Photovoltaic array adjustment method and device, electronic equipment and storage medium

Abstract

The invention provides a photovoltaic array adjustment method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining an operation parameter and an electric energy parameter of a photovoltaic array; predicting a health state of the photovoltaic array based on the operating parameters; determining the electric energy quality of the photovoltaic array based on the electric energy parameters; obtaining a photovoltaic array adjustment instruction determined based on the health state and the electric energy quality; and executing the photovoltaic array adjustment instruction, so that the generated power of the adjusted photovoltaic array does not exceed the maximum operation power corresponding to the health state. According to the method, the problems that the photovoltaic array system is low in operation efficiency and high in equipment safety risk are solved.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation and intelligent control technology, and more specifically, to a photovoltaic array adjustment method, device, electronic device and storage medium. Background Technology

[0002] Photovoltaic power generation, as an important form of clean energy, has been widely applied in various power generation scenarios. Photovoltaic arrays convert solar energy into electrical energy, providing power to the grid. In actual operation, the output efficiency of photovoltaic arrays is affected by various factors such as the sun's position and weather conditions; therefore, their operating status needs to be adjusted accordingly to improve power generation efficiency.

[0003] Currently, adjustments to photovoltaic arrays primarily rely on fixed-angle installations or periodic angle adjustments based on simple timing. While these methods are simple to implement, they cannot adapt to real-time environmental changes, especially under complex conditions such as sudden weather changes or localized shading, making it difficult to maintain optimal power generation. Furthermore, existing adjustment methods typically consider only a single optimization objective, lacking a comprehensive assessment of the overall system operation, resulting in power generation efficiency failing to reach ideal levels. Summary of the Invention

[0004] In view of this, embodiments of this application provide a photovoltaic array adjustment method, apparatus, electronic device, and storage medium to solve the problems of low operating efficiency and high equipment safety risks in photovoltaic array systems.

[0005] In a first aspect, embodiments of this application provide a photovoltaic array adjustment method, the method comprising: Obtain the operating parameters and electrical parameters of the photovoltaic array; Predict the health status of the photovoltaic array based on the operating parameters; The power quality of the photovoltaic array is determined based on the aforementioned power parameters; Obtain photovoltaic array adjustment instructions determined based on the health status and the power quality; the photovoltaic array adjustment instructions include: tilt angle adjustment operation and / or operating voltage adjustment operation; The photovoltaic array adjustment command is executed to ensure that the power generation of the adjusted photovoltaic array does not exceed the maximum operating power corresponding to the healthy state.

[0006] In one feasible implementation, predicting the health status of the photovoltaic array based on the operating parameters includes: Based on the operating parameters, a Markov chain model with dynamic weighting factors is used to predict the probability that the photovoltaic array will be in different health levels within a preset time period in the future. The probabilities of the different health levels are determined as the health status of the photovoltaic array.

[0007] In one feasible implementation, determining the power quality of the photovoltaic array based on the power parameters includes: Based on the electrical energy parameters, the original electrical energy parameter characteristics are generated; The original electrical energy parameter features are input into a multi-scale decomposition network to obtain the trend component, periodic component, and residual component of the photovoltaic array. The multi-scale decomposition network decomposes the original electrical energy parameter features through parallel long-period recurrent neural networks, medium-period recurrent neural networks, and time-series convolutional networks to output the trend component, periodic component, and residual component accordingly. Based on the trend component, the periodic component, and the residual component, the power quality reconstruction characteristics are obtained. Based on the difference between the power quality reconstruction features and the original power parameter features, an anomaly score for power quality is determined; Based on the comparison between the anomaly score and the dynamically determined quality threshold, it is determined whether the power quality is abnormal.

[0008] In one feasible implementation, the dynamic determination step of the quality threshold includes: Based on historical power quality anomaly scores, the quality threshold is adaptively updated using an exponential smoothing method.

[0009] In one feasible implementation, obtaining photovoltaic array adjustment instructions determined based on the health status and the power quality includes: The integrated parameters obtained by fusing the health status and the power quality are input into a preset two-layer decision model; the first layer of the two-layer decision model is used to output the maximum operating power based on the health status information in the integrated parameters; the second layer of the two-layer decision model is used to output the tilt adjustment operation and / or the operating voltage adjustment operation based on the integrated parameters and the maximum operating power. Obtain the photovoltaic array adjustment command output by the two-layer decision model.

[0010] In one feasible implementation, the two-layer decision model is a reinforcement learning model trained based on the Q-learning algorithm; When training the reinforcement learning model, a reward value is calculated using a reward function, wherein the calculation of the reward value includes at least one of the following: Increase reward value in response to increased power generation capacity; In response to power quality being deemed abnormal, the reward value is reduced; In response to an increase in the adjustment range of the tilt angle adjustment operation and / or the operating voltage adjustment operation, the bonus value is reduced.

[0011] In one feasible implementation, the method further includes: In response to being in photovoltaic array adjustment mode, the parameters of the two-level decision model remain unchanged; In response to switching to offline maintenance mode, the parameters of the two-level decision model are updated until convergence.

[0012] Secondly, embodiments of this application also provide a photovoltaic array adjustment device, the device comprising: The parameter acquisition module is used to acquire the operating parameters and electrical energy parameters of the photovoltaic array; The prediction module is used to predict the health status of the photovoltaic array based on the operating parameters; A determining module is used to determine the power quality of the photovoltaic array based on the power parameters; The instruction acquisition module is used to acquire photovoltaic array adjustment instructions determined based on the health status and the power quality; the photovoltaic array adjustment instructions include: tilt angle adjustment operation and / or operating voltage adjustment operation; The adjustment module is used to execute the photovoltaic array adjustment command so that the power generation of the adjusted photovoltaic array does not exceed the maximum operating power corresponding to the healthy state.

