Photovoltaic power generation fault detection method and device based on improved particle swarm optimization
By improving the particle swarm algorithm, using dynamic inertial weights and local search optimization to train the photovoltaic power generation fault diagnosis model, the problem of local optimal solution in traditional methods in photovoltaic power generation fault detection is solved, and high-precision and real-time fault detection is achieved, improving the problem of high-precision and real-time fault detection. System stability.
Patent Information
- Application Number
- CN202411324831.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-23
- Publication Date
- 2024-12-27
AI Technical Summary
Traditional particle swarm algorithms are prone to fall into local optimal solutions when dealing with dynamic and complex photovoltaic power generation fault detection problems, resulting in inaccurate detection results.
The improved particle swarm algorithm is adopted to train the photovoltaic power generation fault diagnosis model by introducing dynamically adjustable inertial weights and local search optimization, and fault detection is carried out in combination with real-time operation data, and fault alarm information is sent through the wireless network.
It realizes high-precision and real-time photovoltaic power generation fault detection, avoids the problems of inaccurate detection and delay, and improves the stability and reliability of the system.
Smart Images
Figure CN119210338A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic power generation fault detection, and in particular to a photovoltaic power generation fault detection method based on an improved particle swarm algorithm. Background Art
[0002] In the prior art, photovoltaic power generation systems, as a representative of clean energy, have been widely used in various scenarios. However, the operational stability of photovoltaic power generation systems is susceptible to a variety of factors, including changes in environmental conditions, aging of electrical equipment, and human operational errors. These factors may cause photovoltaic power generation systems to have varying degrees of failure, such as component damage, inverter abnormality, poor electrical connection, etc. Traditional fault detection methods mainly rely on manual inspections or simple rule settings, and usually use fixed thresholds or expert systems for judgment. These methods have the disadvantages of low detection efficiency, poor real-time performance, and high missed detection rate. Especially in complex environmental conditions and large-scale application scenarios, it is difficult to meet the needs of efficient and accurate fault detection; with the continuous expansion of the scale and complexity of photovoltaic power generation systems, it is particularly important to develop a method that can detect system faults in real time and accurately. As an optimization algorithm based on swarm intelligence, particle swarm optimization has the advantages of fast calculation speed and easy implementation, and has been applied in a variety of optimization problems. However, when dealing with dynamic and complex photovoltaic power generation fault detection problems, traditional particle swarm optimization algorithms are prone to fall into local optimal solutions, resulting in inaccurate detection results. In order to solve the above problems, an improved fault detection method is urgently needed to ensure the safe and stable operation of the photovoltaic power generation system. Summary of the invention
[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a photovoltaic power generation fault detection method based on an improved particle swarm algorithm, aiming to solve the technical problem that the traditional particle swarm algorithm is prone to fall into the local optimal solution when dealing with dynamic and complex photovoltaic power generation fault detection problems, resulting in inaccurate detection results.
[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: The present invention provides a photovoltaic power generation fault detection method based on an improved particle swarm algorithm.
[0005] Preferably, the method comprises the following steps:
[0006] Step S10: Acquire real-time operation data of the photovoltaic power generation system, including voltage signals, current signals, temperature and irradiance parameters;
[0007] Step S20: preprocessing the acquired real-time operation data;
[0008] Step S30: using an improved particle swarm algorithm to train a photovoltaic power generation fault diagnosis model, and optimizing the photovoltaic power generation fault diagnosis model by introducing a dynamically adjustable inertia weight and local search;
[0009] Step S40: input the pre-processed real-time operation data into the fault diagnosis model, and if a fault is detected, output fault alarm information and locate the fault source.
[0010] Preferably, the process of introducing inertia weight and local search to optimize the photovoltaic power generation fault diagnosis model includes:
[0011] Set the initial parameters of the particle swarm, including the initial value of inertia weight, initial velocity and initial position;
[0012] Evaluate the fitness of each particle and update the particle's historical optimal position and global optimal position;
[0013] Dynamically adjust the inertia weight according to the particle's fitness value, and update the position and speed;
[0014] Repeat the iteration until the convergence condition is met and obtain the optimized photovoltaic power generation fault diagnosis model.
