Inverter and photovoltaic power generation anomaly detection method and system thereof

Abstract

The invention discloses an inverter and a photovoltaic power generation anomaly detection method and system thereof, and belongs to the technical field of inverters, and the method comprises the steps: collecting the operation time sequence data and environment perception time sequence data of a specified inverter and a subordinate string of the inverter, and loading predefined equipment topological relation data; generating an inverter power prediction value at a future moment through the prediction model; dynamically generating a power lower limit alarm threshold, and if the actually measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, triggering a fault alarm and generating an abnormal event instruction containing abnormal inverter information; calculating the power deviation of each group string under the jurisdiction of the abnormal inverter, and outputting a positioning result instruction containing target fault group string information; and performing fault root cause reasoning on the target fault group string in combination with a preset rule, and generating a diagnosis conclusion and a maintenance work order. According to the method, full-process automation from anomaly detection and positioning to maintenance decision making is realized, and the problems of early warning lagging, inaccurate positioning and link splitting in a traditional method are solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of inverters, and more particularly relates to an inverter and a photovoltaic power generation anomaly detection method and system thereof. BACKGROUND

[0002] Anomaly detection technology of a photovoltaic power station is crucial for guaranteeing power generation efficiency of the power station and reducing operation and maintenance costs. At present, two types of technical means are mainly relied on in this field: one is an alarm mechanism based on a fixed threshold, which makes fault judgment by setting static limits of parameters such as current and voltage. Although this method is simple and easy to implement, it is difficult to effectively distinguish power changes caused by weather fluctuations (such as cloud cover) and equipment faults, resulting in a high false alarm rate. The other is traditional manual inspection and data analysis. This method not only has an average fault discovery period of more than 72 hours, with a serious response lag, but also is difficult to accurately associate anomalies to specific devices such as inverters and strings, resulting in low operation and maintenance efficiency.

[0003] In recent years, although some researches have tried to introduce a light power prediction model and environmental data, the existing technology still has obvious limitations. On the one hand, some models that combine prediction and diagnosis use fixed parameters, which cannot compensate for dynamic environmental factors such as irradiance mutations in real time, and the problems of false positives and missed detections are still prominent. On the other hand, in fault positioning, existing solutions either rely on mobile devices such as unmanned aerial vehicles for image acquisition, which has the risk of response delay and positioning error, or need to rely on data over several consecutive days for verification, which cannot meet the real-time requirement. In addition, there is a problem of fragmentation of the prediction-diagnosis link in many system architectures, and the data analysis process is lengthy, which cannot achieve efficient real-time linked diagnosis.

[0004] The existing technology discloses a photovoltaic power station real-time monitoring management system and method based on a cloud platform, and belongs to the field of intelligent monitoring of photovoltaic power stations. The system adopts a three-layer architecture of an application layer, a platform layer and an edge layer deployed in the cloud. The edge layer is composed of distributed edge nodes, which acquire data through a data acquisition interface and perform local preprocessing; the platform layer is configured with data storage and machine learning models for power generation prediction and equipment health assessment; the application layer provides a visual monitoring interface and an operation and maintenance terminal, supports multi-dimensional data display and alarm information pushing. The edge layer and the platform layer adopt a dynamic computing task allocation mechanism, which reduces the cloud load while ensuring real-time performance.

[0005] However, the prior art has the following shortcomings: first, its dynamic environmental adaptability is insufficient, and the machine learning model relied on uses fixed parameters, which is difficult to respond to dynamic environmental disturbances such as irradiance mutations in real time, resulting in a high false alarm rate; second, the fault positioning accuracy is limited, mainly relying on unmanned aerial vehicle inspection or historical data verification, which has the risk of response delay and positioning error, and cannot achieve accurate positioning at the string level; in addition, the system has a problem of separation between prediction and diagnosis, resulting in delayed fault warning and inability to form a closed-loop management.

[0006] In summary, the current photovoltaic anomaly detection field needs to solve three major problems: first, how to overcome the confusion between weather interference and real faults and reduce the false alarm rate; second, how to achieve accurate and rapid fault positioning from the power station level to the string level; third, how to connect the prediction and diagnosis links to build an efficient diagnosis system with real-time linkage and significantly shorten the fault response time. SUMMARY

[0007] To solve the problems in the prior art, the present application provides an inverter and a photovoltaic power generation anomaly detection method and system thereof.

[0008] The present application adopts the following technical solutions.

[0009] The first aspect of the present application provides a photovoltaic power generation anomaly detection method for an inverter, comprising:

[0010] Collecting runtime sequence data and environment perception time sequence data of a specified inverter and the group strings under its jurisdiction, and loading predefined device topology relationship data, and preprocessing the collected data;

[0011] Based on the preprocessed data, a future time inverter power prediction value is generated through a prediction model;

[0012] According to the inverter power prediction value and the fluctuation of its actual power sequence, a power lower limit alarm threshold is dynamically generated, and if the measured power of the inverter continuously falls below the power lower limit alarm threshold for a preset period, the inverter is preliminarily determined to be an abnormal inverter, triggering a fault alarm and generating an abnormal event instruction containing the abnormal inverter information;

[0013] In response to the abnormal event instruction, the power deviation of each group string under the jurisdiction of the abnormal inverter is calculated according to the device topology relationship and the pre-constructed power conservation constraint, and a positioning result instruction containing the target fault group string information is generated and outputted;

[0014] Based on the positioning result instruction, the target fault group string is subjected to fault root cause reasoning in combination with a preset rule to generate a diagnosis conclusion and a maintenance work order.