[0013] In one feasible implementation, the prediction module is configured to predict the health status of the photovoltaic array based on the operating parameters, for the purpose of: Based on the operating parameters, a Markov chain model with dynamic weighting factors is used to predict the probability that the photovoltaic array will be in different health levels within a preset time period in the future. The probabilities of the different health levels are determined as the health status of the photovoltaic array.

[0014] In one feasible implementation, a determining module is used to determine the power quality of the photovoltaic array based on the power parameters, for the following purposes: Based on the electrical energy parameters, the original electrical energy parameter characteristics are generated; The original electrical energy parameter features are input into a multi-scale decomposition network to obtain the trend component, periodic component, and residual component of the photovoltaic array. The multi-scale decomposition network decomposes the original electrical energy parameter features through parallel long-period recurrent neural networks, medium-period recurrent neural networks, and time-series convolutional networks to output the trend component, periodic component, and residual component accordingly. Based on the trend component, the periodic component, and the residual component, the power quality reconstruction characteristics are obtained. Based on the difference between the power quality reconstruction features and the original power parameter features, an anomaly score for power quality is determined; Based on the comparison between the anomaly score and the dynamically determined quality threshold, it is determined whether the power quality is abnormal.

[0015] In one feasible implementation, the dynamic determination step of the quality threshold includes: Based on historical power quality anomaly scores, the quality threshold is adaptively updated using an exponential smoothing method.

[0016] In one feasible implementation, the instruction acquisition module is configured to acquire photovoltaic array adjustment instructions determined based on the health status and the power quality, for the purpose of: The integrated parameters obtained by fusing the health status and the power quality are input into a preset two-layer decision model; the first layer of the two-layer decision model is used to output the maximum operating power based on the health status information in the integrated parameters; the second layer of the two-layer decision model is used to output the tilt adjustment operation and / or the operating voltage adjustment operation based on the integrated parameters and the maximum operating power. Obtain the photovoltaic array adjustment command output by the two-layer decision model.

[0017] In one feasible implementation, the two-layer decision model is a reinforcement learning model trained based on the Q-learning algorithm; When training the reinforcement learning model, a reward value is calculated using a reward function, wherein the calculation of the reward value includes at least one of the following: Increase reward value in response to increased power generation capacity; In response to power quality being deemed abnormal, the reward value is reduced; In response to an increase in the adjustment range of the tilt angle adjustment operation and / or the operating voltage adjustment operation, the bonus value is reduced.

[0018] In one feasible implementation, the device further includes: A holding module is used to maintain the parameters of the two-layer decision model unchanged in response to being in photovoltaic array adjustment mode; An update module is used to update the parameters of the two-level decision model until convergence in response to switching to offline maintenance mode.

[0019] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the photovoltaic array adjustment method as described in any one of the first aspects.

[0020] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the photovoltaic array adjustment method as described in any one of the first aspects.

[0021] This application provides a photovoltaic array adjustment method, apparatus, electronic device, and storage medium. By synchronously acquiring the operating parameters and power parameters of the photovoltaic array, predicting the health status based on the operating parameters, evaluating the power quality based on the power parameters, and generating control commands including tilt angle adjustment and / or operating voltage adjustment, compared with existing technologies that only consider a single factor or adopt a fixed adjustment strategy, the following beneficial effects can be achieved: under the premise of ensuring that the actual power generation of the photovoltaic array does not exceed the maximum operating power allowed by its health status, the power generation efficiency is significantly improved, while effectively ensuring the safety of equipment operation and the power quality of the power grid.

[0022] Furthermore, since the power generation of the adjusted photovoltaic array does not exceed the maximum operating power allowed under its healthy state, this solution can ensure that the photovoltaic array will not accelerate its aging or damage due to overload operation, thereby significantly improving the operational reliability and service life of the equipment.

[0023] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart of a photovoltaic array adjustment method provided in an embodiment of this application is shown.

[0026] Figure 2 A flowchart of another photovoltaic array adjustment method provided in an embodiment of this application is shown.

[0027] Figure 3 A flowchart of another photovoltaic array adjustment method provided in an embodiment of this application is shown.

[0028] Figure 4 A schematic diagram of a photovoltaic array adjustment device provided in an embodiment of this application is shown.

[0029] Figure 5A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0031] Photovoltaic power generation, as an important form of clean energy, has been widely applied in various power generation scenarios. Photovoltaic arrays convert solar energy into electrical energy through the photoelectric conversion effect, providing clean energy to the power system. In actual operation, the output efficiency of photovoltaic arrays is affected by multiple factors such as the sun's position, weather conditions, and ambient temperature, requiring timely adjustments based on operating status to improve power generation efficiency.

[0032] Currently, there are two main methods for adjusting the operation of photovoltaic arrays: one is fixed-angle installation, and the other is periodic angle adjustment based on preset time sequences. While these methods are simple in structure and easy to implement, they have significant limitations. First, fixed installations cannot adapt to changes in the sun's position during different seasons and times of day; second, time-series-based adjustments struggle to cope with sudden weather changes and complex conditions such as localized shading. More importantly, existing adjustment methods typically consider only a single optimization objective, lacking a comprehensive understanding and judgment of the system's operating status. This results in photovoltaic arrays operating in suboptimal conditions for extended periods, failing to achieve ideal power generation efficiency.

