Inverter fault prediction method and system

Abstract

The invention provides an inverter fault prediction method and system, and belongs to the technical field of inverter fault prediction, and the method comprises the steps: 1, deploying a multi-modal sensor at a key part of an inverter, and collecting the multi-modal data of the inverter; 2, analyzing the multi-modal data of the inverter to obtain a plurality of data features; 3, determining a comprehensive fault coefficient based on all data characteristics and historical data of the inverter; 4, acquiring real-time state data of the inverter, and determining a risk level of the inverter in combination with the comprehensive fault coefficient; and step 5, determining a fault prediction result based on the comprehensive fault coefficient, the real-time state data of the inverter and a preset prediction network, and outputting a fault prediction report in combination with the risk level of the inverter. And the operation reliability of the inverter and the overall efficiency of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of inverter fault prediction, and particularly to an inverter fault prediction method and system. Background Art

[0002] With the wide application of renewable energy, inverters play a key role in power conversion and control. However, inverters may malfunction during operation, affecting the stability and efficiency of the system.

[0003] Existing fault prediction methods mostly rely on a single monitoring means and often cannot comprehensively reflect the operating state of the inverter. For example, some solutions only analyze current and voltage data and lack comprehensive consideration of temperature and environmental conditions, resulting in insufficient prediction accuracy. Based on the existing technical solutions, the lack of comprehensive integration and analysis of multi-modal data leads to insufficient accuracy and timeliness of fault prediction and cannot effectively reduce the economic losses caused by inverter faults.

[0004] Therefore, the present invention provides an inverter fault prediction method and system. Summary of the Invention

[0005] The present invention provides an inverter fault prediction method and system, which significantly improves the accuracy and timeliness of fault prediction by deploying multi-modal sensors on key components and collecting and analyzing various data features in real time. By integrating the comprehensive fault coefficient with the real-time status data, the risk level of the inverter can be dynamically evaluated, and potential faults can be predicted in advance, which not only reduces the maintenance cost and downtime but also improves the operating reliability of the inverter and the overall efficiency of the system.

[0006] The inverter fault prediction method provided by the present invention includes:

[0007] Step 1: Deploy multi-modal sensors on key components of the inverter and collect multi-modal data of the inverter;

[0008] Step 2: Analyze the multi-modal data of the inverter to obtain several data features;

[0009] Step 3: Determine the comprehensive fault coefficient based on all data features and the historical data of the inverter;

[0010] Step 4: Obtain the real-time status data of the inverter and determine the risk level of the inverter in combination with the comprehensive fault coefficient;

[0011] Step 5: Determine the fault prediction result based on the comprehensive fault coefficient, the real-time status data of the inverter, and a preset prediction network, and output a fault prediction report in combination with the risk level of the inverter.

[0012] The inverter fault prediction method provided by the present invention analyzes multi-modal data of the inverter to obtain several data features, including:

[0013] Based on a preset neural network-modal database, obtain a specific neural network for extracting the features of each modal data;

[0014] Extract the data features of each modal data based on the specific neural network;

[0015] Input the features extracted by each neural network into a multi-layer perceptron in the fusion layer for non-linear mapping, and then extract several cross-modal comprehensive features;

[0016] Extract the evolution trend features of the multi-modal data of the inverter in the time dimension through a preset time series neural network;

[0017] Analyze the evolution trend features through a clustering algorithm, and match the analysis results with the preset fault features, and then obtain several abnormal mode features.

[0018] The inverter fault prediction method provided by the present invention inputs the features extracted by each neural network into a multi-layer perceptron in the fusion layer for non-linear mapping, including:

[0019] Connect the modal data features into a high-dimensional input vector based on a preset connection method;

[0020] Perform non-linear mapping on the high-dimensional input vector through a multi-layer perceptron, and obtain several high-order features containing the features of each mode and the correlation information between the features of each mode based on a preset number of activation functions;

[0021] Convert all high-order features into several cross-modal comprehensive features based on a preset method.