[0015] Optionally, the runtime sequence data comprises direct current side power, voltage and current data of the inverter and the string group;

[0016] The environment perception sequence data comprises irradiance and temperature data.

[0017] Optionally, the prediction model is a dual-channel long short-term memory network model, wherein:

[0018] The first channel is configured to process inverter power sequence data in the runtime sequence data and extract inverter power sequence features;

[0019] The second channel is configured to process the environment perception sequence data and extract environment perception sequence features;

[0020] The prediction model outputs an inverter power prediction value by fusing the inverter power sequence features and the environment perception sequence features.

[0021] Optionally, the dynamically generated power lower limit alarm threshold comprises:

[0022] A standard deviation of an actual power sequence in a recent sliding time window is calculated to quantify the volatility of the actual power sequence;

[0023] The product of the standard deviation and a first adaptive coefficient, the product of the inverter power prediction value and a second adaptive coefficient are calculated respectively, and the larger one of the two products is determined;

[0024] The inverter power prediction value is subtracted by the larger one of the two products, and the calculation result is taken as the power lower limit alarm threshold.

[0025] Optionally, the calculation of the power deviation of each string under the abnormal inverter comprises:

[0026] It is judged whether the total power deviation value of the abnormal inverter exceeds a preset threshold;

[0027] If it exceeds, all the string groups under the abnormal inverter are obtained according to the device topology relationship, and a power conservation constraint equation is established with the total power deviation of the inverter equaling the sum of the power deviations of each string group;

[0028] On the premise of meeting the power conservation constraint equation, the power deviation value of each string group is calculated by an optimization algorithm combined with real-time operation data of each string group.

[0029] Optionally, the calculation of the power deviation value of each string group by the optimization algorithm comprises:

[0030] An optimization problem is constructed with the objective of minimizing the weighted sum of squares of the power deviations of all string groups;

[0031] Each string's power deviation is assigned a weighting factor, which is determined in the following way:

[0032] Calculate the normalized deviations of the real-time operating voltage and current of the string from their normal reference values. After assigning weight coefficients to the voltage and current deviations, perform a weighted summation and use the reciprocal of the summation result as the weight factor of the string.

[0033] Under the premise of satisfying the power conservation constraint equation, solve the optimization problem and calculate the power deviation estimate of each string.

[0034] Optionally, the instruction to generate and output the location result containing the target fault string information includes:

[0035] All calculated string power deviation values ​​are sorted in descending order according to their absolute power deviation values.

[0036] Strings whose absolute power deviation is greater than a preset deviation threshold are selected as candidate fault strings;

[0037] The candidate fault groups are prioritized and sorted based on one or more sorting strategies to generate a priority list;

[0038] The sorting strategies include: sorting in descending order by the absolute value of power deviation, sorting in descending order by the proportion of the power deviation value of a single string to the total power deviation value, or assigning a confidence score to the string based on real-time running data and sorting in descending order by the score.

[0039] The structured data containing the priority list is output as the location result instruction.

[0040] Optionally, the positioning result instruction further includes:

[0041] If the signs of the power deviation values ​​of multiple candidate fault strings are consistent with the signs of the total power deviation value of the inverter, then the multiple candidate fault strings are marked as possibly caused by a common cause and included in the location result instruction.

[0042] A second aspect of the present invention provides a photovoltaic power generation anomaly detection system for an inverter, used to operate the photovoltaic power generation anomaly detection method for an inverter described in the first aspect of the present invention, comprising:

[0043] The data acquisition and preprocessing module is used to acquire runtime timing data and environmental sensing timing data of a specified inverter and its subordinate strings, load predefined device topology relationship data, and preprocess the acquired data.

[0044] The power prediction module is used to generate the inverter power prediction value for future times based on the preprocessed data and through the prediction model.

[0045] The anomaly detection and alarm module is used to dynamically generate a success rate lower limit alarm threshold based on the fluctuation of the inverter power prediction value and its actual power sequence. If the measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, the inverter is initially determined to be an abnormal inverter, triggering a fault alarm and generating an abnormal event command containing the abnormal inverter information.

[0046] The fault location module is used to respond to the abnormal event command, calculate the power deviation of each string under the abnormal inverter according to the device topology relationship and the pre-built power conservation constraints, and generate and output the location result command containing the target fault string information.

[0047] The fault diagnosis and work order generation module is used to perform root cause reasoning on the target fault string based on the location result instruction and in combination with preset rules, and generate diagnostic conclusions and maintenance work orders.

[0048] A third aspect of the present invention provides an inverter that integrates a photovoltaic power generation anomaly detection system as described in the second aspect of the present invention.

[0049] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0050] 1. This invention collects the operating sequence and environmental data of inverters and strings, dynamically generates alarm thresholds by combining them with a prediction model, and performs string-level fault location and root cause diagnosis based on power conservation constraints after an anomaly is detected. This achieves full-process automation from anomaly detection and location to maintenance decision-making, solving the problems of delayed early warning, inaccurate location, and fragmented links in traditional methods.

[0051] 2. By clearly collecting multi-dimensional data such as DC power, voltage, current, irradiance, and temperature, this invention solves the problem that traditional monitoring data has only one dimension and cannot fully depict the operating status of equipment and the impact of the environment, thus providing a reliable data foundation for high-precision prediction and diagnosis.

[0052] 3. This invention solves the problem that a single model is difficult to cope with complex environmental changes by using a dual-channel LSTM model to process power time series and environmental time series features separately and then perform fusion prediction. This results in high-precision power prediction that is more robust to environmental disturbances.