[0033] It is particularly important to point out that existing technical solutions generally suffer from the following shortcomings: Problem 1: The adjustment strategy is too simplistic, only considering one of the angle adjustment or voltage adjustment, lacking synergistic optimization.

[0034] Problem 2: The correlation between operating parameters (such as temperature and irradiance) and electrical energy parameters (such as power factor and harmonics) has not been fully considered.

[0035] Problem 3: Lack of a collaborative analysis and comprehensive decision-making mechanism for operating status and power quality.

[0036] Therefore, there is an urgent need in this field for a technical solution that can simultaneously consider operating parameters and electrical parameters, and achieve continuous optimization of photovoltaic array power generation efficiency by coordinating the adjustment of tilt angle and operating voltage.

[0037] This application provides a photovoltaic array adjustment method, apparatus, electronic device, and storage medium, which are described below through embodiments.

[0038] To facilitate understanding of this embodiment, a photovoltaic array adjustment method disclosed in this application will first be described in detail. For example... Figure 1 The flowchart of a photovoltaic array adjustment method shown includes the following steps: Step 101: Obtain the operating parameters and power parameters of the photovoltaic array.

[0039] In this step, the operating parameters refer to parameters that reflect the physical state and operating performance of the photovoltaic array itself. These parameters may include, but are not limited to, the temperature of the photovoltaic module, output power, insulation resistance, open-circuit voltage, and short-circuit current. The electrical energy parameters refer to parameters that characterize the output electrical energy characteristics of the photovoltaic array. These parameters may include, but are not limited to, output voltage, output current, frequency, harmonic content, and power factor.

[0040] It should be noted that the operating parameters and power parameters can be obtained through corresponding sensors deployed in the photovoltaic array and its grid-connected system, or read from the monitoring system of the photovoltaic inverter. These parameters can be obtained in real time or according to a preset sampling period.

[0041] Step 102: Predict the health status of the photovoltaic array based on the operating parameters.

[0042] In this step, the health status refers to the assessment result of the overall operating condition of the photovoltaic array, reflecting its performance retention level and potential failure risks. By analyzing and processing operating parameters, the health status of the photovoltaic array can be predicted. Specific prediction methods may include statistical analysis based on historical data, simulation calculations based on physical models, or intelligent evaluation based on machine learning algorithms.

[0043] Step 103: Determine the power quality of the photovoltaic array based on the power parameters.

[0044] In this step, the power quality refers to the evaluation result of the output power characteristics of the photovoltaic array, reflecting the degree to which it meets power quality standards. The power quality status can be determined through analysis and processing of power parameters. Specific determination methods may include parameter threshold comparison, waveform feature analysis, and spectrum analysis.

[0045] Step 104: Obtain photovoltaic array adjustment instructions determined based on the health status and the power quality; the photovoltaic array adjustment instructions include: tilt angle adjustment operation and / or operating voltage adjustment operation.

[0046] In this step, the photovoltaic array adjustment command is generated through a preset decision-making mechanism based on the health status and power quality information obtained in the preceding steps. The tilt angle adjustment operation refers to the control command that changes the tilt angle of the photovoltaic panel relative to the horizontal plane; the operating voltage adjustment operation refers to the control command that changes the operating voltage of the photovoltaic array. These adjustment operations can be executed individually or in combination.

[0047] Step 105: Execute the photovoltaic array adjustment command to ensure that the power generation of the adjusted photovoltaic array does not exceed the maximum operating power corresponding to the healthy state.

[0048] In this step, the adjustment instructions obtained in the preceding steps are executed to optimize the operating status of the photovoltaic array. Specifically, this adjustment must ensure that the actual power generation of the adjusted photovoltaic array does not exceed the safe operating limit allowed by its current health condition, thereby improving power generation efficiency while ensuring equipment operational safety.

[0049] This application provides a photovoltaic array adjustment method, apparatus, electronic device, and storage medium. By synchronously acquiring the operating parameters and power parameters of the photovoltaic array, predicting the health status based on the operating parameters, evaluating the power quality based on the power parameters, and generating control commands including tilt angle adjustment and / or operating voltage adjustment, compared with existing technologies that only consider a single factor or adopt a fixed adjustment strategy, the following beneficial effects can be achieved: under the premise of ensuring that the actual power generation of the photovoltaic array does not exceed the maximum operating power allowed by its health status, the power generation efficiency is significantly improved, while effectively ensuring the safety of equipment operation and the power quality of the power grid.

[0050] Furthermore, since the power generation of the adjusted photovoltaic array does not exceed the maximum operating power allowed under its healthy state, this solution can ensure that the photovoltaic array will not accelerate its aging or damage due to overload operation, thereby significantly improving the operational reliability and service life of the equipment.

[0051] In one feasible implementation scheme, such as Figure 2 As shown, predicting the health status of the photovoltaic array based on the operating parameters includes: Step 201: Based on the operating parameters, a Markov chain model with dynamic weighting factors is used to predict the probability that the photovoltaic array will be in different health levels within a preset time period in the future.

[0052] First, the health levels of the photovoltaic array need to be defined, which can be divided into multiple discrete states such as normal, minor fault, moderate fault, and severe fault. The Markov chain model is used to describe the transition patterns between these health states and obtain the state transition probabilities. Unlike traditional Markov chains, this implementation introduces a dynamic weighting factor, which allows the model to assign higher weights to recent historical data when calculating state transition probabilities, while the influence of earlier historical data decays over time. By analyzing the sequence of operating parameters (such as power output, module temperature, etc.), the model can predict the probability distribution of the photovoltaic array being in each of the aforementioned health levels within a preset time period (e.g., the next 24 hours or the next week).

