Photovoltaic power generation performance self-adaptive improvement method and system based on multi-mode perception

Abstract

The invention relates to the field of photovoltaic power generation, and discloses a photovoltaic power generation performance self-adaptive improvement method and system based on multi-modal perception, and the method comprises the following steps: collecting illumination intensity, temperature, wind speed, electrical parameters and meteorological image data in real time through a multi-modal sensor network, constructing a dynamic four-dimensional tensor, and carrying out the real-time collection of the temperature, the wind speed, the electrical parameters and the meteorological image data; an incremental tensor decomposition algorithm is adopted to extract space-time correlation characteristics, and dynamic modeling of the state of the photovoltaic system is achieved; a photovoltaic array is divided into a plurality of sub-regions, collaborative optimization is performed by using a distributed optimization algorithm and a sparse communication protocol, and the power generation efficiency of a system is maximized under the condition of ensuring voltage and temperature constraints; the optimized parameters are mapped to a nonlinear manifold space, control input is adjusted through a gradient descent method based on the Riemannian manifold control law, and global optimal control over the system is achieved. The power generation efficiency of the photovoltaic system in a dynamic environment can be improved, the communication load is reduced, and the stability of the system is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and specifically to a method and system for adaptively improving the performance of photovoltaic power generation based on multi-modal perception. Background Art

[0002] As an important form of clean energy, photovoltaic power generation is widely used in residential, industrial and grid systems. Its basic system usually consists of a photovoltaic array, an inverter, a controller and a monitoring unit, etc., and is a key component in the modern energy structure. However, despite the continuous progress of photovoltaic power generation technology, existing photovoltaic systems still face many technical challenges, especially in terms of system operation efficiency and stability.

[0003] Currently, the operation of most photovoltaic systems still relies on traditional maximum power point tracking (MPPT) algorithms. These algorithms mainly rely on single-sensor data and ignore various environmental factors in the dynamic changes of the system, such as light intensity, temperature, wind speed, etc. This makes it difficult to achieve the optimal power generation efficiency of the system in complex environments. In addition, there is generally a problem of excessive communication overhead in existing technologies. Especially in large-scale photovoltaic systems, frequent data exchange between sub-regions often leads to overloading of the network, affecting the real-time performance and response speed of the system.

[0004] In addition, although manifold control theory has been applied to some optimization control problems, its application in photovoltaic systems is still in a relatively preliminary stage. Traditional optimization methods fail to effectively combine the non-linear characteristics of photovoltaic systems, and most methods cannot fully consider the influence of complex constraint conditions such as temperature and voltage, resulting in the inability to achieve the global optimal control effect. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a method and system for adaptively improving the performance of photovoltaic power generation based on multi-modal perception, which improves the power generation efficiency of the photovoltaic system in a dynamic environment, reduces the communication load and ensures the stability of the system.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for adaptively improving the performance of photovoltaic power generation based on multi-modal perception, comprising the following steps:

[0007] Collect multi-modal data of the photovoltaic power generation system, including light intensity, temperature, wind speed, electrical parameters and meteorological image data;

[0008] Construct the multi-modal data into a dynamic tensor, and extract spatio-temporal correlation features based on an incremental tensor decomposition algorithm;

[0009] Based on a distributed optimization algorithm, perform cooperative control on sub-regions of the photovoltaic array to generate global optimal parameters;

[0010] Map the optimized parameters to the non - linear manifold space and adjust the operating state of the photovoltaic system through the manifold control law.

[0011] Preferably, the multi - modal data includes:

[0012] Spatial dimension: The grid - based distribution data of the photovoltaic array;

[0013] Temporal dimension: The time - series data within a sliding time window;

[0014] Modal dimension: The light intensity matrix, the thermal imaging image, the wind speed sensor data;

[0015] Physical quantity dimension: Voltage, current, temperature.

[0016] Preferably, the step of constructing the multi - modal data into a dynamic tensor includes:

[0017] Organize the collected data of light intensity, temperature, humidity, wind speed and electrical parameters into a four - dimensional tensor where S represents the spatial dimension of the photovoltaic array, T represents the temporal dimension, M represents the modal dimension, and P represents the physical quantity dimension;

[0018] Map the data of the photovoltaic array at each moment to each layer of the tensor, generate a spatio - temporal feature matrix, and maintain the dynamics of time information.