[0053] 4. This invention introduces an adaptive dynamic threshold algorithm based on the standard deviation of a sliding window, which enables the alarm threshold to be automatically adjusted according to the intensity of weather fluctuations. This solves the problem of high false alarm rate caused by the inability of fixed threshold alarm mechanisms to distinguish between weather fluctuations and actual faults.

[0054] 5. This invention establishes a topological constraint equation with the power conservation law as its core and triggers the solution when the total deviation of the inverter is significant. This solves the problem that anomaly detection can only stay at the inverter level and cannot accurately locate the fault string at the bottom level, thus realizing string-level fault location.

[0055] 6. This invention solves the problem of fairly allocating power deviations and accurately identifying the most suspicious fault sources among multiple strings by constructing a weighted least squares optimization problem and dynamically calculating weighting factors based on the real-time voltage and current deviations of each string, thereby improving the accuracy and reliability of the localization algorithm.

[0056] 7. This invention solves the problem of maintenance personnel having difficulty determining the priority order when faced with multiple abnormal signals by comprehensively sorting and filtering the calculated string power deviation using multiple strategies (absolute value, contribution, and confidence level), thereby achieving optimized allocation of maintenance resources.

[0057] 8. This invention solves the problem of identifying and attributing distributed faults caused by the same reason, such as partial shading, by identifying multiple fault strings with consistent power deviation signs and associating them as having a common cause, thereby enhancing the diagnostic capability for complex fault modes.

[0058] 9. This invention solves the problems of fragmented prediction-diagnosis links and poor data flow in the prior art by constructing a modular system that integrates data acquisition, prediction, detection, location and diagnosis functions, and realizes efficient and real-time closed-loop anomaly detection and operation and maintenance linkage.

[0059] 10. By directly integrating the anomaly detection system into the inverter, this invention solves the problems of complex external system deployment and response delay, realizes device-level edge intelligence and self-diagnosis, and improves the intelligence level of a single device and the reliability of the overall system. Attached Figure Description

[0060] Figure 1 This is a flowchart of a method provided according to an embodiment of the present invention;

[0061] Figure 2 This is a schematic diagram of dynamic threshold calculation provided according to an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0063] like Figure 1 As shown, in Embodiment 1, the present invention provides a method for detecting photovoltaic power generation anomalies in an inverter, comprising:

[0064] Step 1: Collect runtime timing data and environmental awareness timing data of the specified inverter and its subordinate strings, load predefined device topology relationship data, and preprocess the collected data.

[0065] Preferably, the device operating data includes:

[0066] The system periodically collects minute-level DC-side power, voltage, and current data for specified inverters and strings via a control database interface, with timestamps and device IDs.

[0067] Preferably, the environmental perception data includes:

[0068] The total horizontal irradiance (W / m²) at key locations of the power plant is collected in real time by controlling the sensor network. 2 The data includes equipment temperature (°C), and key locations in the power station include the area next to the inverter.

[0069] Preferably, the topology relationship data includes a predefined power plant equipment topology relationship mapping library loaded by the control system, such as a JSON structure.

[0070] Preferably, the preprocessing of the collected data includes:

[0071] High-frequency noise interference is eliminated using smoothing filtering technology to ensure clear data trends;

[0072] Standardization is performed to convert data of different dimensions to a uniform scale, which facilitates subsequent model processing.

[0073] It should be noted that by explicitly collecting multi-dimensional data such as DC-side power, voltage, current, irradiance, and temperature, this invention enhances the integrity and reliability of the data system, providing a solid data foundation for subsequent accurate prediction and fault diagnosis.

[0074] Step 2: Based on the preprocessed data, generate the inverter power prediction values ​​for future times using a prediction model.

[0075] Preferably, step 2 includes:

[0076] Extract inverter power timing features and environmental timing features from preprocessed runtime timing data and environmental-aware timing data;

[0077] The inverter power timing features and environmental timing features are input into a pre-trained prediction model to obtain the inverter power prediction value at future times.

[0078] The prediction model is updated periodically or triggered by new operational data to maintain its prediction accuracy.

[0079] More preferably, extracting the inverter power timing features includes:

[0080] Extract minute-level power timing data from the target inverter within a recent time window (e.g., 120 minutes);

[0081] Extracting environmental temporal features includes:

[0082] Extract minute-level environmental awareness time-series data related to the target inverter within the same time window.

[0083] More preferably, the prediction model is a dual-channel long short-term memory (LSTM) network model, where the first channel is used to process the inverter power timing features, the second channel is used to process the environmental timing features, and the inverter power prediction value for future times is output by fusing the dual-channel features.

[0084] It should be noted that this invention achieves high-precision prediction of inverter power by using a dual-channel LSTM model to process power timing and environmental timing features separately, thereby improving the model's adaptability to complex environmental changes.

[0085] More preferably, the triggering conditions for the triggered update prediction model include any of the following:

[0086] Performance threshold trigger: The recent average prediction error exceeds the preset performance threshold, for example, the average prediction error in the last 24 hours exceeds the preset performance threshold.

[0087] Environmental change triggers: The intensity of fluctuations in environmental sensing data exceeds the historical fluctuation range, such as the intensity of fluctuations in irradiance or temperature exceeding twice the historical average;

[0088] Data volume trigger: The accumulated amount of new data reaches a preset scale, such as when the accumulated amount of newly collected power and environmental data reaches a preset time length scale (e.g., 120 minutes).