[0053] First, the health levels of the photovoltaic array need to be defined, which can be divided into multiple discrete states such as normal, minor fault, moderate fault, and severe fault. The Markov chain model is used to describe the transition patterns between these health states. Unlike traditional Markov chains, this implementation introduces a dynamic weighting factor, which allows the model to assign higher weights to recent historical data when calculating state transition probabilities, while the influence of earlier historical data decays over time.

[0054] In practice, the historical health status data of the photovoltaic array is first collected, including but not limited to parameters such as component temperature, output power, and insulation resistance, as well as the corresponding photovoltaic system performance ratio (PR). This data can be obtained through the photovoltaic array's monitoring system, and the time span should be long enough to cover different seasons, weather conditions, and other operating conditions.

[0055] The collected data is cleaned to remove outliers, missing values, and duplicates. For example, temperature or power data that significantly deviates from the normal range can be removed or corrected by comparing with data from adjacent time points or using statistical methods. The cleaned data is then organized, arranged chronologically, and standardized to meet the requirements of subsequent modeling. Standardization can be achieved using the Z-score standardization method, converting the data into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0056] Rolling sampling is performed with a 10-day sampling interval, meaning that a data segment is selected every 10 days starting from the data start time. Within each sampling interval, the health state of the photovoltaic array is determined according to the health state classification criteria. Then, the state transition probability, i.e., the probability of transitioning from one healthy state to another, is calculated. The state transition probability can be calculated by statistically analyzing the ratio of the number of state transitions within different sampling intervals to the total number of samples.

[0057] Next, using the calculated state transition probabilities, a multi-state transition probability matrix is ​​constructed. The rows of this matrix represent the current state, the columns represent the next state, and the elements in the matrix represent the probability of transitioning from the current state to the next state.

[0058] The core principle of the Markov chain model is as follows: Suppose the state space of the system contains There are several states, denoted as... , , , ..., State transition probability matrix It is The matrix is ​​represented as follows:

[0059] Among them, elements From state (For example, transitioning from a normal state to a new state) (e.g., a minor fault state) state transition probability, and satisfying: for any , ,have And for any It has the following constraints:

[0060] Given an initial state probability vector ,go through After step transfer Step-state probability transition matrix It can be obtained through recursive formula The calculation yields the final result. .

[0061] The dynamic weighting factor is calculated as follows: First, define a time decay function. This is used to measure the degree of influence of historical data on the current state. Time decay function. The form is:

[0062] in Indicates time, This is the attenuation coefficient. This makes the impact of recent historical data on the state transition probability greater than that of earlier historical data. For example, it can be adjusted according to actual needs. Larger The value indicates that the influence of historical data on the current state decays more quickly. When set... A value of 0.1 indicates that the impact of historical data decreases by 10% for each additional time unit (e.g., one day).

[0063] Then, based on the time decay function Calculate the dynamic weights of state transitions within each sampling interval. Assume that from state... Transition to state The number of transitions within the k-th sampling interval is Then dynamic weight The calculation formula is:

[0064] in, This represents the total number of sampling intervals. Indicates the first Each sampling interval's time point This represents the time point of the m-th sampling interval.

[0065] Finally, using dynamic weighting factors Calculate the improved state transition probability. From state... Transition to state Improved state transition probability The calculation formula is as follows:

[0066] By introducing dynamic weighting factors, the improved Markov chain model can dynamically adjust the state transition probability based on real-time data, better reflect the dynamic changes in the health status of the photovoltaic array, and improve the model's prediction accuracy and adaptability.

[0067] Step 202: Determine the probability of the different health levels as the health status of the photovoltaic array.

[0068] This step clarifies the quantitative representation of health status. Based on the improved Markov chain model constructed in step 201, the probability distribution of future states can be calculated using the state transition probability matrix and the initial probability vector.

[0069] Specifically, an initial probability vector is determined based on the state distribution of the first sampling interval in historical data, and then the state probability distribution at future times is calculated using a recursive formula. The recursive formula is as follows:

[0070] The health status is not a single deterministic judgment, but is represented in the form of a probability distribution vector. For example, the model output could be: the probability that the photovoltaic array is in a normal state in the future is 75%, the probability that it is in a slightly faulty state is 20%, and the probability that it is in a moderately faulty state is 5%.

[0071] This probability-based representation method can more delicately reflect the uncertainty and gradual process of equipment health status, providing a richer and more accurate information foundation for subsequent maintenance decisions and operation optimization.

[0072] Finally, model validation and optimization are conducted to ensure the accuracy and reliability of the model. Specifically, this includes comparing the simulated health state time series with the actual historical state series, calculating the state matching rate (a matching rate ≥ 85% is considered a valid model), and analyzing the Pearson correlation coefficient between simulated state changes and photovoltaic system performance ratio (PR) fluctuations (a Pearson correlation coefficient > 0.6 indicates that the model can reflect the state-PR linkage pattern). Simultaneously, a data-driven update mechanism is established, recalculating the transition probability matrix quarterly to track component aging trends, ultimately forming a hybrid optimization model that balances static statistical regularities with dynamic environmental adaptability.

[0073] Through the above steps 201 and 202, the health status is upgraded from a vague qualitative judgment to a precise quantitative probabilistic prediction.