[0019] Preferably, the step of extracting spatio - temporal correlation features based on the incremental tensor decomposition algorithm includes:

[0020] Decompose the dynamic tensor x t into the product of a core tensor and factor matrices U (n) (n = 1, 2, 3, 4), where the core tensor represents spatio - temporal features, and the factor matrices represent the features of each dimension, R n is the rank of the factor matrix, and N n is the size of the corresponding dimension;

[0021] Use the incremental tensor decomposition algorithm to update the core tensor and factor matrices by minimizing the reconstruction error, and the optimization objective function is:

[0022]

[0023] where, represents the Frobenius norm of the dynamic tensor reconstruction error, α t-k is the time decay factor, λ is the regularization parameter, is the factor matrix U(n) Frobenius norm;

[0024] Core tensor based on incremental update and factor matrix U (n) Extract spatio-temporal features, and then construct an environmental adaptability model for the photovoltaic array.

[0025] Preferably, the step of generating global optimal parameters by coordinately controlling sub-regions of the photovoltaic array based on the distributed optimization algorithm includes:

[0026] Divide the photovoltaic array into multiple sub-regions, each sub-region contains multiple photovoltaic units, and assign a local optimization objective to each sub-region;

[0027] Use the alternating direction multiplier method to coordinately optimize the sub-regions, and achieve global optimality by iteratively updating the local control variables;

[0028] Calculate the local optimization function, and generate global optimal control parameters according to the global optimization constraints.

[0029] Preferably, the optimization objective function for coordinately controlling sub-regions of the photovoltaic array based on the distributed optimization algorithm is:

[0030]

[0031] where f i (x i ) is the local optimization objective function of the i-th sub-region, x i is the control variable of the i-th sub-region, A i is the constraint matrix, b i is the constraint vector, and K is the total number of sub-regions;

[0032] In each iteration, the local optimization variable is updated by the following update formula:

[0033]

[0034] where is the control variable of the i-th sub-region in the k-th iteration, ρ is the step size factor, is the gradient of the local objective function, λ i is the Lagrange multiplier, is the transpose of the constraint matrix of the i-th sub-region;

[0035] Converge to the global optimal parameters through multiple iterative updates.

[0036] Preferably, the constraint conditions in the coordinated optimization process include:

[0037] Global voltage constraint: where V i is the voltage of the i-th sub-region, and V max is the maximum voltage of the system;

[0038] Global temperature constraint: where T i is the temperature of the i-th sub-region, and T th is the maximum temperature threshold of the system;

[0039] Communication constraint between sub-regions: Only the exchange of control information with adjacent sub-regions is allowed.

[0040] Preferably, the step of mapping the optimized parameters to the non-linear manifold space includes:

[0041] Model the current-voltage characteristics of the photovoltaic system as a Riemannian manifold where the manifold has a non-linear geometric structure;

[0042] By calculating the geodesic distance of the system, establish a control objective function and map it to the manifold space;

[0043] Based on the manifold control law, optimize the operating state of the photovoltaic system by adjusting the control variables.

[0044] Preferably, the step of adjusting the operating state of the photovoltaic system by the manifold control law includes:

[0045] Define the control objective function of the photovoltaic system as:

[0046]

[0047] where x(t) is the state of the system at time t, and x opt is the global optimal state, the geodesic distance on;

[0048] By optimizing the above control objective function, obtain the optimal control input u(t) and map it to the manifold to control the parameters of the system;

[0049] Use the gradient descent method to update the control input u(t) and keep the system in the optimized state.

[0050] The present invention also provides a photovoltaic power generation performance adaptive improvement system based on multi-modal perception, including:

[0051] A multi-modal data acquisition module for acquiring illumination, temperature, wind speed and electrical parameters;

[0052] A dynamic tensor decomposition module configured to perform incremental tensor decomposition and extract spatio-temporal features;

[0053] A distributed optimization module for collaboratively controlling sub-regions and generating globally optimal parameters;

[0054] A manifold embedding control module configured to map parameters to a non-linear manifold space and adjust the system operating state.

[0055] The present invention provides a method and system for adaptively improving the power generation performance of photovoltaic power generation based on multi-modal perception.