[0022] The inverter fault prediction method provided by the present invention determines a comprehensive fault coefficient based on all data features and the historical data of the inverter, including:

[0023] Through a preset adaptive weighting mechanism, obtain the fusion weight corresponding to the output of each neural network according to the real-time deviation degree of each type of data, and then determine the weight of each cross-modal comprehensive feature;

[0024] Determine the comprehensive fault coefficient based on each cross-modal comprehensive feature, the weight of each cross-modal comprehensive feature, the abnormal mode feature, and the historical data of the inverter:

[0025]

[0026] where s iis the comprehensive feature of the i-th cross-modal, and n1 is the number of comprehensive features of cross-modal. is the weight of the comprehensive feature of the i-th cross-modal, m u is the u-th abnormal mode feature, h j is the j-th historical data feature, n2 is the number of abnormal mode features, t is the time difference from the current time to the historical data, and γ is the preset time decay coefficient.

[0027] The inverter fault prediction method provided by the present invention obtains the real-time status data of the inverter and determines the risk level of the inverter in combination with the comprehensive fault coefficient, including:

[0028] Based on a preset monitoring device, obtain the real-time status data of the inverter, analyze the real-time status data of the inverter, and obtain several features;

[0029] Based on the historical data of the inverter, determine the numerical range of each feature in the normal working state;

[0030] According to the historical fault data, determine the range of each feature in the fault state;

[0031] Based on the numerical range of the inverter in the normal working state and the range in the fault state, determine several level risk level thresholds;

[0032] Based on the features of the inverter and the comprehensive fault coefficient, determine the real-time risk coefficient of the inverter;

[0033] Based on the real-time risk coefficient of the inverter and the risk level threshold, determine the risk level of the inverter.

[0034] The inverter fault prediction method provided by the present invention determines the real-time risk coefficient of the inverter based on the features of the inverter and the comprehensive fault coefficient, including:

[0035] Analyze the features of the inverter to obtain several real-time status features of the inverter;

[0036] Based on the real-time status features of the inverter and the comprehensive fault coefficient, determine the real-time risk coefficient of the inverter:

[0037]

[0038] Among them, R1 is the real-time risk coefficient of the inverter, d p is the p-th real-time status feature of the inverter, w p is the weight coefficient corresponding to the p-th real-time status feature of the inverter, d q is the q-th real-time status feature of the inverter, n3 is the number of real-time status features of the inverter, ∝d qis the conversion coefficient of the real-time state feature of the q-th preset inverter, α1 is the preset weight coefficient corresponding to the comprehensive fault coefficient, is the preset interaction coefficient of the real-time state feature of the p-th inverter and the real-time state feature of the q-th inverter, θ is the preset parameter for controlling the smoothness of the exponential function, and ε is the preset weight coefficient of the comprehensive fault coefficient and the deviation of the real-time state feature.

[0039] The inverter fault prediction method provided by the present invention, the fault prediction report includes: risk source, risk level, predicted fault time and type, real-time state data analysis, and preventive maintenance plan.

[0040] The inverter fault warning system provided by the present invention includes:

[0041] Data acquisition module: Deploy multi-modal sensors on key components of the inverter and collect multi-modal data of the inverter.

[0042] Data analysis module: Analyze the multi-modal data of the inverter to obtain several data features.

[0043] Coefficient determination module: Determine the comprehensive fault coefficient based on all data features and historical data of the inverter.

[0044] Level determination module: Obtain the real-time state data of the inverter and determine the risk level of the inverter in combination with the comprehensive fault coefficient.

[0045] Report generation module: Determine the fault prediction result based on the comprehensive fault coefficient, real-time data of the inverter, and the preset prediction network, and output the fault prediction report in combination with the risk level of the inverter.

[0046] Compared with the prior art, the beneficial effects of the present application are as follows:

[0047] By deploying multi-modal sensors on key components, real-time collecting and analyzing various data features, the accuracy and timeliness of fault prediction are significantly improved. By combining the comprehensive fault coefficient with the real-time state data, the risk level of the inverter can be dynamically evaluated, potential faults can be warned in advance, not only the maintenance cost and downtime are reduced, but also the operation reliability of the inverter and the overall efficiency of the system are improved. Description of the Drawings

[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0049] Figure 1It is a schematic flowchart of the inverter fault warning method provided by an embodiment of the present invention;

[0050] Figure 2 It is a schematic structural diagram of the inverter fault warning system provided by an embodiment of the present invention. Detailed implementation manners