[0074] This improvement makes the core constraint in step 105—"ensuring the adjusted power output of the photovoltaic array does not exceed the maximum operating power corresponding to the health state"—precisely enforceable. The system no longer simply and drastically switches power on and off based on "normal" or "fault" conditions, but can dynamically and precisely adjust the power limit according to the probability distribution of different health levels. For example, when the probability of a "minor fault" increases, the system can predictively and slightly reduce the power limit to protect the equipment, rather than waiting for a fault to occur before taking action. This maximizes the power generation potential without sacrificing equipment safety, achieving a precise balance between safety and efficiency.

[0075] It provides a precise and dynamic safety boundary for the entire adaptive adjustment system. This enables the reinforcement learning decision model to intelligently balance "increasing power generation" and "avoiding equipment failure" based on clearly quantified health risks when optimizing power generation efficiency. In this way, it maximizes power generation efficiency while ensuring the long-term safe operation of the equipment, achieving a leap from passive response to proactive prediction and protection.

[0076] In an alternative implementation, such as Figure 3 As shown, determining the power quality of the photovoltaic array based on the power parameters includes: Step 301: Based on the electrical energy parameters, generate the original electrical energy parameter features.

[0077] This step aims to prepare standardized data input for subsequent analysis. Specifically, after collecting raw electrical energy parameters such as voltage and current from the photovoltaic system, the first step is to utilize... Transformation techniques are used to convert the signal from a three-phase stationary coordinate system to a synchronous rotating coordinate system to extract the DC and AC components characterizing power quality. Subsequently, key features such as harmonic amplitude, frequency, phase, and flicker are analyzed from the transformed signal to construct a multi-dimensional time series. This series is then embedded and mapped using a time convolution filter, transforming it into a latent spatial feature representation more suitable for deep network processing. This refers to the characteristics of the original electrical energy parameters generated in this step.

[0078] Step 302: Input the original electrical energy parameter features into a multi-scale decomposition network to obtain the trend component, periodic component, and residual component of the photovoltaic array; wherein, the multi-scale decomposition network decomposes the original electrical energy parameter features through parallel long-period recurrent neural networks, medium-period recurrent neural networks, and time-series convolutional networks to output the trend component, the periodic component, and the residual component accordingly.

[0079] This step is the core of in-depth power quality analysis. The multi-scale decomposition network employs a parallel hybrid architecture specifically designed to capture variation patterns at different time scales in power data. Long-period recurrent neural networks are responsible for capturing slowly changing trend components, such as long-term degradation of device performance or slow drift caused by environmental factors.

[0080] The medium-period recurrent neural network is responsible for capturing the periodic components of regular fluctuations, such as the periodic fluctuations in daily power generation caused by changes in solar radiation intensity.

[0081] Temporal convolutional networks are responsible for capturing transient, bursty residual components, such as singularities and noise caused by switching actions, transient shadows, or faults.

[0082] The weight coefficients of each component are dynamically calculated using an attention mechanism, and the softmax function is used to ensure weight normalization. Finally, the final feature representations of the three components are obtained by element-wise weighting.

[0083] Step 303: Based on the trend component, the periodic component and the residual component, the power quality reconstruction features are reconstructed.

[0084] This step completes feature reconstruction and fusion. First, the weighted trend, periodic, and residual features are reconstructed using a mirror decoder. Learnable basis vectors are used to perform linear transformations on each component, generating corresponding reconstructed features. Then, the three reconstructed component features are concatenated and fused using a multilayer perceptron with LeakyReLU as the activation function, ultimately generating a unified power quality reconstruction feature. This reconstruction process not only verifies the effectiveness of the decomposition but also generates a denoised fused feature representation.

[0085] Step 304: Based on the difference between the power quality reconstruction features and the original power parameter features, determine the power quality anomaly score.

[0086] The core of this step lies in quantifying the degree of power quality anomalies by calculating the "reconstruction error." The basic principle is that a well-trained decomposition and reconstruction model can accurately learn and reproduce the power characteristic patterns under normal operating conditions. Specifically, the "power quality reconstruction characteristics" are the ideal state reconstructed by the model based on the learned normal patterns, while the "original power parameter characteristics" represent the real-time operating state of the system. When the system is operating normally, the difference between the two is minimal; once a power quality anomaly occurs, the model will be unable to accurately reconstruct the abnormal pattern, leading to a significant deviation. This deviation value is the anomaly score. By comprehensively calculating multi-dimensional errors such as overall reconstruction error, trend deviation, periodic inaccuracy, and residual strength, a precise quantitative assessment of power quality problems is formed.

[0087] Step 305: Based on the comparison result between the anomaly score and the dynamically determined quality threshold, determine whether the power quality is abnormal.

[0088] This step enables the final adaptive judgment. By comparing the anomaly score with the quality threshold, a comparison result can be obtained, thereby determining whether the power quality is abnormal.

[0089] In one embodiment, the dynamic determination step of the quality threshold includes: Based on historical power quality anomaly scores, the quality threshold is adaptively updated using an exponential smoothing method.

[0090] In other words, the aforementioned quality threshold is not a fixed value, but is dynamically updated using exponential smoothing based on historical anomaly scores. This allows the quality threshold to adaptively track the normal fluctuation range of the system, avoiding false alarms caused by changes in normal operating conditions such as seasons and weather. The smoothing coefficient controls the weight ratio between historical data and current data. The initial threshold is set based on the statistical quantiles of normal samples in the training set. The final anomaly determination rule combines comprehensive anomaly scores with residual mutations: when the anomaly score exceeds the dynamic threshold, or when a sudden fluctuation in the residual component exceeds a certain fixed threshold, it is determined to be a power quality anomaly. This mechanism balances the sensitivity and robustness of detecting gradual anomalies and sudden events.