[0056] It has the following beneficial effects:

[0057] 1. By modeling the current-voltage characteristics of a photovoltaic array and adopting a distributed optimization algorithm, the present invention can significantly improve the power generation efficiency of a photovoltaic system and ensure optimal performance under different environmental conditions while satisfying voltage and temperature constraints.

[0058] 2. The present invention adopts a sparse communication protocol, which only allows similar sub-regions to exchange information according to the similarity of each sub-region, reducing unnecessary communication overhead, thus effectively reducing the communication load of the system and improving the communication efficiency of the entire system.

[0059] 3. Through the manifold control law in the Riemannian manifold space, the present invention can adjust the control input according to the dynamic state of the photovoltaic system to ensure that the system is always in an optimized state. This control method based on geodesic distance can make the system state more accurately approach the global optimum, thereby achieving more efficient energy conversion.

[0060] 4. The present invention can respond to environmental changes in real time and perform dynamic adjustment based on multi-modal perception data, enabling the photovoltaic system to adapt to different operating environments, maintain the best working state, and avoid performance degradation caused by environmental changes.

[0061] 5. By comprehensively considering various constraint conditions such as temperature and voltage and using a global optimization objective for coordinated adjustment, the present invention can effectively avoid potential risks such as excessive temperature while ensuring high-efficiency power generation of the photovoltaic system, enhancing the stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0063] Figure 2 It is a schematic diagram of the system architecture of the present invention.

[0064] Among them, 10 is a multi-modal data acquisition module; 20 is a dynamic tensor decomposition module; 30 is a distributed optimization module; 40 is a manifold embedding control module. DETAILED DESCRIPTION OF THE INVENTION

[0065] Next, in combination with the accompanying drawings of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0066] Please refer to the attached Figure 1 , the present invention provides a method for adaptively improving the performance of photovoltaic power generation based on multi-modal perception, aiming to integrate various environmental and electrical parameter data, and combine advanced mathematical and control algorithms to achieve the performance improvement of the photovoltaic power generation system. This method fully considers the multi-modal data characteristics in the operation of the photovoltaic system, and through accurate modeling and optimized control, achieves the purpose of improving the power generation efficiency of the photovoltaic system and extending the system life.

[0067] As Figure 1 shown, the method for adaptively improving the performance of photovoltaic power generation based on multi-modal perception may include the following steps:

[0068] S1. Collect multi-modal data of the photovoltaic power generation system;

[0069] S2. Construct the multi-modal data into a dynamic tensor, and extract spatio-temporal correlation features based on the incremental tensor decomposition algorithm;

[0070] S3. Based on the distributed optimization algorithm, perform cooperative control on the sub-regions of the photovoltaic array to generate global optimal parameters;

[0071] S4. Map the optimized parameters to the non-linear manifold space, and adjust the operation state of the photovoltaic system through the manifold control law.

[0072] The following is a detailed description of each step in the method of the present invention, and a comprehensive elaboration is made on the specific implementation principles, technical details and processes of each step.

[0073] For step S1, it includes the collection of multi-modal data of the photovoltaic power generation system, specifically involving the real-time collection and recording of light intensity, temperature, wind speed, electrical parameters and meteorological image data. These data provide the basis for subsequent optimization and control, and can ensure that the state information of the photovoltaic system is accurately reflected, thereby supporting efficient performance improvement.

[0074] First, the system includes a light intensity sensor, a temperature and humidity sensor, a wind speed sensor, a current and voltage sensor, and a meteorological image acquisition device. The light intensity sensor collects the direct light intensity and scattered light intensity in the area. The temperature and humidity sensor monitors the temperature of the photovoltaic module and the ambient temperature. The wind speed sensor measures wind speed data. The current and voltage sensor is used to collect the electrical parameters (such as voltage, current, etc.) of the system, and the meteorological image acquisition device records the meteorological image data of the environment (such as cloud changes, weather conditions).

[0075] The light intensity sensor can obtain the light intensity of the surrounding environment of the photovoltaic system in real time. The data acquisition time can be set according to actual needs, usually sampled every minute or every hour. Specifically, the light intensity data includes solar radiation intensity and scattered light intensity, reflecting the change of solar energy. The acquisition of light intensity data can be completed by an integrated photovoltaic sensor module.

