includes meteorological data and State of Storage (SOC).

[0008] Preferably, step S1 further includes preprocessing the historical data, which specifically includes the following steps: S11: Data cleaning: Use box plots to remove outliers and ensure data quality; S12: Feature standardization: Standardize the data using the mean-standard deviation normalization formula to eliminate the influence of dimensions; S13: Time alignment: Align data of different frequencies according to a fixed time window, and handle missing data by interpolation or padding to ensure consistency in the time dimension.

[0009] Preferably, step S2 includes the following steps: S21: Constructing a data detection model for distributed photovoltaic and energy storage systems based on deep belief networks; S22: Using the training set, the data detection model of the distributed photovoltaic and its energy storage system is subjected to unsupervised pre-training layer by layer using the restricted Boltzmann machine training method to extract photovoltaic current data features; S23: Supervised fine-tuning training is performed on the pre-trained data detection model of distributed photovoltaic and its energy storage system using photovoltaic current data characteristics and target power data to form a fine-tuned and optimized data detection model of distributed photovoltaic and its energy storage system.

[0010] Preferably, step S3 includes the following steps: S31: Initialize the relevant parameters of the second-order oscillatory firefly algorithm, including the number of fireflies, the maximum attraction coefficient, the light absorption coefficient, and the maximum number of iterations; S32: Initialize the space vectors of all fireflies; S33: Calculate the attraction of all fireflies; S34: Calculate the fitness of all fireflies; S35: Iteratively update the firefly position based on a second-order oscillation mechanism; S36: Determine if the current iteration count exceeds the maximum iteration count. If the current iteration count does not exceed the maximum iteration count, increment the iteration count by 1 and return to step S33; if the current iteration count exceeds the maximum iteration count, terminate the iteration. S37: After the iteration terminates, the firefly position vector with the smallest fitness value is selected. The selected firefly position vector is transformed into the optimal output weight of the deep belief network. The optimal output weight is then substituted into the fine-tuned and optimized data detection model of the distributed photovoltaic and energy storage system to obtain the optimized data detection model of the distributed photovoltaic and energy storage system.

[0011] Preferably, t The iteration of the ... i m The spatial vector of each firefly is: , in, This indicates that the elements of the spatial vector are real numbers; K N This represents the number of neurons in the last hidden layer of a deep belief network. N This represents the number of neurons in the output layer of a deep belief network. i It is the firefly's identification number; θ This represents the number of fireflies.

[0012] Preferably, the formula for calculating the attractiveness of fireflies is: , in, β 0 represents the maximum attraction coefficient; γ It is the light absorption coefficient; p imjn The distance between fireflies.

[0013] Preferably, the formula for calculating the fitness of fireflies is: , in, , For the actual output matrixY and fitted output matrix The Middle p Line 1 q The element values ​​of the column, M , C Represents the actual output matrix Y And the fitted output matrix has M OK C The column, the actual output matrix Y pq Target power data from the test set.

[0014] Preferably, the firefly position update formula is: , in, x im and x jn Fireflies i m and j n The corresponding coordinates; t This represents the number of iterations. β im，jn The attractiveness of fireflies; φ This is the balance coefficient; α Step size factor; random It is a random number within [0, 1].

[0015] Preferably, in the early stage of the second-order oscillating firefly algorithm, the balance coefficient is configured as follows: In the later iterations of the second-order oscillating firefly algorithm, the balance coefficients are configured as follows: The threshold for separating the early and late stages of iteration is the ratio of the current iteration number to the maximum iteration number, which is 50%-70%.

[0016] Compared with the prior art, the above-described technical solution of the present invention has the following advantages: 1. This invention utilizes the powerful nonlinear feature extraction and state reconstruction capabilities of Deep Belief Networks (DBNs), combined with optimization algorithms for precise tuning of key parameters, to construct a high-precision data detection model for distributed photovoltaic and energy storage systems. This enables the system to predict the critical operating states of devices or locations without directly installed sensor nodes based on measured data from a limited number of intelligent fusion terminals within a region. This significantly reduces the number of physical sensors and communication modules required at the hardware level, directly lowering equipment investment and subsequent maintenance costs.