[0091] In one feasible implementation, obtaining photovoltaic array adjustment instructions determined based on the health status and the power quality includes: The integrated parameters obtained by fusing the health status and the power quality are input into a preset two-layer decision model; the first layer of the two-layer decision model is used to output the maximum operating power based on the health status information in the integrated parameters; the second layer of the two-layer decision model is used to output the tilt angle adjustment operation and / or the operating voltage adjustment operation based on the integrated parameters and the maximum operating power; and the photovoltaic array adjustment command output by the two-layer decision model is obtained.

[0092] The integrated parameters are represented in the form of state vectors, and their construction method is as follows:

[0093] in, This represents the system state vector at time t, which is also known as the comprehensive parameter. The health status probability vector is output by the dynamic health status prediction model, and each element represents the probability that the photovoltaic array is in the corresponding health level. This indicates an abnormal power quality score. For vector concatenation, The L1 norm represents the residual component and reflects the intensity of sudden anomalies in the electrical signal.

[0094] One specific implementation of the two-level decision model is to construct a two-level Q-table to achieve hierarchical decision-making. A Q-table is a data structure used to store state-action values, where each Q-value represents the long-term expected reward of performing a certain action in a specific state.

[0095] First-level decision-making (health status adjustment layer): based on the health status information in the comprehensive parameters, i.e., the health status probability vector. The first-level Q-table is queried. Based on the health status assessment results, this Q-table outputs the optimal power control command, i.e., the maximum operating power. This maximum operating power is specifically represented by the power derating factor. Its value range is usually 0.5 to 1.0, where 1.0 indicates full power operation.

[0096] Second-level decision-making (power quality optimization layer): based on the aforementioned comprehensive parameters. and the power derating factor of the first layer output The second-level Q-table is then queried. Given the known power constraints, this Q-table comprehensively considers the power quality and outputs optimal operating parameter adjustment commands, namely the tilt angle adjustment operation and / or operating voltage adjustment operation. These adjustment operations collectively constitute the action vector. :

[0097] in, This refers to the adjustment amount of the photovoltaic panel tilt angle. This is the working voltage correction amount. The power derating factor determined for the first layer.

[0098] This action vector It is the complete photovoltaic array adjustment command output by the two-level decision model (by querying the two-level Q table). This command includes both power control commands and operating parameter adjustment commands. Through this hierarchical decision mechanism, dynamic optimization of power quality is achieved while prioritizing the health of the equipment.

[0099] In an optional implementation, the two-layer decision model is a reinforcement learning model trained using the Q-learning algorithm. During training of the reinforcement learning model, a reward value is calculated using a reward function, and the calculation of the reward value includes at least one of the following: The reward value is increased in response to an increase in power generation; the reward value is decreased in response to an abnormal power quality; and the reward value is decreased in response to an increase in the adjustment range of the tilt angle adjustment operation and / or the operating voltage adjustment operation.

[0100] This means that the aforementioned two-level decision-making model (or two-level Q-table) is not based on pre-defined rules, but rather is trained by allowing the agent to continuously "try and learn" in simulated or real environments. Q-learning is one implementation method, the core of which is to learn an evaluation function called the Q-value, which can predict the long-term cumulative reward of performing a certain adjustment action in a specific system state.

[0101] During the training of the reinforcement learning model, a reward value is calculated using a reward function. The reward function acts like a "conductor," directly determining the model's learning direction and final behavior. Its design aims to guide the model to make optimal decisions under complex constraints, specifically including the following key reward and penalty components: Reward and punishment mechanism 1: In response to an increase in power generation capacity, increase the reward value: By encouraging the model to take actions that improve power generation efficiency, the primary goal of optimization is to directly impact the economic benefits of the photovoltaic system. When the model performs an adjustment, if the actual power generation of the system increases compared to before, the reward function will award a positive reward value. This incentivizes the model to actively explore control strategies that maximize power generation.

[0102] Reward and Penalty Mechanism 2: In response to power quality being judged as abnormal, the reward value will be reduced. By constraining the model's behavior, grid security and power quality are ensured. This prevents the model from taking dangerous actions that could degrade power quality (such as excessive harmonics or voltage flicker) in pursuit of blindly increasing power generation. Once the power quality detection module determines that the current state is abnormal, the reward function applies a significant negative reward value (i.e., a penalty). This heavy penalty strongly discourages the model from choosing the same action that would cause the abnormality again in similar future states.

[0103] Reward and Penalty Mechanism 3: In response to an increase in the adjustment range of the tilt angle adjustment operation and / or operating voltage adjustment operation, the reward value will be reduced. By encouraging smooth, precise control, overly drastic or frequent adjustments are avoided. The reward function calculates the magnitude of the adjustment (e.g., the sum of the squares of angle changes or voltage changes) and applies a slight negative reward value to actions with excessively large magnitudes. This helps improve system stability, reduce mechanical wear on actuators, and avoid unnecessary shocks to the power grid.

[0104] In summary, this reward function cleverly balances the three sometimes conflicting objectives of "power generation efficiency," "power quality," and "control stability" through a "one reward, two penalties" mechanism, thereby guiding the reinforcement learning model to eventually converge to an intelligent control strategy that can generate power efficiently and operate safely and stably.

[0105] During model training, the system employs a decision-making strategy that combines exploration and exploitation. The system randomly explores the possible actions, with a certain probability of success; otherwise, it selects the action considered optimal. This is achieved by designing a multi-objective reward function. To guide the direction of model learning:

[0106] This function contains three key items: (1) Power generation improvement reward item:

[0107] (2) Penalty items for range of motion: ( () and (3) Power quality anomaly penalty items ( ).