[0051] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] Embodiment 1

[0053] The inverter fault prediction method provided by an embodiment of the present invention, as Figure 1 shown, includes:

[0054] Step 1: Deploy multimodal sensors on key components of the inverter and collect multimodal data of the inverter;

[0055] Step 2: Analyze the multimodal data of the inverter to obtain several data features;

[0056] Step 3: Determine a comprehensive fault coefficient based on all data features and historical data of the inverter;

[0057] Step 4: Obtain real-time status data of the inverter and determine the risk level of the inverter in combination with the comprehensive fault coefficient;

[0058] Step 5: Determine a fault prediction result based on the comprehensive fault coefficient, real-time status data of the inverter and a preset prediction network, and output a fault prediction report in combination with the risk level of the inverter.

[0059] In this embodiment, the key components of the inverter refer to the components that play a decisive role in the performance and reliability of the inverter during its operation, including: power switches: electronic components responsible for power conversion and control, transformers: devices used for voltage conversion, which affect the energy transmission efficiency, radiators: components that prevent the inverter from overheating and ensure stable operation, control circuits: circuits that monitor and adjust the operating state of the inverter;

[0060] In this embodiment, a multimodal sensor is a sensor capable of simultaneously collecting multiple types of data, providing comprehensive information, including: a temperature sensor: monitoring the temperature changes of key components of the inverter; a current sensor: measuring the current flow of the inverter in real time; a vibration sensor: detecting mechanical vibrations during equipment operation to identify potential faults; an environmental sensor: monitoring environmental factors around the inverter, such as humidity and air pressure;

[0061] In this embodiment, multimodal data refers to a collection of data collected from different sensors in different formats, which can comprehensively reflect the operating state of the inverter. For example, temperature and current data: combining the temperature and current information during the operation of the inverter; vibration and noise data: analyzing mechanical vibration and noise signals to identify faults; environmental condition data: monitoring the impact of the external environment of the inverter on its performance.

[0062] In this embodiment, data features are indicators or attributes extracted by analyzing multimodal data that help to judge the state of the inverter, including: average temperature: the average value of the inverter temperature within a specific time; current fluctuation amplitude: the amplitude of the change in the current signal, reflecting the load stability; vibration frequency: the key frequency feature for identifying potential mechanical faults; environmental humidity: environmental condition data that affects the performance of the inverter.

[0063] In this embodiment, the comprehensive fault coefficient is calculated based on all data features and historical data, and is an indicator used to measure the fault risk of the inverter.

[0064] In this embodiment, the risk level of the inverter is evaluated according to the comprehensive fault coefficient, indicating the possibility and severity of a fault, including: low risk: the probability of a fault is low and the equipment is operating normally; medium risk: there is a certain possibility of a fault; high risk: the probability of a fault is high and immediate measures need to be taken.

[0065] In this embodiment, the preset prediction network is constructed through machine learning or deep learning algorithms, and is a model for analyzing data and predicting faults, including: a neural network model: predicting the future fault conditions of the inverter by training historical data; a decision tree algorithm: constructing a decision model based on feature data to assist in judging the fault risk.

[0066] The beneficial effects of the above technical solutions are as follows: By deploying multimodal sensors on key components, collecting and analyzing multiple data features in real time, the accuracy and timeliness of fault prediction are significantly improved. By combining the comprehensive fault coefficient with real-time status data, the risk level of the inverter can be dynamically evaluated, and potential faults can be pre-warned in advance, which not only reduces the maintenance cost and downtime, but also improves the operating reliability of the inverter and the overall efficiency of the system.

[0067] Embodiment 2

[0068] The inverter fault prediction method provided by the embodiment of the present invention analyzes multi-modal data of the inverter to obtain a number of data features, including:

[0069] Based on a preset neural network-modal database, obtain a specific neural network for extracting the features of each modal data;

[0070] Based on the specific neural network, extract the data features of each modal data;

[0071] Input the features extracted by each neural network into the multi-layer perceptron in the fusion layer for non-linear mapping, and then extract a number of cross-modal comprehensive features;

[0072] Extract the evolution trend features of the multi-modal data of the inverter in the time dimension through a preset time series neural network;

[0073] Analyze the evolution trend features through a clustering algorithm, and match the analysis results with the preset fault features, so as to obtain a number of abnormal mode features.