[0017] 2. This invention does not treat photovoltaics and energy storage as independent entities. Instead, it integrates the Second-Order Oscillating Firefly Algorithm (SOO-FA) with DBN, treating the operational data of the coupled system as a whole for training and learning. This achieves unified and collaborative perception of the source-storage coordinated operation status. DBN can progressively mine the deep features and patterns hidden in massive heterogeneous operational data, effectively overcoming the shortcomings of traditional methods in superficial state identification of nonlinear, strongly coupled systems. The introduced SOO-FA is used to optimize the key parameters of DBN. This algorithm, combined with a second-order oscillation mechanism, effectively overcomes the problems of traditional optimization algorithms easily getting trapped in local optima and slow convergence speed. Through the rapid global optimization of SOO-FA, the constructed data detection model of the distributed photovoltaic and energy storage system is ensured to have both high accuracy and high computational efficiency, enabling rapid response to the time-varying state of the system. This meets the stringent real-time requirements of distributed energy systems for data acquisition and lays a solid foundation for subsequent real-time control. Attached Figure Description

[0018] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0019] Figure 1 This is a flowchart of the data detection method for the distributed photovoltaic and its energy storage system of the present invention.

[0020] Figure 2 This is a flowchart of the data preprocessing method in step S1 of the present invention.

[0021] Figure 3 This is a flowchart of step S2 of the present invention.

[0022] Figure 4 This is a flowchart of step S3 of the present invention. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0024] like Figure 1 As shown, the present invention provides a data detection method for distributed photovoltaic and its energy storage system, comprising the following steps: S1: Obtain historical data from distributed photovoltaic and its energy storage systems and divide it into training set, test set and verification set. The historical data includes input data and target power data. The input data includes photovoltaic current data. The target power data includes photovoltaic active power data, energy storage charging power and energy storage discharging power. S2: Construct a data detection model for distributed photovoltaic and energy storage systems based on Deep Belief Network (DBN), and train the data detection model for distributed photovoltaic and energy storage systems using the training set to obtain a fine-tuned and optimized data detection model for distributed photovoltaic and energy storage systems. S3: The output weights of the finely tuned and optimized data detection model of the distributed photovoltaic and its energy storage system are optimized using the test set and the second-order oscillating firefly algorithm to obtain the optimized data detection model of the distributed photovoltaic and its energy storage system. S4: Using the validation set, verify the detection accuracy of the optimized data detection model for distributed photovoltaic and its energy storage system; S5: Input the real-time collected photovoltaic current data into the optimized data detection model of the distributed photovoltaic and its energy storage system to obtain the target power data.

[0025] This invention treats photovoltaics and energy storage as a coupled system. By integrating SOO-FA and DBN, the operating data of this coupled system is used as a whole for training and learning, thereby achieving unified and collaborative perception of the photovoltaic-energy storage coordinated operation status.

[0026] In one specific embodiment, the historical data also includes meteorological data (temperature, irradiance), energy storage state (State of Charge, SOC), etc. Using the above data together with the photovoltaic current data as input data for model training can effectively avoid the problem of decreased prediction accuracy of the model under non-ideal conditions.

[0027] To ensure the accuracy of the predicted performance of the trained model, the time span of the historical data needs to be as long as possible to ensure a sufficient sample size. Preferably, the historical data in this application is data from the past two years.

[0028] like Figure 2 As shown, in a specific embodiment, step S1 further includes preprocessing the historical data, which specifically includes the following steps: S11: Data Cleaning: Use box plots to remove outliers and ensure data quality. For example, set upper and lower limits for data values ​​outside the interquartile range and treat them as outliers to be removed. S12: Feature Standardization: Use the mean-standard deviation normalization formula Data is standardized to eliminate the influence of units of measurement, among which, X norm The data is standardized after normalization; X This is the original data; μ The mean of the original data; σ The standard deviation of the original data; S13: Time alignment: Align data of different frequencies according to a fixed time window (e.g., 15 minutes or 1 hour), and handle missing data by interpolation or padding to ensure consistency in the time dimension.

[0029] DBN is composed of multiple Restricted Boltzmann Machines (RBMs) stacked together, including visible and hidden layers. It is a deep learning architecture based on a probabilistic generative model. It has powerful non-linear feature extraction capabilities, strong robustness in virtual acquisition, and can adapt to high-dimensional heterogeneous data scenarios.