[0089] Step 3: Based on the fluctuation of the inverter power prediction value and its actual power sequence, dynamically generate a power lower limit alarm threshold. If the measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, the inverter is initially determined to be an abnormal inverter, triggering a fault alarm and generating an abnormal event command containing the abnormal inverter information.

[0090] Preferably, step 3 includes:

[0091] Based on the fluctuation of the inverter power prediction value and its actual power sequence, the power lower limit alarm threshold at the current moment is dynamically calculated;

[0092] The measured power of the inverter collected in real time is compared with the calculated dynamic lower limit alarm threshold. If the measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, it is preliminarily determined that the inverter is in an abnormal operating state, triggering a fault alarm and generating an abnormal event command containing information about the abnormal inverter.

[0093] More preferably, the dynamically calculated power lower limit alarm threshold includes:

[0094] Calculate the standard deviation of the actual power sequence within a recent sliding time window to quantify the volatility of actual power;

[0095] Calculate the product of the standard deviation and the first adaptive coefficient, and the product of the inverter power prediction value and the second adaptive coefficient, respectively, and determine the larger of the two products;

[0096] The larger of the two products is subtracted from the inverter power prediction value, and the result is used as the power lower limit alarm threshold.

[0097] Specifically, the dynamic threshold calculation function is defined as:

[0098]

[0099] in, σ is the predicted inverter power output from the prediction model at time t; t-k:t α is the standard deviation of the actual power value in the first k minutes, used to quantify volatility; α and β are the first adaptive coefficient and the second adaptive coefficient, respectively. For example, α = 2 and β = 0.1.

[0100] The power lower limit alarm threshold is:

[0101]

[0102] More preferably, the abnormal event instruction includes the abnormal inverter ID, the timestamp of the abnormality, and the degree of power deviation (such as the percentage of the measured inverter power being lower than the lower power limit alarm threshold).

[0103] It should be noted that this invention introduces an adaptive dynamic threshold algorithm based on the standard deviation of a sliding window, which enables real-time adjustment of alarm thresholds, effectively distinguishes between weather fluctuations and real faults, and significantly reduces the false alarm rate.

[0104] Step 4: In response to the abnormal event command, calculate the power deviation of each string under the abnormal inverter according to the device topology relationship and the pre-built power conservation constraints, and generate and output a location result command containing the faulty string information.

[0105] Preferably, step 4 includes:

[0106] Step 4.1: Receive the abnormal event instruction from Step 3, obtain the specified abnormal inverter ID, query the measured power value and corresponding predicted power value of the abnormal inverter at the time of the abnormality, calculate the total power deviation value of the inverter, and determine whether the absolute value of the total power deviation value exceeds the preset inverter power deviation threshold. If it does not exceed the threshold, end the current positioning process; if it exceeds the threshold, determine it as a significant abnormality and trigger the subsequent string-level positioning process.

[0107] More preferably, step 4.1 includes:

[0108] Calculate the total power deviation of the inverter:

[0109] ΔP inv =P inv_measured -P inv_predicted

[0110] Among them, P inv_measured and P inv_predicted These are the measured power value of the abnormal inverter at the moment the abnormality occurred and its corresponding predicted power value, respectively.

[0111] Determine whether the absolute value of the deviation reaches or exceeds the preset inverter power deviation threshold δ. If it does not exceed the threshold, it is determined to be a slight fluctuation, and the current positioning process ends. If it exceeds the threshold, it is determined to be a significant anomaly, and the subsequent string-level positioning process is triggered.

[0112] Specifically, the inverter power deviation threshold δ = inverter rated power × set ratio, and its typical value is usually 5% to 10% of the inverter rated power. The specific value needs to be adjusted according to the power plant scale, environmental conditions and fault type. The set ratio can be set by the user.

[0113] Step 4.2: Based on the equipment topology data, automatically associate and obtain all strings under the abnormal inverter, and establish the topological constraint relationship equation between the total power deviation of the inverter and the power deviation of all its strings based on the power conservation law of photovoltaic system.

[0114] More preferably, the topological constraint equation includes a power deviation conservation constraint equation, whereby the total power deviation of the inverter is equal to the sum of the power deviations of all its strings, i.e.:

[0115]

[0116] Where, ΔP str,i Let be the power deviation to be solved for the i-th string, and n be the total number of strings.

[0117] It should be noted that this invention achieves accurate fault location from the inverter level to the string level by establishing a power conservation constraint equation between the inverter and the string, and triggering string-level deviation calculation when there is a significant anomaly.

[0118] Step 4.3: Under the premise of satisfying the topological constraint relationship equation, and in combination with the real-time operating data of each string, calculate the most likely power deviation value of each string using the weighted least squares optimization method.

[0119] More preferably, step 4.3 includes:

[0120] Construct the following constrained optimization problem:

[0121]

[0122] Where, ω i Let be the weight factor for the i-th string, reflecting the probability of the string being abnormal. Its calculation method is as follows:

[0123]

[0124] Where, ΔV i and ΔI i γ and λ are the deviations of the voltage and current of the i-th string from the normal values ​​(after normalization), which can be calculated based on historical or predicted values; γ and λ are the weighting coefficients of the voltage and current deviations, which can be set to 0.5 to 1.0; κ is a minimum constant (e.g., 1e-6) to prevent the denominator from being zero.

[0125] The solution to the optimization problem is:

[0126]

[0127] This solution indicates that the smaller the weight of the string (i.e., the more obvious the abnormal signs), the greater the power deviation assigned, thus enabling the initial identification of faulty strings.