[0074] In this embodiment, the preset neural network-modal database is a set containing multiple neural networks for feature extraction of different modal data (such as temperature, current, vibration, etc.), which is used to improve the efficiency and accuracy of data analysis. For example, the temperature feature extraction network: specifically used to extract the features of temperature data, the current feature extraction network: processes current signals and extracts relevant features, and the vibration feature extraction network: a neural network designed for vibration signals.

[0075] In this embodiment, the fusion layer is a layer in the neural network, which is used to integrate the features from different neural networks for subsequent processing. For example, feature splicing: splicing the features extracted from different modalities together by rows or columns, weighted fusion: fusing according to the importance of the features by giving different weights;

[0076] In this embodiment, the multi-layer perceptron is the most basic feedforward neural network, which consists of multiple layers and can realize complex function mapping, including: input layer: receives the input of the fused features, hidden layer: performs non-linear processing of the features, output layer: outputs the comprehensive features or classification results;

[0077] In this embodiment, non-linear mapping refers to the process of converting input data into output data through a neural network, making the relationship between data more complex and flexible. For example, activation functions: such as ReLU, Sigmoid, etc., help to realize non-linear feature extraction, and polynomial regression: non-linearly model the data through polynomial functions;

[0078] In this embodiment, the cross-modal comprehensive feature refers to the feature extracted by fusing different modal data, forming information with a higher dimension. For example, the combined feature of temperature and current: reflecting the comprehensive operating state of the device under specific conditions, and the combined feature of vibration and environmental conditions: helping to judge the impact of the external environment on the inverter.

[0079] In this embodiment, the preset time-series neural network is a neural network specifically used to process time-series data, capable of capturing the laws of data changes over time, including: Long Short-Term Memory Network (LSTM): effectively processing time-series data with long-term dependencies, and Gated Recurrent Unit (GRU): an improved RNN that can better capture time-series features;

[0080] In this embodiment, the evolution trend feature refers to the change laws and trends of the inverter's multi-modal data in the time dimension. For example, the temperature rising trend: a continuous increase in temperature may indicate an increased risk of failure, and the current fluctuation trend: irregular current changes may reflect device abnormalities.

[0081] In this embodiment, the clustering algorithm is an unsupervised learning method used to group data points into similar categories for easy analysis and pattern recognition. For example, K-means clustering: dividing data into K clusters by calculating distances, and hierarchical clustering: establishing a tree structure to gradually merge or split the data set;

[0082] In this embodiment, the preset fault features are key indicators defined based on historical data and expert experience, used to judge the fault state of the inverter, including: high-temperature alarm threshold: exceeding a certain temperature value is regarded as a fault risk, and current mutation pattern: a mutated current signal indicates a possible fault occurrence;

[0083] In this embodiment, the abnormal mode feature refers to the feature indicating potential faults or abnormal states of the device mined by analyzing data. For example, a high vibration peak within a short period: may indicate a mechanical fault, and a frequent temperature fluctuation pattern: indicates system instability.

[0084] The beneficial effects of the above technical solutions are as follows: By analyzing multi-modal data, rich data features are extracted. The neural network is used to extract features for each modal data, and then non-linear mapping is performed through a multi-layer perceptron to obtain cross-modal comprehensive features. At the same time, a time-series neural network is used to capture the evolution trend of data in the time dimension, and the clustering algorithm is used to analyze the evolution features and match them with the preset fault features, thereby identifying abnormal modes, improving the comprehensiveness and accuracy of fault prediction, and promoting the realization of intelligent monitoring of the inverter.

[0085] Embodiment 3

[0086] The inverter fault prediction method provided by the embodiment of the present invention inputs the features extracted by each neural network into a multi-layer perceptron in a fusion layer for non-linear mapping, including:

[0087] Connect the feature of each modality data into a high-dimensional input vector based on a preset connection method;

[0088] Perform non-linear mapping on the high-dimensional input vector through a multi-layer perceptron, and obtain a number of high-order features containing the features of each modality and the correlation information between the features of each modality based on a number of preset activation functions;

[0089] Convert all high-order features into a number of cross-modal comprehensive features based on a preset method.