[0030] like Figure 3 As shown, in a specific embodiment, step S2 includes the following steps: S21: Construct a data detection model for distributed photovoltaic and energy storage systems based on DBN. Specifically, this includes the following steps: S211: Initialize the relevant parameters of DBN, including the number of hidden layers, the number of neurons in the hidden layers, and the learning rate; S212: Initialize the relevant parameters of RBM, including the input layer weight matrix, the bias of the visible layer, the bias of the hidden layer, and the maximum number of training iterations. ,in, i Number the neurons in the visible layer a i For the bias of the visible layer, v i For visible layer status, j Number the neurons in the hidden layer b j For hidden layer bias, h j This is the hidden layer state. w ij For the visible layer i The first neuron and the hidden layer j The connection weights of each neuron.

[0031] S22: Using the training set, the data detection model of the distributed photovoltaic and its energy storage system is subjected to unsupervised pre-training layer by layer using the restricted Boltzmann machine training method to extract photovoltaic current data features. Specifically, this includes the following steps: S221: Input photovoltaic current data, calculate the conditional distribution of neurons in the current training layer and the previous hidden layer respectively, and generate Gibbs sampling points; S222: Use the maximum likelihood estimation method... The connection weights of the current hidden layer, the bias of the visible layer, and the bias of the hidden layer are updated based on the Gibbs sampling points. P ( v | h (in the hidden layer state) h Generate input samples v S223: Determine if the maximum number of iterations for a single layer has been reached. If not, repeat the calculation and update. If it has been reached, proceed to the next iteration. S224: Finally, determine if the current training layer is equal to the total number of hidden layers. If not, return to retrain the next layer until all hidden layers have been trained.

[0032] S23: Supervised fine-tuning training is performed on the pre-trained distributed photovoltaic and energy storage system data detection model using photovoltaic current data features and target power data to form a fine-tuned and optimized distributed photovoltaic and energy storage system data detection model. Specifically, this includes the following steps: S231: An output layer is added to the top layer (last hidden layer) of the pre-trained DBN. The number of neurons in the output layer is consistent with the dimension of the prediction target. The output layer is connected to the top layer of the DBN through output weights. S232: The photovoltaic current data from the pre-training stage is input into the pre-trained RBM network. Features are extracted layer by layer through forward propagation and finally passed to the output layer to calculate the predicted target power. S233: The error between the predicted target power and the target power is defined as the loss function. Backpropagation is used to update the weights and bias parameters of all layers. The weights include the inter-layer connection weights and output weights of the RBM. The bias parameters include the bias of the visible layer and the bias of the hidden layer. S234: Iteration is performed until the loss function convergence threshold is reached to complete the supervised fine-tuning.

[0033] DBN can uncover the deep features and patterns hidden in massive heterogeneous operational data layer by layer. This invention overcomes the shortcomings of traditional methods in identifying the state of nonlinear and strongly coupled systems by using DBN to build models.

[0034] The Second-Order Oscillating Firefly Algorithm (SOO-FA) is an intelligent optimization algorithm improved upon the traditional Firefly Algorithm (FA). Its core lies in introducing a second-order oscillation mechanism to optimize the firefly's position update strategy, thereby enhancing the algorithm's optimization capabilities in complex problems. It not only overcomes local optima and achieves global optima but also boasts a fast convergence speed, meeting real-time requirements, adapting to complex systems, and exhibiting strong robustness.