[0128] It should be noted that this invention constructs a weighted least squares optimization problem and dynamically sets weighting factors based on voltage and current deviations, thereby achieving intelligent identification and priority ranking of faulty strings and improving positioning accuracy and efficiency.

[0129] Step 4.4: Based on the power deviation values ​​calculated for each string, sort them in descending order according to their absolute deviation values, and generate and output a location result instruction containing a priority list of fault strings.

[0130] It should be noted that this invention generates a priority list by sorting and filtering the power deviation of the strings using multiple strategies, thereby achieving rapid identification of faulty strings and optimized allocation of operation and maintenance resources.

[0131] More preferably, step 4.4 further includes a strategy for filtering and prioritizing fault sequences:

[0132] All calculated string power deviation values ​​are sorted in descending order according to their absolute power deviation values.

[0133] Strings whose absolute power deviation is greater than a preset deviation threshold are selected as candidate fault strings;

[0134] The candidate fault groups are prioritized and sorted based on one or more sorting strategies to generate a priority list;

[0135] The sorting strategies include: sorting in descending order by the absolute value of power deviation, sorting in descending order by the proportion of the power deviation value of a single string to the total power deviation value, or assigning a confidence score to the string based on real-time running data and sorting in descending order by the score.

[0136] The structured data containing the priority list is output as the location result instruction.

[0137] More preferably, if the signs of the power deviation values ​​of multiple candidate fault strings are consistent with the signs of the total power deviation value of the inverter, then the multiple strings are marked as possibly caused by a common cause and included in the location result instruction to prompt further analysis.

[0138] Finally, a structured list containing high-priority fault string IDs, estimated power deviation values, and confidence scores is output.

[0139] It should be noted that this invention enhances the system's diagnostic capabilities for complex fault scenarios by identifying multiple fault sequences with consistent symbols and marking them as potentially caused by a common cause.

[0140] Step 5: Based on the positioning result instructions, combined with real-time operating data, historical data patterns and preset expert knowledge base rules, perform root cause reasoning on the target fault series to generate diagnostic conclusions and maintenance work orders.

[0141] Preferably, step 5 includes:

[0142] Step 5.1: Call the preset expert knowledge base rules to perform pattern matching. The expert knowledge base contains multiple diagnostic rules consisting of conditions and conclusions.

[0143] The condition part of the rule comprehensively judges multiple characteristics such as power descent rate, ambient temperature, irradiance stability, voltage bias, current fluctuation and historical data patterns.

[0144] The conclusion section of the rule outputs the corresponding fault cause and confidence level.

[0145] An example rule is: if the power descent rate is >80%, the ambient temperature is >25°C, and the irradiance is stable, then the cause of the fault is a damaged DC fuse with a confidence level of 92%.

[0146] Step 5.2: Output diagnostic conclusions and structured repair work orders;

[0147] The diagnostic conclusion should include the faulty equipment, location information, inferred cause of the fault and its confidence level (e.g., DC fuse failure - 92% confidence level, PID attenuation - 85% confidence level, component microcrack / hot spot - 78% confidence level).

[0148] The maintenance work order is structured data, which includes faulty equipment information, location information, diagnostic conclusions, and targeted maintenance suggestions (e.g., check and replace fuse F07-08, activate PID recovery function, use infrared thermal imager to scan and locate damaged components and replace them).

[0149] Step 5.3: The generated maintenance work order will be automatically pushed to the operation and maintenance management system or mobile terminal to trigger the subsequent maintenance process.

[0150] It should be noted that this invention collects the operating sequence and environmental data of the inverter and strings, dynamically generates alarm thresholds by combining them with a prediction model, and performs string-level fault location and root cause diagnosis based on power conservation constraints after detecting an anomaly. This achieves full automation of the entire process from anomaly detection and location to maintenance decision-making, significantly improving the real-time performance and accuracy of photovoltaic power plant fault response.

[0151] In Embodiment 2, this invention provides a photovoltaic power generation anomaly detection system for an inverter, used to operate the photovoltaic power generation anomaly detection method for an inverter described in Embodiment 1, comprising:

[0152] The data acquisition and preprocessing module is used to acquire runtime timing data and environmental sensing timing data of a specified inverter and its subordinate strings, load predefined device topology relationship data, and preprocess the acquired data.

[0153] The power prediction module is used to generate the inverter power prediction value for future times based on the preprocessed data and through the prediction model.

[0154] The anomaly detection and alarm module is used to dynamically generate a success rate lower limit alarm threshold based on the fluctuation of the inverter power prediction value and its actual power sequence. If the measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, the inverter is initially determined to be an abnormal inverter, triggering a fault alarm and generating an abnormal event command containing the abnormal inverter information.

[0155] The fault location module is used to respond to the abnormal event command, calculate the power deviation of each string under the abnormal inverter according to the device topology relationship and the pre-built power conservation constraints, and generate and output the location result command containing the target fault string information.

[0156] The fault diagnosis and work order generation module is used to perform root cause reasoning on the target fault string based on the location result instruction and in combination with preset rules, and generate diagnostic conclusions and maintenance work orders.

[0157] It should be noted that this invention achieves full-process integration and automated operation of photovoltaic power generation anomaly detection by constructing a modular system that includes data acquisition, prediction, detection, location and diagnosis.

[0158] In Embodiment 3, the present invention provides an inverter that integrates a photovoltaic power generation anomaly detection system as described in Embodiment 2 of the present invention.