[0015] Preferably, the training process of the fault diagnosis model includes initializing the model using historical fault data.
[0016] Preferably, the local search includes introducing a neighborhood-based search strategy.
[0017] Preferably, the preprocessing step of the real-time operation data includes filtering and denoising of the data, normalization processing and elimination of abnormal data.
[0018] Preferably, the fault alarm information can be sent to a remote monitoring terminal via a wireless network for remote monitoring and rapid response.
[0019] Preferably, the device comprises:
[0020] Data acquisition module: used to obtain real-time operation data of photovoltaic power generation system;
[0021] Data preprocessing module: used to preprocess the acquired real-time operation data;
[0022] Fault diagnosis module: used to train the photovoltaic power generation fault diagnosis model using the improved particle swarm algorithm and detect the faults of the photovoltaic power generation system through the model;
[0023] Result output module: used to display fault detection results and issue alarms.
[0024] The beneficial effect of the present invention is that compared with the traditional particle swarm algorithm, which is easy to fall into the local optimal solution when dealing with dynamic and complex photovoltaic power generation fault detection problems, resulting in inaccurate detection results, the present application achieves high-precision and real-time photovoltaic power generation fault detection by improving the optimized use of the particle swarm algorithm, thereby avoiding the problems of inaccurate detection and delay, and improving the stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0026] Figure 1 It is a flow chart of the first embodiment of the photovoltaic power generation fault detection method based on the improved particle swarm algorithm of the present invention. DETAILED DESCRIPTION
[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0028] Embodiment 1: Figure 1 FIG. 1 is a flow chart of a first embodiment of a photovoltaic power generation fault detection method based on an improved particle swarm algorithm according to the present invention.
[0029] Step S10: Acquire real-time operation data of the photovoltaic power generation system, including voltage signals, current signals, temperature and irradiance parameters;
[0030] Step S20: preprocessing the acquired real-time operation data;
[0031] It should be noted that real-time operating data include parameters such as voltage, current, temperature and irradiance, among which the voltage signal represents the current voltage value of the system and is used to monitor voltage changes; the current signal represents the current intensity in the system and is used to analyze the changing trend of the current; the temperature parameter is the temperature of the photovoltaic components and the environment, which is used to evaluate the impact of temperature on power generation efficiency; the irradiance parameter reflects the intensity of solar radiation and is an important factor affecting the efficiency of photovoltaic power generation; the preprocessing methods include data cleaning, that is, removing outliers or missing data to avoid these data affecting subsequent analysis; data smoothing, that is, reducing noise in the data through filtering or moving average methods; normalization, that is, converting data of different scales to a unified standard for easy comparison and analysis; time alignment, that is, ensuring that the timestamps of different data sources are consistent for synchronous analysis.
[0032] It should be understood that in order to better implement fault detection, the process of obtaining real-time operating data of the photovoltaic power generation system can be: in the photovoltaic power station, voltage sensors, current sensors, temperature sensors and light intensity sensors are installed. These sensors collect various data in real time, such as the voltage and current values of each photovoltaic panel, the temperature of the environment and components, and the intensity of solar radiation. The sensors transmit the data to the central control system through wired connections or wireless communication modules (such as Zigbee or LoRa).
[0033] It should be understood that in order to better ensure the accuracy and reliability of the data, the process of preprocessing the acquired real-time operation data includes data filtering and denoising, normalization processing and abnormal data removal. The specific implementation steps may be:
[0034] (1) Filtering and denoising
[0035] The real-time voltage signal obtained may contain high-frequency noise due to external interference or sensor noise. In order to eliminate this noise, a low-pass filter can be used for processing. For example, in a photovoltaic system, the sampling frequency of the voltage signal is 1000Hz, but part of the signal generates high-frequency noise due to electromagnetic interference. A low-pass filter with a cutoff frequency of 100Hz can be applied to retain only the effective low-frequency signal and remove the high-frequency noise, thereby obtaining smoother voltage data.