[0108] in for Power generation at the current moment, For the theoretical maximum power, This is an indicator of abnormal power quality. The value is 0 or 1. When the indicator is 0, it means that the current power quality is normal and no abnormality has been detected. When the indicator is 1, it means that the power quality detection system has determined that there is an abnormal condition, such as excessive harmonics, voltage flicker, or other problems that do not meet the power quality standards.

[0109] After completing action selection and execution, the system updates the Q-value using the Q-learning algorithm, with the following update formula:

[0110] in, For state Next execution action vector Expected returns The learning rate controls the weight of new information over old information. For state Next action The instant reward obtained afterward; This is a discount factor used to balance the importance of current and future rewards; In the new state Choose the optimal action And calculate its largest value.

[0111] It represents the old state. Execute the old action Subsequently, the assessment of the new state Next, execute a future action. Its long-term value.

[0112] By iterating this process repeatedly until the model converges, the optimal control strategy is finally extracted from the trained Q-table, forming a stable and reliable photovoltaic array adjustment scheme.

[0113] In an optional implementation, the method further includes: In response to being in photovoltaic array adjustment mode, the parameters of the two-level decision model remain unchanged; in response to switching to offline maintenance mode, the parameters of the two-level decision model are updated until convergence.

[0114] This method balances system stability and self-evolution capabilities by introducing a dual-mode operation mechanism. When the system is in photovoltaic array adjustment mode, the parameters of the two-layer decision model remain locked. This design ensures the absolute reliability and response efficiency of the online control process. The model executes stable and predictable decisions based on existing mature experience, completely avoiding the operational risks that may arise from exploratory learning in a real production environment.

[0115] When the system switches to offline maintenance mode, the model's parameter update process is initiated. In this mode, the system operates in a safe simulation environment, and the two-layer decision model continuously optimizes itself through reinforcement learning algorithms. Its parameters are continuously updated until performance stabilizes. This process enables the model to learn new operating data, adapt to changes in equipment status, and solidify the optimization results into the parameters, thus reserving a more comprehensive control strategy for the next stage of online operation.

[0116] This mode-switching mechanism creates a virtuous cycle of learning: online mode ensures the immediate security and stability of system operation, while offline mode provides continuous optimization space. Through the alternation of these two modes, the system maintains the reliability of decision-making in a production environment while gaining the ability to continuously evolve, thereby maximizing long-term operational benefits.

[0117] Based on the same technical concept, embodiments of this application also provide a photovoltaic array adjustment device, such as... Figure 4 As shown, the photovoltaic array adjustment device includes: The parameter acquisition module 401 is used to acquire the operating parameters and electrical energy parameters of the photovoltaic array.

[0118] The prediction module 402 is used to predict the health status of the photovoltaic array based on the operating parameters.

[0119] The determination module 403 is used to determine the power quality of the photovoltaic array based on the power parameters.

[0120] The instruction acquisition module 404 is used to acquire photovoltaic array adjustment instructions determined based on the health status and the power quality; the photovoltaic array adjustment instructions include: tilt angle adjustment operation and / or operating voltage adjustment operation.

[0121] The adjustment module 405 is used to execute the photovoltaic array adjustment command so that the power generation of the adjusted photovoltaic array does not exceed the maximum operating power corresponding to the healthy state.

[0122] In one feasible implementation, the prediction module is configured to predict the health status of the photovoltaic array based on the operating parameters, for the purpose of: Based on the operating parameters, a Markov chain model with dynamic weighting factors is used to predict the probability that the photovoltaic array will be in different health levels within a preset time period in the future.

[0123] The probabilities of the different health levels are determined as the health status of the photovoltaic array.

[0124] In one feasible implementation, a determining module is used to determine the power quality of the photovoltaic array based on the power parameters, for the following purposes: Based on the electrical energy parameters, the original electrical energy parameter characteristics are generated.

[0125] The original electrical energy parameter features are input into a multi-scale decomposition network to obtain the trend component, periodic component, and residual component of the photovoltaic array. The multi-scale decomposition network decomposes the original electrical energy parameter features through parallel long-period recurrent neural networks, medium-period recurrent neural networks, and time-series convolutional networks to output the trend component, periodic component, and residual component accordingly.

[0126] Based on the trend component, the periodic component, and the residual component, the power quality reconstruction characteristics are obtained.

[0127] Based on the difference between the power quality reconstruction features and the original power parameter features, an anomaly score for power quality is determined.

[0128] Based on the comparison between the anomaly score and the dynamically determined quality threshold, it is determined whether the power quality is abnormal.

[0129] In one feasible implementation, the dynamic determination step of the quality threshold includes: Based on historical power quality anomaly scores, the quality threshold is adaptively updated using an exponential smoothing method.

[0130] In one feasible implementation, the instruction acquisition module is configured to acquire photovoltaic array adjustment instructions determined based on the health status and the power quality, for the purpose of: The integrated parameters obtained by fusing the health status and the power quality are input into a preset two-layer decision model; the first layer of the two-layer decision model is used to output the maximum operating power based on the health status information in the integrated parameters; the second layer of the two-layer decision model is used to output the tilt adjustment operation and / or the operating voltage adjustment operation based on the integrated parameters and the maximum operating power.

[0131] Obtain the photovoltaic array adjustment command output by the two-layer decision model.

[0132] In one feasible implementation, the two-layer decision model is a reinforcement learning model trained based on the Q-learning algorithm.

[0133] When training the reinforcement learning model, a reward value is calculated using a reward function, wherein the calculation of the reward value includes at least one of the following: Increase reward value in response to increased power generation.