[0159] It should be noted that by integrating the anomaly detection system into the inverter, the present invention achieves intelligent monitoring and fault self-diagnosis at the device level, thereby improving the overall reliability and operation and maintenance efficiency of the photovoltaic system.

[0160] This invention provides an application example of a photovoltaic power generation anomaly detection method for inverters in Embodiment 4. Based on the photovoltaic power generation anomaly detection method for inverters described in Embodiment 1, the following detailed description uses an inverter from a photovoltaic power station as an example, including:

[0161] Step 1: Collect runtime timing data and environmental awareness timing data of the specified inverter and its subordinate strings, load predefined device topology relationship data, and preprocess the collected data.

[0162] Preferably, step 1 includes:

[0163] Control the database interface to periodically collect minute-level DC-side power, voltage, and current data (with timestamps and device IDs) of specified inverters and strings;

[0164] Control the sensor network to collect the total horizontal irradiance (W / m²) at key locations in the power plant in real time. 2 The data includes equipment temperature (°C), and key locations in the power station include the area next to the inverter;

[0165] The control system loads a predefined power plant equipment topology mapping library (such as a JSON structure);

[0166] Preprocess the collected raw data:

[0167] Smoothing filtering techniques are used to eliminate high-frequency noise interference and ensure clear data trends; standardization processing is performed to convert data of different dimensions to a uniform scale, which facilitates subsequent model processing.

[0168] Step 2: Based on the preprocessed data, generate the inverter power prediction values ​​for future times using a prediction model.

[0169] Preferably, step 2 includes:

[0170] Historical time-series data for the most recent 120 minutes was extracted from the preprocessed data and used as model input. This includes inverter power time-series characteristics and environmental time-series characteristics (irradiance, temperature).

[0171] The aforementioned features are input into a pre-trained dual-channel LSTM prediction model. The first channel of the model processes power time-series features, and the second channel processes environmental time-series features. By fusing the information from both channels, the model outputs the inverter power prediction value for a future time (e.g., one minute later).

[0172] More preferably, it also includes a dynamically performance-driven update mechanism:

[0173] Triggering conditions: The model update will be automatically triggered when the system detects any of the following conditions.

[0174] Performance threshold trigger: The average prediction error over the past 24 hours exceeds a preset threshold;

[0175] Environmental changes trigger the event: fluctuations in irradiance or temperature exceed twice the historical average.

[0176] Data volume trigger: Newly collected 120 minutes of power and environmental data.

[0177] Update cycle: A dynamic cycle is used instead of a fixed time interval (such as 7 days). For example, if the environment fluctuates drastically, the system may trigger updates multiple times within 24 hours; if the environment is stable, the update cycle can be extended to 7 days.

[0178] Step 3: Based on the fluctuation of the inverter power prediction value and its actual power sequence, dynamically generate a power lower limit alarm threshold. If the measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, the inverter is initially determined to be an abnormal inverter, triggering a fault alarm and generating an abnormal event command containing the abnormal inverter information.

[0179] Preferably, step 3 includes:

[0180] Based on the power prediction value obtained in step 2 and the volatility of the recent actual power sequence, the lower limit alarm threshold for power is dynamically calculated. The specific calculation formula is as follows:

[0181] The dynamic threshold calculation function is defined as follows:

[0182]

[0183] in, σ is the predicted inverter power output from the prediction model at time t; t-k:t α is the standard deviation of the actual power value in the first k minutes, used to quantify volatility; α and β are adaptive coefficients; take a sliding window of k = 60 minutes, calculate the standard deviation of the actual power in the first 60 minutes, and the adaptive coefficients α = 2 and β = 0.1.

[0184] This standard deviation sliding window is designed to capture the intensity of sudden weather changes in real time.

[0185] When irradiance fluctuates drastically (e.g., from sunny to cloudy), σ increases → the threshold is automatically widened to avoid false alarms; when the weather is stable, σ approaches 0 → the threshold is tightened to a fixed tolerance (±10%) to improve fault sensitivity.

[0186] The power lower limit alarm threshold is:

[0187]

[0188] Specifically, such as Figure 2 As shown, this illustrates the principle of adaptive adjustment of the dynamic threshold according to the volatility σ. When σ increases, the threshold range widens; when σ decreases, the threshold range narrows, effectively avoiding false alarms. The actual power curve represents the power value measured during actual system or equipment operation, the predicted power curve represents the model's prediction structure for power, and the gray shaded area represents the confidence interval of the prediction error, i.e., the range of deviation between the predicted and actual values.

[0189] When σ is small (e.g., σ = 0.5), the gray shaded area is narrow, indicating that the prediction result is relatively accurate and the error is small; when σ is large (e.g., σ = 8.2), the gray shaded area is wide, indicating that the prediction result has greater uncertainty and the error range is larger.

[0190] The false alarm avoidance point shown in the figure refers to a key judgment threshold set during the prediction process to avoid misjudging a normal state as an abnormal state (i.e., false alarm).

[0191] The system compares the measured power of the inverter with the dynamically calculated lower limit alarm threshold in real time. If the measured power of the inverter is found to be lower than the lower limit alarm threshold for three consecutive prediction cycles (i.e., 30 minutes, with each prediction cycle being 10 minutes), the inverter is initially determined to be an abnormal inverter. A fault alarm is then triggered, and an abnormal event command is generated. The command contains information such as the abnormal inverter ID, timestamp, and power deviation degree.

[0192] Step 4: In response to the abnormal event command, calculate the power deviation of each string under the abnormal inverter according to the device topology relationship and the pre-built power conservation constraints, and generate and output a location result command containing the faulty string information.