[0036] (2) Normalization
[0037] Data collected by different sensors may have different dimensions and ranges, for example, the voltage range is between 0-1000V, while the temperature range may be between -20℃ and 80℃. In order to enable these data to be analyzed in the same model, the data needs to be normalized; for example, the voltage data is between 0-1000V, and the temperature data is between -20℃ and 80℃; the Min-Max normalization method can be used to convert the voltage and temperature data to the range of 0-1 respectively; for example, for the voltage signal V, the normalized value is (V-0) / (1000-0), and for the temperature signal T, the processed value is (T+20) / (80+20).
[0038] (3) Abnormal data elimination
[0039] In actual operation, extreme values or abnormal data may be generated due to sensor failure or abnormal environment. These data will have a negative impact on the analysis results, so it is necessary to remove abnormal data. For example, during a day's operation, the light sensor shows a value exceeding the solar surface irradiance due to a failure. You can set a reasonable threshold range (for example, 0-1200W / m 2 ), remove the illuminance data beyond this range. If the illuminance data of a certain sampling is 1500W / m 2 , the data can be considered abnormal and directly excluded from subsequent analysis.
[0040] Through the above steps, the acquired real-time operation data of the photovoltaic power generation system can be effectively preprocessed to ensure the accuracy, comparability and reliability of the data, thereby laying the foundation for subsequent fault detection and analysis.
[0041] Step S30: using an improved particle swarm algorithm to train a photovoltaic power generation fault diagnosis model, and optimizing the photovoltaic power generation fault diagnosis model by introducing a dynamically adjustable inertia weight and local search;
[0042] It should be noted that the improved particle swarm algorithm refers to:
[0043] Inertia weight: The inertia weight in the particle swarm algorithm is used to control the speed and exploration ability of particles. The traditional particle swarm algorithm uses a fixed inertia weight, which may not be able to balance between global search and local search. The introduction of inertia weight means that the weight value will be dynamically adjusted with the number of iterations - a larger inertia weight is used in the early stage of the algorithm to enhance the global search ability; the inertia weight is gradually reduced in the later stage of the algorithm to enhance the local search ability;
[0044] Local search optimization: Adding a local search optimization mechanism on the basis of the particle swarm algorithm can help particles better explore the optimal solution in the current area. When a particle enters a potential good solution area, the local search optimization mechanism will conduct a more detailed search in the area, thereby improving the accuracy of the solution;
[0045] The specific implementation steps of introducing inertia weight and local search optimization of the photovoltaic power generation fault diagnosis model may be:
[0046] (1) Initialize particle swarm
[0047] For example, when an inverter in a photovoltaic power generation system fails, the parameters of the fault diagnosis model are first initialized as the position information of the particles. These parameters may include voltage thresholds, temperature limits, etc. The speed of each particle is initialized to a random value, indicating the adjustment speed of the model parameters. In the initial stage, the position of a particle may correspond to a voltage threshold of 400V and a temperature limit of 50°C, while the speed determines how these thresholds are adjusted in the next step.
[0048] (2) Dynamically adjust inertia weight
[0049] A larger inertia weight (for example, 0.9) is used to ensure that particles can escape from the local optimal solution and find more suitable model parameters. When the particle swarm gradually converges, the inertia weight is gradually reduced (for example, to 0.4) to refine the search range and lock it near the optimal solution. As the algorithm iterates, the inertia weight is gradually reduced from 0.9 to 0.4, so that particles may try larger parameter changes in the early stage (for example, from 400V to 380V), and only make fine adjustments in the later stage (for example, from 380V to 378V).
[0050] formula:
[0051]
[0052] Among them, w max is the initial inertia weight (e.g. 0.9), w min is the minimum inertia weight (e.g. 0.4), max_iter is the maximum number of iterations, and t is the current number of iterations. At the 50th iteration, if w max =0.9, w min =0.4, and the maximum number of iterations is 100, then the inertia weight w(50) will be:
[0053]
[0054] (3) Particle position update
[0055] Each particle adjusts its position according to the current speed and the updated inertia weight. The position update corresponds to the adjustment of the model parameters.
[0056] formula:
[0057]
[0058] x i (t+1)=x i (t)+v i (t+1)
[0059] Among them, v i (t) is the velocity of particle i at time t, x i (t) is the position of particle i at time t, is the historical optimal position of particle i, g is the global optimal position, c1 and c2 are learning factors, and r1 and r2 are random numbers.