[0076] The temperature and humidity sensor is responsible for measuring the temperature of the photovoltaic module and the surrounding environment in real time, and monitors each photovoltaic unit through a temperature sensor. The working principle of this sensor is based on the thermocouple or thermistor principle, and the data acquisition accuracy can be adjusted according to system requirements. The acquisition of temperature data is an important parameter for the influencing factors of photovoltaic power generation efficiency, because the increase in temperature will lead to a decrease in the conversion efficiency of photovoltaic cells.

[0077] The wind speed sensor measures the wind speed based on the ultrasonic principle or the mechanical principle, and monitors the influence of wind speed on heat dissipation and power generation efficiency. The wind speed data provides an effective reference for the change of ambient temperature. Especially when the temperature change of the photovoltaic system is large, the wind speed data can help to achieve more accurate optimization control.

[0078] The current and voltage sensor collects electrical parameters such as current, voltage and power, and transmits the data to the processing module through the corresponding data acquisition module. The current and voltage sensor can monitor the working state of the photovoltaic array in real time, especially under the conditions of system load or light change, helping to evaluate the power generation efficiency.

[0079] The meteorological image acquisition device, such as a high-resolution camera or an infrared thermal imaging device, is used to monitor the meteorological conditions around the photovoltaic system in real time. The meteorological image data helps to judge the impact of weather changes on the photovoltaic system, such as cloud cover or sudden weather conditions, so as to optimize the adjustment strategy of the photovoltaic system.

[0080] The multi-modal data includes sensor data in multiple dimensions. All the data is synchronously collected by the data acquisition module and transmitted to the central data processing module. The data transmission process uses wireless communication technology or wired communication technology to ensure that the system can transmit and process the collected data in real time. Through the synchronously transmitted multi-modal data, the system can comprehensively monitor each photovoltaic module and its working environment status, and then accurately predict the power generation efficiency of the system.

[0081] Next, the collected multi-modal data will be denoised and normalized through the data preprocessing module to ensure the accuracy and consistency of the data. The denoising process includes data filtering and outlier detection. The data normalization process includes normalizing or standardizing different modal data so that each parameter has the same dimension, which is convenient for subsequent modeling and analysis.

[0082] Finally, the processed multi-modal data is constructed into a dynamic tensor, which contains multiple dimensions such as time, space, modality, and physical quantity, and can comprehensively describe the operating state of the photovoltaic power generation system.

[0083] Through the acquisition and processing of this multi-modal data, the system can comprehensively understand the environment and working state of the photovoltaic array, providing support for subsequent steps, especially in the extraction of spatio-temporal features and the system optimization process, ensuring the accuracy and real-time nature of decision-making.

[0084] For step S2, in this embodiment, by constructing and decomposing the collected multi-modal data, the spatio-temporal correlation features of the photovoltaic array are extracted, and then support is provided for the optimal control of the photovoltaic system. Specifically, this step includes two main sub-processes: data construction and tensor decomposition. Through these two sub-processes, the potential patterns and dynamic changes in the operating state of the photovoltaic array can be effectively mined, providing a data basis for subsequent environmental adaptability modeling.

[0085] First, the collected data of light intensity, temperature, humidity, wind speed, and electrical parameters are organized into a four-dimensional dynamic tensor. The structure of this tensor is defined as:

[0086]

[0087] Among them,

[0088] S represents the spatial dimension of the photovoltaic array, representing different sub-regions or modules;

[0089] T represents the time dimension, indicating the timestamp of data collection, usually in minutes or hours;

[0090] M represents the modality dimension, including data types such as light intensity, temperature, and wind speed;

[0091] P represents the dimension of physical quantities, including electrical parameters such as voltage, current, and power.

[0092] After organizing these data into a four-dimensional tensor, it is convenient to map the data of each moment of the photovoltaic array. The data of the photovoltaic array at each moment form a slice of the tensor through mapping, and then a spatio-temporal feature matrix is generated. This spatio-temporal feature matrix not only maintains the dynamics of time information but also accurately reflects the diversity of the spatial dimension. Therefore, the four-dimensional tensor X t provides sufficient structured data for subsequent spatio-temporal feature extraction.

[0093] After the data construction is completed, further, in step S2, the spatio-temporal correlation features are extracted by an incremental tensor decomposition algorithm.