[0193] Preferably, step 4 includes:

[0194] Receive the abnormal event command generated in step 3, which includes the abnormal inverter ID;

[0195] Based on the power plant topology database, automatically associate all strings under the abnormal inverter;

[0196] The topological constraint equations based on the principle of energy conservation are constructed. These equations include a power deviation conservation constraint equation, where the total power deviation of the inverter is equal to the sum of the power deviations of all its strings, i.e.:

[0197]

[0198] Where, ΔP str,i Let be the power deviation to be solved for the i-th string, and n be the total number of strings.

[0199] Under the premise of satisfying the aforementioned topological constraint equations, and combining the real-time operating data of each string, the most likely power deviation value of each string is calculated using a weighted least squares optimization method, including:

[0200] Construct the following constrained optimization problem:

[0201]

[0202] Where, ω i Let be the weight factor for the i-th string, reflecting the probability of the string being abnormal. Its calculation method is as follows:

[0203]

[0204] Where, ΔV i and ΔI i γ and λ are the deviations of the voltage and current of the i-th string from the normal values ​​(after normalization), which can be calculated based on historical or predicted values; γ and λ are the weighting coefficients of the voltage and current deviations, which can be set to 0.5 to 1.0; κ is a minimum constant (e.g., 1e-6) to prevent the denominator from being zero.

[0205] The solution to the optimization problem is:

[0206]

[0207] This solution indicates that the smaller the weight of the string (i.e., the more obvious the abnormal signs), the greater the power deviation assigned, thus enabling the initial identification of faulty strings.

[0208] Based on the power deviation values ​​calculated for each string, sort them in descending order according to their absolute deviation values, generate and output a location result instruction containing a priority list of faulty strings, format it as structured data (such as JSON), and clearly indicate the ID of the string most likely to fail and its estimated power loss value.

[0209] Furthermore, it also includes strategies for filtering and prioritizing fault sequences, including:

[0210] All calculated string power deviation values ​​are sorted in descending order according to their absolute power deviation values.

[0211] Strings with an absolute power deviation value greater than a preset deviation threshold are selected as candidate fault strings. The preset deviation threshold is calculated similarly to the inverter power deviation threshold δ and can be set to 5% to 10% of the string's rated power.

[0212] The candidate fault groups are prioritized and sorted based on one or more sorting strategies to generate a priority list;

[0213] The sorting strategies include: sorting in descending order by the absolute value of power deviation, sorting in descending order by the proportion of the power deviation value of a single string to the total power deviation value, or assigning a confidence score to the string based on real-time running data and sorting in descending order by the score.

[0214] The structured data containing the priority list is output as the location result instruction.

[0215] More preferably, if the sign of the power deviation value of multiple candidate fault strings is consistent with the sign of the total power deviation value of the inverter, then the multiple strings are marked as possibly caused by a common cause (such as partial shading) and included in the location result instruction to prompt further analysis; if multiple strings deviate significantly from the normal range, then they are marked as high-priority fault strings and processed first.

[0216] Finally, a structured list containing high-priority fault string IDs, estimated power deviation values, and confidence scores is output.

[0217] Step 5: Based on the positioning result instructions, combined with real-time operating data, historical data patterns and preset expert knowledge base rules, perform root cause reasoning on the target fault series to generate diagnostic conclusions and maintenance work orders.

[0218] Preferably, step 5 includes:

[0219] Receive the positioning result instruction from step 4.

[0220] By combining real-time data (such as power drop amplitude, temperature, voltage bias, current fluctuation, historical data patterns, etc.), location information, and preset expert knowledge base rules, the root cause of the fault is inferred.

[0221] The system generates diagnostic conclusions and targeted repair recommendations. For example, the diagnostic conclusions include DC fuse damage - 92% confidence level, PID attenuation - 85% confidence level, and component microcrack / hot spot - 78% confidence level. The targeted repair recommendations include inspecting and replacing fuse F07-08, activating the PID recovery function, and using an infrared thermal imager to scan and locate the damaged component and replace it.

[0222] The system automatically generates structured maintenance work orders, which include faulty equipment, location information, diagnostic conclusions, and maintenance suggestions, and pushes them to the operation and maintenance management system or mobile terminals.

[0223] For example, the expert knowledge base rules include knowledge base matching rules:

[0224] IF power descent rate >80% AND ambient temperature >25℃ AND stable irradiance:

[0225] THEN Fault Cause = "DC fuse failure" (92% confidence level).

[0226] More preferably, the confidence level calculation in the diagnostic conclusion includes:

[0227]

[0228] Where, ΔP inv <δ,ΔP inv Let δ be the total power deviation of the inverter, and δ be the power deviation threshold of the inverter. When ΔP inv When ≥δ, C=0%.

[0229] Furthermore, the power plant monitoring system report interface provides analysis and control capabilities:

[0230] When a user clicks on an abnormal period, the first layer of the system control interface displays: a comparison chart of actual power vs. predicted power and dynamic threshold range, intuitively showing the abnormality;

[0231] The second layer displays trend charts of key environmental parameters (irradiance, temperature) during the same period, which helps to analyze whether the anomalies are affected by the environment.

[0232] The third layer displays related historical similar cases, providing a reference for repairs.

[0233] Users can perform interactive operations such as copying and downloading data and triggering tasks (such as starting infrared scanning) through the interface.