[0060] For example, the voltage threshold of a particle after adjustment changes from 380V to 385V, and the temperature limit changes from 50℃ to 48℃. This process reflects the optimization of the diagnostic model parameters. For example, if the voltage threshold of a particle was 400V and the speed was -20V / step before, after the inertia weight is adjusted, the new speed may become -10V / step, and the new voltage threshold may be adjusted to 390V. The updated speed and position calculations are:
[0061] v i (t+1)=0.65×(-5)+2.0×0.5×(385-380)+2.0×0.5×(390-380)
[0062] x i (t+1)=380+11.75
[0063] (4) Local search optimization
[0064] When a particle in the particle swarm gradually approaches the possible global optimal solution during the iteration process (for example, the voltage threshold is close to 380V and the temperature limit is close to 45℃), the local search algorithm is enabled to conduct a detailed search around the particle to ensure that more accurate fault diagnosis model parameters are found. For example, if the position of a particle is close to the combination of 380V and 45℃, the local search algorithm may make fine adjustments within the range of 380±2V and 45±1℃ to further optimize the diagnosis effect.
[0065] (5) Iterate until convergence
[0066] The above steps are iterated continuously until the algorithm converges, that is, the global optimal solution no longer changes significantly, or the preset maximum number of iterations is reached. The final model parameters will be used to monitor the operating status of the photovoltaic power generation system in real time and diagnose potential faults in a timely manner.
[0067] Step S40: input the preprocessed data into the fault diagnosis model, and if a fault is detected, output fault alarm information and locate the fault source.
[0068] It should be noted that the fault alarm information includes the following:
[0069] Fault Type: Clearly indicates the type of fault detected, such as abnormal voltage, excessive current, high temperature, or insufficient irradiance.
[0070] Fault severity: Faults are graded according to their impact, such as minor faults, major faults, or critical faults, to help operations personnel assess priorities.
[0071] Fault time: Record the exact time when the fault is detected in order to analyze and track the fault development process.
[0072] Fault location: Preliminary location of the faulty component or system area, such as a PV panel, inverter, power distribution system, etc.
[0073] It is understandable that after a fault is detected, the system's alarm information will not only provide the above content, but also include recommended operating measures or repair steps. This information can be notified to the operation and maintenance personnel in a variety of ways, such as through the monitoring system interface display, SMS or email notification, etc., so as to quickly respond and handle it.
[0074] It should be understood that the process for better locating the fault source may specifically be:
[0075] Data retrospective analysis: Use historical data to compare and analyze current data to identify abnormal trends. For example, by analyzing the voltage, current and other data over the past few minutes or hours, determine whether the fault occurred gradually or instantaneously. If the fault occurs gradually, it may be related to component aging or environmental changes; if it occurs instantaneously, it may be caused by an emergency such as a short circuit or external interference.
[0076] Multi-point data comparison: By comparing the data from different sensors or different photovoltaic arrays at the same time, it is possible to identify which area or component has abnormal parameters. For example, among multiple photovoltaic components, if only one group has a significantly lower voltage, it can be located that the group of components may have a fault.
[0077] Fault pattern recognition: Use the pattern recognition function in the fault diagnosis model to match the current fault characteristics with the known fault pattern library. For example, certain specific combinations of voltage fluctuations and temperature rise may point to inverter failures, and the system can quickly identify the specific fault source through comparison.
[0078] Embodiment 2: The training process of the fault diagnosis model includes initializing the model using historical fault data
[0079] It should be noted that historical fault data includes fault type: various types of faults that occurred in the past, such as abnormal voltage, excessive current, component overheating, inverter failure, etc.; fault occurrence time: record the specific time of each fault in order to analyze the correlation between the fault and environmental factors; fault cause: detailed analysis results after the fault occurs, including the root cause of the fault, such as aging, poor connection, external interference, etc.; environmental data: environmental conditions when the fault occurs, such as temperature, humidity, irradiance, wind speed, etc., to help understand the impact of the environment on the fault; repair measures: the repair measures taken for each fault and their effects, to evaluate the effectiveness of the model's response to faults.