[0094] Specifically, Tucker decomposition is used to decompose the dynamic tensor χ t into components. Tucker decomposition decomposes a multi-dimensional tensor χ t into the product of a core tensor and multiple factor matrices U (n) , which is expressed as:

[0095]

[0096] where,

[0097] is the core tensor, representing the core information of spatio-temporal features;

[0098] are factor matrices, representing the features of each dimension of the tensor.

[0099] R n is the rank of the factor matrix, which determines the extraction dimension of the features;

[0100] N n is the size of the corresponding dimension, representing the number of elements in that dimension.

[0101] Through Tucker decomposition, the high-dimensional data can be effectively reduced in dimension, and the spatio-temporal features in the data can be extracted. These features can reflect the dynamic changes of the photovoltaic array at different times, spaces, and physical quantities, providing a basis for subsequent model training and optimization.

[0102] Next, an incremental tensor decomposition algorithm is used to optimize the result of Tucker decomposition. This algorithm updates the core tensor and the factor matrix U (n) by minimizing the reconstruction error. The optimization objective function is expressed as:

[0103]

[0104] Wherein:

[0105] is the Frobenius norm of the tensor reconstruction error, representing the difference between the decomposition result and the actual data;

[0106] α t-k is the time decay factor, which is used to adjust the influence weight of data at different time points on the model, ensuring that the contribution of historical data to the model gradually decreases;

[0107] λ is the regularization parameter to prevent overfitting;

[0108] is the Frobenius norm of the factor matrix U (n) to control the complexity of the matrix.

[0109] By minimizing the objective function, the incremental tensor decomposition algorithm can continuously update the core tensor and the factor matrix U (n) , and extract the spatio-temporal correlation features in the photovoltaic array data through the optimization process. The incremental update method of this algorithm enables the model to quickly adapt to new data and optimize the extraction of spatio-temporal features in real time in an environment where the data is constantly changing.

[0110] Finally, based on the spatio-temporal features extracted from the incrementally updated core tensor and the factor matrix U (n) provide data support for constructing the environmental adaptability model of the photovoltaic array.

[0111] For step S3, in this embodiment, the sub-regions of the photovoltaic array are collaboratively controlled based on the distributed optimization algorithm to generate the global optimal parameters. This step divides the photovoltaic array, sets local optimization objectives, and uses the alternating direction method of multipliers (ADMM) for iterative optimization to finally determine the globally optimal control parameters. This process can improve the power generation efficiency of the photovoltaic array and ensure the stable operation of the system.

[0112] First, the photovoltaic array is divided into multiple sub-regions. Each sub-region contains multiple photovoltaic units, and these photovoltaic units can be grouped according to the similarity of physical location or electrical parameters. Each sub-region assigns a local optimization objective to a specific control module to ensure the optimization of the operating state of each sub-region, and finally realizes the overall optimization through global collaborative control.

[0113] In this embodiment, the goal of the distributed optimization algorithm is to achieve the global optimum by minimizing the local optimization function of each sub-region. The specific optimization objective function is:

[0114]

[0115] Among them,

[0116] f i (x i ) is the local optimization objective function of the $i$-th sub-region, representing the performance optimization objective of this sub-region;

[0117] x i is the control variable of the $i$-th sub-region, used to adjust the working state of the photovoltaic unit;

[0118] A i is the constraint matrix, describing the constraint conditions of this sub-region;

[0119] b i is the constraint vector, representing the specific values of each constraint condition;

[0120] K is the total number of sub-regions.

[0121] The purpose of this objective function is to minimize the local optimization objectives of all sub-regions while ensuring that the constraint conditions of each sub-region are satisfied. The constraint conditions involve voltage, temperature, and communication limitations between sub-regions.

[0122] During the optimization process, the local optimization variables are updated iteratively each time to promote the optimization of the global system. The specific update formula is:

[0123]

[0124] Among them,

[0125] is the control variable of the $i$-th sub-region in the $k$-th iteration;

[0126] ρ is the step size factor, controlling the amplitude of each update;

[0127] is the local objective function 's gradient, representing the direction and rate of change of the objective function;

[0128] λ i is the Lagrange multiplier of the $i$-th sub-region, used to balance the constraint conditions and the optimization objective;

[0129] is the transpose of the constraint matrix of the $i$-th sub-region, used to adjust the control variable under the constraint conditions.

[0130] Through multiple iterations, the local control variables gradually approach the global optimal solution, and finally the optimal control parameters of all sub-regions are obtained.