[0035] like Figure 4 As shown, in a specific embodiment, step S3 includes the following steps: S31: Initialize the relevant parameters of the second-order oscillating firefly algorithm, including the number of fireflies, the maximum attraction coefficient, the light absorption coefficient, and the maximum number of iterations; in a specific embodiment, the number of fireflies is 50-200, and the maximum number of iterations is 100-500. S32: Initialize the spatial vectors of all fireflies, where, t The iteration of the ... i m The spatial vector of the fireflies is ,in, This indicates that the elements of the spatial vector are real numbers; K N This represents the number of neurons in the last hidden layer of a deep belief network. N This represents the number of neurons in the output layer of a deep belief network. i It is the firefly's identification number; θ This represents the number of fireflies; S33: Calculate the attraction of all fireflies. ,in, β 0 represents the maximum attraction coefficient; γ It is the light absorption coefficient; p imjn The distance between fireflies; S34: Calculate the fitness of all fireflies; i m A firefly t The fitness in the next iteration is defined as: ,in, , For the actual output matrix Y and fitted output matrix The Middle p Line 1 q The element values ​​of the column, M , C Represents the actual output matrix YAnd the fitted output matrix has M OK C The column, the actual output matrix Y pq Target power data from the test set; S35: The firefly position is iteratively updated based on a second-order oscillation mechanism. The position update formula is: ,in, x im and x jn Fireflies i m and j n The corresponding coordinates; t This represents the number of iterations. β im，jn The attractiveness of fireflies; φ This is the balance coefficient; α Step size factor; random A random number within the range [0, 1]; S36: Determine if the current iteration count has exceeded [the limit]. t max If the current iteration count does not exceed the maximum iteration count, increment the iteration count by 1 and return to step S33; if the current iteration count exceeds the maximum iteration count, terminate the iteration. S37: After the iteration terminates, the firefly position vector with the smallest fitness value is selected, and the selected firefly position vector is transformed into the optimal output weight of DBN. The optimal output weight is then substituted into the fine-tuned and optimized data detection model of distributed photovoltaic and energy storage system to obtain the optimized data detection model of distributed photovoltaic and energy storage system.

[0036] In a specific embodiment, during step S35, when iteratively updating the firefly position based on the second-order oscillation mechanism, the balance coefficient is configured as follows during the early iteration phase of the second-order oscillatory firefly algorithm: This configuration allows individual fireflies to have a larger search step size in the early stages of iteration, enabling them to quickly traverse a wider weight space and cover more possible optimal solution regions. In the later stages of the second-order oscillating firefly algorithm, the balance coefficient is configured as follows: This setting allows individual fireflies to have higher search accuracy in the later stages of iteration, enabling fireflies to finely adjust weights within the optimal region and gradually approach the precise value of "minimum error", thus maximizing the power prediction accuracy of DBN.

[0037] Specifically, the threshold for separating the early and late stages of iteration is a ratio of 50%-70% of the current iteration count to the maximum iteration count, preferably 60%. t / tmax When it is less than 60%, it belongs to the early stage of iteration. t / t max When the percentage is ≥60%, it is considered to be in the later stage of the iteration.

[0038] This invention introduces the SOO-FA algorithm to optimize key parameters of DBN. This algorithm, combined with a second-order oscillation mechanism, effectively overcomes the problems of traditional optimization algorithms being prone to getting trapped in local optima and having slow convergence speed. Through the rapid global optimization of SOO-FA, the constructed DBN virtual acquisition model is ensured to have both high accuracy and high computational efficiency, enabling it to respond quickly to the time-varying state of the system. This meets the stringent real-time requirements of distributed energy systems for data acquisition and lays a solid foundation for subsequent real-time control.

[0039] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A data detection method for a distributed photovoltaic and its energy storage system, characterized in that, Includes the following steps: S1: Obtain historical data from distributed photovoltaic and its energy storage systems and divide it into training set, test set and verification set. The historical data includes input data and target power data. The input data includes photovoltaic current data. The target power data includes photovoltaic active power data, energy storage charging power and energy storage discharging power. S2: Construct a data detection model for distributed photovoltaic and energy storage systems based on deep belief networks, and train the data detection model for distributed photovoltaic and energy storage systems using the training set to obtain a fine-tuned and optimized data detection model for distributed photovoltaic and energy storage systems. S3: The output weights of the finely tuned and optimized data detection model of the distributed photovoltaic and its energy storage system are optimized using the test set and the second-order oscillating firefly algorithm to obtain the optimized data detection model of the distributed photovoltaic and its energy storage system. S4: Using the validation set, verify the detection accuracy of the optimized data detection model for distributed photovoltaic and its energy storage system; S5: Input the real-time collected photovoltaic current data into the optimized data detection model of the distributed photovoltaic and its energy storage system to obtain the target power data.