[0080] It is understandable that during the model initialization process, these historical fault data not only help provide training samples for the model, but can also be used to build a fault mode library to provide reference for the model. These data help the model understand the possible faults that may occur under similar conditions and improve the accuracy of fault detection.
[0081] It should be understood that in order to better utilize historical fault data to initialize the model, the specific process may be:
[0082] Data cleaning and preprocessing: First, the historical fault data is cleaned and preprocessed, including removing incomplete or erroneous data, smoothing abnormal data, and normalizing different data dimensions (such as standardizing data of different dimensions). For example, the voltage data of different photovoltaic modules is normalized to the range of 0-1 to make them comparable in the model.
[0083] Feature extraction and selection: Extract key features from historical fault data, such as the frequency of voltage fluctuations, temperature change trends, inverter operating current, etc., and select features closely related to the fault for model training. For example, for inverter faults, special attention can be paid to the current and temperature features.
[0084] Data labeling: Label historical data, that is, label the data according to the fault type and cause. For example, a certain voltage data is labeled as "abnormal voltage fault" and another temperature data is labeled as "overheating fault". These labeled data will be used for supervised learning to help the model identify and classify different faults.
[0085] Model training and validation: The preprocessed historical data is input into the fault diagnosis model for preliminary training. The model learns the relationship between different features in the data and the fault type by continuously adjusting internal parameters. After training, part of the historical data is used as a validation set to test and evaluate the performance of the model to ensure that the model can effectively identify and classify faults.
[0086] Build a fault mode library: Based on historical data, build a fault mode library to record the typical characteristics and environmental conditions of different fault types. When the model is initialized, the information in this library is used to optimize the model so that it can quickly identify known fault modes. For example, a certain combination of voltage fluctuation and temperature increase can be identified as an inverter overheating fault.
[0087] Continuous optimization and updating: After the initial training and verification of the model, continuous training and optimization are carried out by introducing new fault data, so that the model can continuously adapt to and learn new fault types and patterns. The library of historical data should also be updated at any time to ensure the accuracy and real-time performance of the model.
[0088] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
Claims
1. A photovoltaic power generation fault detection method based on improved particle swarm algorithm, characterized in that: The method comprises the following steps: Step S10: Acquire real-time operation data of the photovoltaic power generation system, including voltage signals, current signals, temperature and irradiance parameters; Step S20: preprocessing the acquired real-time operation data; Step S30: using an improved particle swarm algorithm to train a photovoltaic power generation fault diagnosis model, and optimizing the photovoltaic power generation fault diagnosis model by introducing a dynamically adjustable inertia weight and local search; Step S40: input the pre-processed real-time operation data into the fault diagnosis model, and if a fault is detected, output fault alarm information and locate the fault source.
2. The method according to claim 1, characterized in that In step S30, the process of introducing inertia weight and local search to optimize the photovoltaic power generation fault diagnosis model includes: Set the initial parameters of the particle swarm, including the initial value of inertia weight, initial velocity and initial position; Evaluate the fitness of each particle and update the particle's historical optimal position and global optimal position; Dynamically adjust the inertia weight according to the particle's fitness value, and update the position and speed; Repeat the iteration until the convergence condition is met and obtain the optimized photovoltaic power generation fault diagnosis model.
3. The method according to claim 1, characterized in that The training process of the fault diagnosis model includes initializing the model using historical fault data.
4. The method according to claim 1, characterized in that The local search includes introducing a neighborhood-based search strategy.
5. The method according to claim 1, characterized in that The preprocessing steps of the real-time operation data include data filtering and denoising, normalization processing and abnormal data elimination.
6. The method according to claim 1, characterized in that The fault alarm information can be sent to a remote monitoring terminal via a wireless network for remote monitoring and rapid response.
7. A photovoltaic power generation fault detection device based on an improved particle swarm algorithm, characterized in that: The device comprises: Data acquisition module: used to obtain real-time operation data of photovoltaic power generation system; Data preprocessing module: used to preprocess the acquired real-time operation data; Fault diagnosis module: used to train the photovoltaic power generation fault diagnosis model using the improved particle swarm algorithm and detect the faults of the photovoltaic power generation system through the model; Result output module: used to display fault detection results and issue alarms.