[0134] In response to power quality being deemed abnormal, the reward value is reduced.

[0135] In response to an increase in the adjustment range of the tilt angle adjustment operation and / or the operating voltage adjustment operation, the bonus value is reduced.

[0136] In one feasible implementation, the device further includes: A holding module is used to maintain the parameters of the two-layer decision model unchanged in response to being in photovoltaic array adjustment mode.

[0137] An update module is used to update the parameters of the two-level decision model until convergence in response to switching to offline maintenance mode.

[0138] Figure 5 A schematic diagram of an electronic device provided in this application embodiment includes: a processor 501, a storage medium 502, and a bus 503. The storage medium 502 stores machine-readable instructions executable by the processor 501. When the electronic device runs the photovoltaic array adjustment method as described in the embodiment, the processor 501 communicates with the storage medium 502 via the bus 503, and the processor 501 executes the machine-readable instructions to perform the steps as described in the embodiment.

[0139] In this embodiment, the storage medium 502 may also execute other machine-readable instructions to perform other methods as described in the embodiment. For details on the specific execution steps and principles, please refer to the description of the embodiment, which will not be repeated here.

[0140] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to perform the steps as described in the embodiments.

[0141] In this embodiment, the computer program, when run by the processor, can also execute other machine-readable instructions to perform other methods as described in the embodiments. For details on the specific execution steps and principles, please refer to the description of the embodiments, which will not be repeated here.

[0142] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0143] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0144] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0145] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0146] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A photovoltaic array adjustment method, characterized in that, The method includes: Obtain the operating parameters and electrical parameters of the photovoltaic array; Predict the health status of the photovoltaic array based on the operating parameters; The power quality of the photovoltaic array is determined based on the aforementioned power parameters; Obtain photovoltaic array adjustment instructions determined based on the health status and the power quality; the photovoltaic array adjustment instructions include: tilt angle adjustment operation and / or operating voltage adjustment operation; The photovoltaic array adjustment command is executed to ensure that the power generation of the adjusted photovoltaic array does not exceed the maximum operating power corresponding to the healthy state.

2. The method according to claim 1, characterized in that, Predicting the health status of the photovoltaic array based on the operating parameters includes: Based on the operating parameters, a Markov chain model with dynamic weighting factors is used to predict the probability that the photovoltaic array will be in different health levels within a preset time period in the future. The probabilities of the different health levels are determined as the health status of the photovoltaic array.

3. The method according to claim 1, characterized in that, Determining the power quality of the photovoltaic array based on the aforementioned power parameters includes: Based on the electrical energy parameters, the original electrical energy parameter characteristics are generated; The original electrical energy parameter features are input into a multi-scale decomposition network to obtain the trend component, periodic component, and residual component of the photovoltaic array. The multi-scale decomposition network decomposes the original electrical energy parameter features through parallel long-period recurrent neural networks, medium-period recurrent neural networks, and time-series convolutional networks to output the trend component, periodic component, and residual component accordingly. Based on the trend component, the periodic component, and the residual component, the power quality reconstruction characteristics are obtained. Based on the difference between the power quality reconstruction features and the original power parameter features, an anomaly score for power quality is determined; Based on the comparison between the anomaly score and the dynamically determined quality threshold, it is determined whether the power quality is abnormal.

4. The method according to claim 3, characterized in that, The dynamic determination step of the quality threshold includes: Based on historical power quality anomaly scores, the quality threshold is adaptively updated using an exponential smoothing method.

5. The method according to claim 1, characterized in that, Obtaining photovoltaic array adjustment instructions determined based on the health status and the power quality includes: The integrated parameters obtained by fusing the health status and the power quality are input into a preset two-layer decision model; the first layer of the two-layer decision model is used to output the maximum operating power based on the health status information in the integrated parameters; the second layer of the two-layer decision model is used to output the tilt adjustment operation and / or the operating voltage adjustment operation based on the integrated parameters and the maximum operating power. Obtain the photovoltaic array adjustment command output by the two-layer decision model.

6. The method according to claim 5, characterized in that, The two-layer decision model is a reinforcement learning model trained based on the Q-learning algorithm; When training the reinforcement learning model, a reward value is calculated using a reward function, wherein the calculation of the reward value includes at least one of the following: Increase reward value in response to increased power generation capacity; In response to power quality being deemed abnormal, the reward value is reduced; In response to an increase in the adjustment range of the tilt angle adjustment operation and / or the operating voltage adjustment operation, the bonus value is reduced.

7. The method according to claim 5, characterized in that, The method further includes: In response to being in photovoltaic array adjustment mode, the parameters of the two-level decision model remain unchanged; In response to switching to offline maintenance mode, the parameters of the two-level decision model are updated until convergence.

8. A photovoltaic array adjustment device, characterized in that, The device includes: The parameter acquisition module is used to acquire the operating parameters and electrical energy parameters of the photovoltaic array; The prediction module is used to predict the health status of the photovoltaic array based on the operating parameters; A determining module is used to determine the power quality of the photovoltaic array based on the power parameters; The instruction acquisition module is used to acquire photovoltaic array adjustment instructions determined based on the health status and the power quality; the photovoltaic array adjustment instructions include: tilt angle adjustment operation and / or operating voltage adjustment operation; The adjustment module is used to execute the photovoltaic array adjustment command so that the power generation of the adjusted photovoltaic array does not exceed the maximum operating power corresponding to the healthy state.

9. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the photovoltaic array adjustment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the photovoltaic array adjustment method as described in any one of claims 1 to 7.

Images