[0131] During the collaborative optimization process, the constraint conditions are crucial to the optimization process, mainly including the following aspects:

[0132] Global voltage constraint:

[0133]

[0134] where V i is the voltage of the i-th sub-region, and V max is the maximum voltage of the system. This constraint ensures that the voltage of the entire PV system is within a safe range.

[0135] Global temperature constraint:

[0136]

[0137] where T i is the temperature of the i-th sub-region, and T th is the maximum temperature threshold of the system. This constraint is used to ensure that the system does not overheat and avoid a reduction in system efficiency or damage caused by excessive temperature.

[0138] Communication constraint between sub-regions:

[0139] This constraint requires that sub-regions can only exchange control information with adjacent sub-regions, ensuring the efficiency of information transfer and reducing communication costs. This constraint helps to improve the efficiency of the optimization algorithm and the real-time response ability of the system.

[0140] To further reduce communication overhead, a sparse communication protocol is adopted. By calculating the similarity of the feature matrices between sub-regions (for example, using cosine similarity), only sub-regions with a relatively high similarity are allowed to exchange information. The cosine similarity calculation formula is:

[0141]

[0142] If the similarity is greater than a preset threshold θ, then the two sub-regions are allowed to exchange intermediate variables.

[0143] Through these constraint conditions, the system can coordinate the operating states of each sub-region while ensuring safety and stability, and achieve global optimality.

[0144] In actual implementation, the Alternating Direction Method of Multipliers (ADMM) is applied to update the local optimization variables. The advantage of the ADMM method is that it can decompose a large-scale optimization problem into multiple smaller sub-problems and achieve global optimality through iterative solution. In each iteration, by updating the control variables of each sub-region, the coordination and synchronization between the various parts of the system are ensured, and finally a global optimal solution that satisfies all constraint conditions is obtained.

[0145] Finally, after multiple iterative updates, the control variables of all sub-regions converge to the global optimal value, thus obtaining the final global optimal control parameters. This process ensures the best performance of the photovoltaic array under different environmental conditions and enables dynamic adjustment according to real-time data, improving the adaptability and reliability of the system.

[0146] For step S4, in this embodiment, the optimized parameters are mapped to the non-linear manifold space, and the operating state of the photovoltaic system is adjusted through the manifold control law to achieve the optimized control of the system. The specific implementation process includes modeling the current-voltage characteristics of the photovoltaic system as a Riemannian manifold, establishing and mapping the control objective function to the manifold space, and adjusting the control input through the manifold control law to optimize the system operating state.

[0147] First, the current-voltage characteristics of the photovoltaic system are modeled as a Riemannian manifold. The non-linear geometric structure of the manifold can better reflect the dynamic characteristics of the system. Assume the state space of the photovoltaic system is where each state represents the operating state of the photovoltaic system, such as parameters like voltage, current, temperature, etc. The choice of the manifold is based on the physical characteristics of the system, and the geometric properties of the manifold can effectively capture the non-linear relationship of the current-voltage characteristics.

[0148] Then, the control objective function is established by calculating the geodesic distance of the system. The geodesic is the shortest path between two points on the manifold, and the difference between system states can be measured through the metric structure of the manifold. Specifically, the geodesic distance of the system is:

[0149]

[0150] where,

[0151] x(t) is the state of the system at time t;

[0152] x opt is the global optimal state;

[0153] is the geodesic distance between two points on the manifold, representing the difference between the system state and the optimal state.

[0154] Based on this geodesic distance, the control objective function is constructed:

[0155]

[0156] The significance of this control objective function is that minimizing this objective function can make the state x(t) of the system approach the global optimal state x opt , thus achieving the optimization of the system performance.

[0157] Next, based on the manifold control law, the operating state of the photovoltaic system is optimized by adjusting the control input. The control objective is to minimize the above control objective function. To achieve this optimization goal, first define the control input u(t), and obtain the optimal control input u opt (t) by optimizing the control objective function. In the manifold space, the control input u(t) needs to be mapped onto the manifold so that the system state evolves along the shortest path of the manifold to approach the global optimal state.