[0090] In this embodiment, the preset connection method refers to the rule or algorithm adopted when combining the feature of each modality data, aiming to effectively integrate information from different sources, including: serial connection: splicing different modality features in sequence into a high-dimensional vector, weighted connection: adding after giving different weights according to the importance of the features to form a high-dimensional input vector;

[0091] In this embodiment, the high-dimensional input vector is a vector formed by connecting multiple modality features, usually having multiple dimensions and being able to represent rich information. For example, the combination of temperature, vibration and current features: forming a vector containing multiple values, such as [temperature value, vibration amplitude, current value], the combination of environmental features and working features: such as [environmental humidity, temperature, load current].

[0092] In this embodiment, the number of preset activation functions is a function used in a multi-layer perceptron, which determines the output of a neuron, thereby introducing non-linear characteristics. For example, ReLU (Rectified Linear Unit): the output is the larger value of the input value and zero, and is often used in the hidden layer, Sigmoid function: the output value range is between 0 and 1, suitable for binary classification tasks, Tanh function: the output range is between -1 and 1, suitable for features with negative values;

[0093] In this embodiment, the high-order feature is a feature obtained through non-linear mapping, usually containing the deep information of the features of each modality and their mutual relationship. For example, feature interaction: such as the feature jointly affected by temperature and current, characterizing the relationship between the two, combined feature: such as the product of vibration and current, reflecting a certain potential fault mode.

[0094] In this embodiment, the preset method refers to the specific algorithm or steps used to process and transform high-order features to obtain the final comprehensive features. For example, principal component analysis (PCA): reducing the dimension of high-order features and extracting the most important components, linear regression: establishing a model based on high-order features for prediction;

[0095] In this embodiment, converting high-order features into cross-modal comprehensive features means integrating the processed multi-modal features into new features for subsequent analysis or prediction. For example, the comprehensive health index: an indicator calculated based on multiple high-order features, used to characterize the overall health status of the inverter; the fault prediction indicator: an indicator constructed through comprehensive features, indicating the probability of the inverter failing.

[0096] The beneficial effects of the above technical solution are as follows: By inputting multi-modal features into a multi-layer perceptron, non-linear mapping and feature fusion are achieved. A high-dimensional input vector is constructed using a preset connection method, and high-order features are extracted through the multi-layer perceptron to capture complex correlation information between modalities, improving the accuracy and robustness of fault prediction, enabling the model to more effectively identify potential faults, and promoting the progress of intelligent monitoring and maintenance of inverters.

[0097] Embodiment 4

[0098] The inverter fault prediction method provided by the embodiment of the present invention determines a comprehensive fault coefficient based on all data features and the historical data of the inverter, including:

[0099] Through a preset adaptive weighting mechanism, according to the real-time deviation degree of various types of data, the fusion weight corresponding to each neural network output is obtained, and then the weight of each cross-modal comprehensive feature is determined;

[0100] Based on each cross-modal comprehensive feature, the weight of each cross-modal comprehensive feature, the abnormal mode feature, and the historical data of the inverter, a comprehensive fault coefficient is determined:

[0101]

[0102] where s i is the i-th cross-modal comprehensive feature, n1 is the number of cross-modal comprehensive features, is the weight of the i-th cross-modal comprehensive feature, m u is the u-th abnormal mode feature, h j is the j-th historical data feature, n2 is the number of abnormal mode features, t is the time difference between the current time and the historical data, and γ is a preset time decay coefficient.

[0103] In this embodiment, the preset adaptive weighting mechanism is an algorithm that dynamically adjusts weights according to the characteristics of current data, which can assign appropriate weights to different features, thereby improving the prediction accuracy of the model. For example, adaptive weight adjustment: when the system temperature rises abnormally, the weight of the temperature feature is increased to reflect its importance at the current moment; weighting based on the degree of data abnormality: when the vibration signal deviates greatly from the normal range, the adaptive weighting mechanism will increase the weight of the vibration feature to more prominently show the abnormality of the current data;

[0104] In this embodiment, the real-time deviation degree refers to the deviation degree between the current data value and the standard value or historical mean value, which is used to judge the abnormality of data. For example, temperature deviation: the current temperature is 10 degrees higher than the historical mean value, indicating a large deviation; current deviation: the current fluctuation range is large, deviating from the normal range by more than 50%, indicating a potential fault.