2. The data detection method for distributed photovoltaic and its energy storage system according to claim 1, characterized in that, The input data also includes meteorological data and energy storage status.

3. The data detection method for distributed photovoltaic and its energy storage system according to claim 1, characterized in that, Step S1 also includes preprocessing the historical data, which specifically includes the following steps: S11: Data cleaning: Use box plots to remove outliers and ensure data quality; S12: Feature standardization: Standardize the data using the mean-standard deviation normalization formula to eliminate the influence of dimensions; S13: Time alignment: Align data of different frequencies according to a fixed time window, and handle missing data by interpolation or padding to ensure consistency in the time dimension.

4. The data detection method for distributed photovoltaic and its energy storage system according to claim 1, characterized in that, Step S2 includes the following steps: S21: Constructing a data detection model for distributed photovoltaic and energy storage systems based on deep belief networks; S22: Using the training set, the data detection model of the distributed photovoltaic and its energy storage system is subjected to unsupervised pre-training layer by layer using the restricted Boltzmann machine training method to extract photovoltaic current data features; S23: Supervised fine-tuning training is performed on the pre-trained data detection model of distributed photovoltaic and its energy storage system using photovoltaic current data characteristics and target power data to form a fine-tuned and optimized data detection model of distributed photovoltaic and its energy storage system.

5. The data detection method for distributed photovoltaic and energy storage systems according to claim 1, characterized in that, Step S3 includes the following steps: S31: Initialize the relevant parameters of the second-order oscillatory firefly algorithm, including the number of fireflies, the maximum attraction coefficient, the light absorption coefficient, and the maximum number of iterations; S32: Initialize the space vectors of all fireflies; S33: Calculate the attraction of all fireflies; S34: Calculate the fitness of all fireflies; S35: Iteratively update the firefly position based on a second-order oscillation mechanism; S36: Determine if the current iteration count exceeds the maximum iteration count. If the current iteration count does not exceed the maximum iteration count, increment the iteration count by 1 and return to step S33; if the current iteration count exceeds the maximum iteration count, terminate the iteration. S37: After the iteration terminates, the firefly position vector with the smallest fitness value is selected. The selected firefly position vector is transformed into the optimal output weight of the deep belief network. The optimal output weight is then substituted into the fine-tuned and optimized data detection model of the distributed photovoltaic and energy storage system to obtain the optimized data detection model of the distributed photovoltaic and energy storage system.

6. The data detection method for distributed photovoltaic and its energy storage system according to claim 5, characterized in that, t The iteration of the ... i m The spatial vector of each firefly is: ， in, The elements of the spatial vector are real numbers. K N This represents the number of neurons in the last hidden layer of a deep belief network. N This represents the number of neurons in the output layer of a deep belief network. i It is the firefly's identification number; θ This represents the number of fireflies.

7. The data detection method for distributed photovoltaic and energy storage systems according to claim 5, characterized in that, The formula for calculating the attractiveness of fireflies is: ， in, β 0 represents the maximum attraction coefficient; γ It is the light absorption coefficient; p imjn The distance between fireflies.

8. The data detection method for distributed photovoltaic and its energy storage system according to claim 5, characterized in that, The formula for calculating the fitness of fireflies is: ， in, , For the actual output matrix Y and fitted output matrix The Middle p Line number q The element values ​​of the column, M , C Represents the actual output matrix Y And the fitted output matrix has M OK C The column, the actual output matrix Y pq Target power data from the test set.

9. The data detection method for distributed photovoltaic and energy storage systems according to claim 5, characterized in that, The formula for updating firefly positions is: ， in, x im and x jn Fireflies i m and j n The corresponding coordinates; t This represents the number of iterations. β im，jn The attractiveness of fireflies; φ This is the balance coefficient; α Step size factor; random It is a random number within [0, 1].

10. The data detection method for distributed photovoltaic and energy storage systems according to claim 9, characterized in that, In the early iterations of the second-order oscillating firefly algorithm, the balance coefficients are configured as follows: ; In the later iterations of the second-order oscillating firefly algorithm, the balance coefficient is configured as follows: The threshold for separating the early and late stages of iteration is the ratio of the current iteration number to the maximum iteration number, which is 50%-70%.

Images