[0158] The optimization process of the control input is achieved by the gradient descent method. Specifically, the gradient descent method minimizes the control objective function J by iteratively updating the control input u(t). The formula for updating the control input is:

[0159]

[0160] where,

[0161] u(t) (k) is the control input in the k-th iteration;

[0162] α is the learning rate, which controls the magnitude of each update;

[0163] is the gradient of the control objective function J(u(t)) with respect to the control input u(t), representing the change direction of the control input.

[0164] Through continuous iteration, the control input of the system gradually approaches the optimal control input u opt (t), so that the system state tends to the global optimal state.

[0165] In addition, to ensure that the system always remains in the optimized state, the system uses the gradient descent method to update the control input in each iteration, and maintains the system state close to the global optimal state x opt by adjusting the control input. This process can effectively improve the performance of the system, ensure the maximum power generation efficiency of the photovoltaic array, and avoid the influence of adverse factors such as overheating temperature.

[0166] In this process, the optimized parameters are mapped to the non-linear manifold space through the manifold control law, and the control input is adjusted through the manifold control law, finally realizing the optimized operating state of the photovoltaic system. This process can effectively improve the power generation efficiency of the system, while ensuring the stability and adaptability of the system.

[0167] Finally, the optimization process can achieve the following effects: by optimizing the control input, the state x(t) of the system gradually approaches the global optimal state x opt , thereby maximizing the power generation efficiency of the photovoltaic system and ensuring that the constraint conditions such as voltage and temperature are satisfied.

[0168] Generally speaking, the present invention first collects the light intensity, temperature, wind speed, electrical parameters and meteorological image data of the photovoltaic system in real time through a multi-modal sensor network, constructs a dynamic four-dimensional tensor to fuse spatio-temporal features; then adopts an incremental tensor decomposition algorithm to extract spatio-temporal correlation features, and realizes the dynamic update of the core tensor through Tucker decomposition and regularization optimization; further divides the photovoltaic array into sub-regions, and uses the distributed alternating direction method of multipliers (ADMM) combined with a sparse communication protocol for collaborative optimization to maximize the power generation efficiency and meet the global voltage and temperature constraints; finally maps the optimized parameters to the non-linear manifold space, constructs a control objective function based on the geometric properties of the Riemannian manifold, and adjusts the control input through the gradient descent method to make the system converge to the optimal state along the geodesic. This method can achieve efficient power generation, low communication load and stable operation of the photovoltaic system in a dynamic environment.

[0169] The adaptive improvement system for photovoltaic power generation performance based on multi-modal perception described below can be correspondingly referred to the method for adaptive improvement of photovoltaic power generation performance based on multi-modal perception described above.

[0170] Please refer to the attached Figure 2 , the present invention also provides an adaptive improvement system for photovoltaic power generation performance based on multi-modal perception, including:

[0171] A multi-modal data acquisition module 10 for acquiring light, temperature, wind speed and electrical parameters;

[0172] A dynamic tensor decomposition module 20 configured to perform incremental tensor decomposition and extract spatio-temporal features;

[0173] A distributed optimization module 30 for performing collaborative control on sub-regions and generating global optimal parameters;

[0174] A manifold embedding control module 40 configured to map parameters to a non-linear manifold space and adjust the operating state of the system.

[0175] The system of this embodiment can be used to execute the above method embodiment, and its principle and technical effect are similar, so it will not be elaborated here.

[0176] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for adaptively improving photovoltaic power generation performance based on multimodal perception, characterized in that: The following steps are involved: Collect multimodal data of photovoltaic power generation systems, including light intensity, temperature, wind speed, electrical parameters and meteorological image data; Constructing the multimodal data into a dynamic tensor, and extracting spatiotemporal correlation features based on an incremental tensor decomposition algorithm; Based on the distributed optimization algorithm, the sub-areas of the photovoltaic array are coordinated to generate the global optimal parameters; The optimized parameters are mapped to the nonlinear manifold space, and the operating state of the photovoltaic system is adjusted through the manifold control law.

2. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 1 is characterized in that: The multimodal data includes: Spatial dimension: grid distribution data of photovoltaic arrays; Time dimension: time series data within a sliding time window; Modal dimensions: light intensity matrix, thermal imaging image, wind speed sensor data; Physical quantity dimensions: voltage, current, temperature.

3. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 1 is characterized in that: The step of constructing the multimodal data into a dynamic tensor comprises: Organize the collected light intensity, temperature, humidity, wind speed and electrical parameter data into a four-dimensional tensor Where S represents the spatial dimension of the photovoltaic array, T represents the time dimension, M represents the modal dimension, and P represents the physical quantity dimension; The photovoltaic array data at each moment is mapped to each layer of the tensor to generate a spatiotemporal feature matrix and maintain the dynamic nature of the time information.

4. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 3 is characterized in that: The step of extracting spatiotemporal correlation features based on the incremental tensor decomposition algorithm comprises: The dynamic tensor is decomposed by Tucker Decomposition into core tensors And the factor matrix U (n) (n=1,2,3,4), where the core tensor Represents spatiotemporal features, factor matrix Represents the characteristics of each dimension, R n is the rank of the factor matrix, N n is the size of the corresponding dimension; The incremental tensor decomposition algorithm is used to update the core tensor and factor matrix by minimizing the reconstruction error. The optimization objective function is: in, represents the Frobenius norm of the dynamic tensor reconstruction error, α t-k is the time decay factor, λ is the regularization parameter, is the factor matrix U (n) The Frobenius norm of ; Core tensors based on incremental updates And the factor matrix U (n) Extract spatiotemporal features and then construct an environmental adaptability model of the photovoltaic array.

5. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 1, characterized in that: The step of collaboratively controlling the sub-areas of the photovoltaic array based on the distributed optimization algorithm to generate global optimal parameters includes: Divide the photovoltaic array into multiple sub-regions, each sub-region contains multiple photovoltaic units, and assign a local optimization target to each sub-region; The sub-regions are collaboratively optimized using the alternating direction multiplier method, and the local control variables are iteratively updated to achieve the global optimum. Calculate the local optimization function and generate the global optimal control parameters according to the global optimization constraints.

6. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 5 is characterized in that: The optimization objective function for the coordinated control of the sub-areas of the photovoltaic array based on the distributed optimization algorithm is: Among them, f i (x i ) is the local optimization objective function of the i-th sub-region, x i is the control variable of the ith sub-region, A i is the constraint matrix, b i is the constraint vector, K is the total number of sub-regions; In each iteration, the local optimization variables are updated by the following update formula: in, is the control variable of the ith sub-region in the kth iteration, ρ is the step size factor, is the gradient of the local objective function, λ i is the Lagrange multiplier, is the transpose of the constraint matrix of the i-th sub-region; Through multiple iterative updates, the global optimal parameters are converged.

7. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 6 is characterized in that: The constraints in the collaborative optimization process include: Global voltage constraints: Where V i is the voltage of the ith sub-region, V max is the maximum voltage of the system; Global temperature constraint: Where T i is the temperature of the ith sub-region, T th is the maximum temperature threshold of the system; Communication constraints between sub-regions: Control information is only allowed to be exchanged with adjacent sub-regions.

8. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 1, characterized in that: The step of mapping the optimized parameters to the nonlinear manifold space comprises: Modeling the current-voltage characteristics of a photovoltaic system as a Riemann manifold The manifold Has nonlinear geometric structure; By calculating the geodesic distance of the system, the control objective function is established and mapped to the manifold space; Based on the manifold control law, the operating state of the photovoltaic system is optimized by adjusting the control variables.

9. The method for adaptively improving photovoltaic power generation performance based on multimodal perception according to claim 8, characterized in that: The step of adjusting the operating state of the photovoltaic system by the manifold control law comprises: The control objective function of the photovoltaic system is defined as: Among them, x(t) is the state of the system at time t, x opt is the global optimal state, The geodesic distance on ; By optimizing the above control objective function, the optimal control input u(t) is obtained and mapped to the manifold to control the system parameters; Use gradient descent to update the control input u(t) and keep the system in an optimized state.

10. A photovoltaic power generation performance adaptive improvement system based on multimodal perception, used to execute the photovoltaic power generation performance adaptive improvement method based on multimodal perception as claimed in any one of claims 1 to 9, characterized in that: include: Multimodal data acquisition module for obtaining light, temperature, wind speed and electrical parameters; a dynamic tensor decomposition module configured to perform incremental tensor decomposition and extract spatiotemporal features; Distributed optimization module, used to coordinate the sub-areas and generate global optimal parameters; The manifold embedding control module is configured to map parameters to the nonlinear manifold space and adjust the system operation state.

Images