[0105] In this embodiment, the fusion weight is a weight value calculated through an adaptive weighting mechanism and assigned to different comprehensive features to indicate the importance of each feature at the current moment. For example, temperature feature weight: when the device temperature is high, a larger fusion weight is given to the temperature feature; vibration feature weight: when the vibration signal fluctuates significantly, the fusion weight of this feature is increased to make the prediction result more reflect the current fault risk.

[0106] In this embodiment, the weight of the comprehensive feature is the contribution value of each cross-modal comprehensive feature during the fusion process, which is assigned according to the adaptive weighting mechanism. For example, the weight of the temperature-current comprehensive feature: during the operation of the device, if the combined feature of temperature and current fluctuates greatly, a higher weight is given to it; the weight of the current-vibration comprehensive feature: when this combined feature shows an abnormality, its weight is increased to reflect a possible fault.

[0107] In this embodiment, the preset time decay coefficient is a parameter used to control the influence of historical data. The longer the time, the smaller the influence of the data on the current prediction. For example, a smaller decay coefficient for short-term data: the temperature data within the past 1 hour decays less because it is more representative; a larger decay coefficient for long-term data: a larger decay coefficient is applied to the historical data 3 months ago to weaken its influence on the current prediction and prevent outdated data from affecting the prediction.

[0108] The beneficial effects of the above technical solutions are as follows: Through the adaptive weighting mechanism, various data features and historical data are fused to effectively determine the comprehensive fault coefficient. By dynamically obtaining the fusion weight, it is ensured that the weights of cross-modal features accurately reflect the real-time deviation, enhancing the sensitivity of fault identification. Combining the abnormal mode features and the time decay coefficient can adapt to the state changes of the inverter in real time, significantly improving the accuracy and timeliness of fault prediction, and promoting the intelligentization and optimized management of inverter maintenance.

[0109] Embodiment 5

[0110] The inverter fault prediction method provided by the embodiment of the present invention obtains the real-time state data of the inverter and determines the risk level of the inverter in combination with the comprehensive fault coefficient, including:

[0111] Based on a preset monitoring device, obtain the real-time state data of the inverter, analyze the real-time state data of the inverter, and obtain several features.

[0112] Determine the numerical range of each feature in the normal operating state based on the historical data of the inverter;

[0113] Determine the range of each feature in the fault state according to the historical fault data;

[0114] Determine several hierarchical risk level thresholds based on the numerical range of the inverter in the normal operating state and the range in the fault state;

[0115] Determine the real-time risk coefficient of the inverter based on the features of the inverter and the comprehensive fault coefficient;

[0116] Determine the risk level of the inverter based on the real-time risk coefficient of the inverter and the risk level threshold.

[0117] In this embodiment, the preset monitoring device refers to a device or system for real-time collection of inverter status data, usually configured with sensors, data acquisition modules, etc., including: temperature sensor: real-time monitoring of the internal temperature of the inverter to ensure it is within a safe range, current sensor: recording the current changes of the inverter to help identify abnormal fluctuations, vibration sensor: detecting the vibration during the operation of the inverter to judge whether there are mechanical faults;

[0118] In this embodiment, the features of the inverter refer to the key indicators extracted from the real-time status data for describing the working state of the inverter, including: output voltage: reflecting the ability of the inverter to convert electrical energy, which should be within the set range under normal circumstances, operating frequency: referring to the frequency range at which the inverter operates, abnormal fluctuations may indicate a fault, heat dissipation: related to the heat dissipation performance of the inverter, excessive heat dissipation may cause the device to overheat.

[0119] In this embodiment, the hierarchical risk level threshold is a numerical range set according to the features and historical data of the inverter for dividing different risk levels. For example, low-risk threshold: when the operating temperature of the inverter is below 60 °C, it is defined as a low-risk state, medium-risk threshold: when the temperature is between 60 °C and 80 °C, it is regarded as a medium-risk and requires enhanced monitoring, high-risk threshold: when it exceeds 80 °C, it is marked as a high-risk state and immediate measures need to be taken.