[0234] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0235] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for detecting photovoltaic power generation anomalies in an inverter, characterized in that, include: Collect runtime timing data and environmental sensing timing data of a specified inverter and its subordinate strings, load predefined device topology relationship data, and preprocess the collected data; Based on the preprocessed data, the inverter power prediction values ​​for future times are generated through a prediction model; Based on the fluctuation of the inverter power prediction value and its actual power sequence, a power lower limit alarm threshold is dynamically generated. If the measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, the inverter is initially determined to be an abnormal inverter, triggering a fault alarm and generating an abnormal event command containing the abnormal inverter information. In response to the abnormal event command, the power deviation of each string under the abnormal inverter is calculated according to the device topology and the pre-built power conservation constraints, and a location result command containing the target fault string information is generated and output. Based on the location result instructions, and combined with preset rules, the root cause reasoning of the target fault string is performed to generate diagnostic conclusions and maintenance work orders.

2. The photovoltaic power generation anomaly detection method for an inverter according to claim 1, characterized in that: The runtime timing data includes DC-side power, voltage, and current data for the inverter and strings; The environmental sensing time-series data includes irradiance and temperature data.

3. The photovoltaic power generation anomaly detection method for an inverter according to claim 2, characterized in that: The prediction model is a dual-channel long short-term memory network model, wherein: The first channel is used to process the inverter power timing data in the runtime timing data and extract the inverter power timing features; The second channel is used to process the environmental perception time-series data and extract environmental perception time-series features; The prediction model outputs a predicted inverter power value by fusing the inverter power timing characteristics with the environmental perception timing characteristics.

4. The photovoltaic power generation anomaly detection method for an inverter according to claim 1, characterized in that: The dynamically generated power lower limit alarm threshold includes: Calculate the standard deviation of the actual power series within a recent sliding time window to quantify the volatility of the actual power series; Calculate the product of the standard deviation and the first adaptive coefficient, and the product of the inverter power prediction value and the second adaptive coefficient, respectively, and determine the larger of the two products; The larger of the two products is subtracted from the inverter power prediction value, and the result is used as the power lower limit alarm threshold.

5. The photovoltaic power generation anomaly detection method for an inverter according to claim 1, characterized in that: The calculation of the power deviation of each string under the abnormal inverter includes: Determine whether the total power deviation of the abnormal inverter exceeds a preset threshold; If the number of strings exceeds the limit, then all strings under its jurisdiction are obtained according to the equipment topology, and a power conservation constraint equation is established in which the total power deviation of the inverter is equal to the sum of the power deviations of each string. Under the premise of satisfying the power conservation constraint equation, and combined with the real-time operating data of each string, the power deviation value of each string is calculated by an optimization algorithm.

6. The photovoltaic power generation anomaly detection method for an inverter according to claim 5, characterized in that: The calculation of the power deviation value of each string using the optimization algorithm includes: Construct an optimization problem with the objective of minimizing the weighted sum of squared power deviations of all strings; Each string's power deviation is assigned a weighting factor, which is determined in the following way: Calculate the normalized deviations of the real-time operating voltage and current of the string from their normal reference values. After assigning weight coefficients to the voltage and current deviations, perform a weighted summation and use the reciprocal of the summation result as the weight factor of the string. Under the premise of satisfying the power conservation constraint equation, solve the optimization problem and calculate the power deviation estimate of each string.

7. The photovoltaic power generation anomaly detection method for an inverter according to claim 1, characterized in that: The instruction for generating and outputting the location result containing target fault string information includes: All calculated string power deviation values ​​are sorted in descending order according to their absolute power deviation values. Strings whose absolute power deviation is greater than a preset deviation threshold are selected as candidate fault strings; The candidate fault groups are prioritized and sorted based on one or more sorting strategies to generate a priority list; The sorting strategies include: sorting in descending order by the absolute value of power deviation, sorting in descending order by the proportion of the power deviation value of a single string to the total power deviation value, or assigning a confidence score to the string based on real-time running data and sorting in descending order by the score. The structured data containing the priority list is output as the location result instruction.

8. The photovoltaic power generation anomaly detection method for an inverter according to claim 7, characterized in that: The location result instruction also includes: If the signs of the power deviation values ​​of multiple candidate fault strings are consistent with the signs of the total power deviation value of the inverter, then the multiple candidate fault strings are marked as possibly caused by a common cause and included in the location result instruction.

9. A photovoltaic power generation anomaly detection system for an inverter, used to operate the photovoltaic power generation anomaly detection method for an inverter according to any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module is used to acquire runtime timing data and environmental sensing timing data of a specified inverter and its subordinate strings, load predefined device topology relationship data, and preprocess the acquired data. The power prediction module is used to generate the inverter power prediction value for future times based on the preprocessed data and through the prediction model. The anomaly detection and alarm module is used to dynamically generate a success rate lower limit alarm threshold based on the fluctuation of the inverter power prediction value and its actual power sequence. If the measured power of the inverter is continuously lower than the power lower limit alarm threshold for a preset period, the inverter is initially determined to be an abnormal inverter, triggering a fault alarm and generating an abnormal event command containing the abnormal inverter information. The fault location module is used to respond to the abnormal event command, calculate the power deviation of each string under the abnormal inverter according to the device topology relationship and the pre-built power conservation constraints, and generate and output the location result command containing the target fault string information. The fault diagnosis and work order generation module is used to perform root cause reasoning on the target fault string based on the location result instruction and in combination with preset rules, and generate diagnostic conclusions and maintenance work orders.

10. An inverter, characterized in that, A photovoltaic power generation anomaly detection system integrating an inverter as described in claim 9.

Images