[0120] In this embodiment, the real-time risk coefficient is a numerical value calculated based on the features of the inverter and the comprehensive fault coefficient, used to represent the current risk level of the inverter. If the comprehensive fault coefficient is 0.75 and the output voltage deviates from the normal range by 20%, the real-time risk coefficient may be 0.9, indicating a high risk. When the real-time risk coefficient is 0.4, it means the inverter is operating stably with a low risk; while 0.8 means close monitoring is required.

[0121] The beneficial effects of the above technical solution are as follows: By analyzing real-time status data and comparing historical data, an efficient risk level assessment mechanism is provided. By establishing the numerical ranges of characteristics in normal and faulty states and formulating risk level thresholds, the risk assessment becomes more accurate. The calculation of the real-time risk coefficient in combination with the comprehensive fault coefficient can timely identify potential faults of the inverter, optimize maintenance decisions, significantly reduce the likelihood of faults occurring, improve the safety and reliability of the equipment, and realize the intelligence of fault prediction.

[0122] Embodiment 6

[0123] The inverter fault prediction method provided by the embodiment of the present invention determines the real-time risk coefficient of the inverter based on the characteristics of the inverter and the comprehensive fault coefficient, including:

[0124] Analyze the characteristics of the inverter to obtain several real-time status characteristics of the inverter;

[0125] Determine the real-time risk coefficient of the inverter based on the real-time status characteristics of the inverter and the comprehensive fault coefficient:

[0126]

[0127] where, R1 is the real-time risk coefficient of the inverter, d p is the p-th real-time status characteristic of the inverter, w p is the weight coefficient corresponding to the p-th real-time status characteristic of the inverter, d q is the q-th real-time status characteristic of the inverter, n3 is the number of real-time status characteristics of the inverter, ∝d q is the preset conversion coefficient of the q-th real-time status characteristic of the inverter, α1 is the preset weight coefficient corresponding to the comprehensive fault coefficient, is the preset interaction coefficient of the p-th real-time status characteristic and the q-th real-time status characteristic of the inverter, θ is the preset parameter for controlling the smoothness of the exponential function, and ε is the preset weight coefficient of the deviation between the comprehensive fault coefficient and the real-time status characteristic.

[0128] In this embodiment, the preset conversion coefficient is a parameter used to convert a certain real-time status characteristic of the inverter into a comparable standardized value, usually set according to the physical meaning and importance of the characteristic. For example, the temperature conversion coefficient: If the temperature range of the inverter is from 0°C to 100°C, a conversion coefficient of 0.01 may be set to convert the actual temperature value into a value between 0 and 1 for easy comparison with other characteristics. The current conversion coefficient: Assuming the normal range of the current is from 0 to 50A, a conversion coefficient of 0.02 can be set to map the current value to the standard range of 0 to 1 for easy analysis;

[0129] In this embodiment, the preset interaction coefficient is a parameter used to describe the degree of mutual influence between different real-time state characteristics, reflecting the importance of the relationship between characteristics in risk calculation. For example, the interaction coefficient between temperature and current: in some cases, the increase in temperature will affect the stability of the current, and the interaction coefficient may be set to 0.5, indicating a strong mutual relationship. The interaction coefficient between current and vibration: if the increase in inverter current is usually accompanied by an increase in vibration, an interaction coefficient of 0.3 is set, indicating that their mutual influence is small but still exists.

[0130] The beneficial effects of the above technical solutions are as follows: Through the calculation of the real-time risk coefficient, based on the inverter characteristics and the comprehensive fault coefficient, accurate risk assessment is provided. By integrating multiple state characteristics and weight coefficients, it can dynamically reflect the operating conditions of the inverter, effectively identify potential faults. Through the preset conversion and interaction coefficients, the flexibility and accuracy of risk assessment are ensured, thereby improving the maintenance efficiency, reducing the fault downtime, and realizing the intelligentization and optimization of inverter management.

[0131] Embodiment 7

[0132] The inverter fault prediction method and fault prediction report provided by the embodiments of the present invention include: risk sources, risk levels, the time and type of predicted faults, real-time state data analysis, and preventive maintenance plans.

[0133] In this embodiment, the risk sources are specific factors or conditions identified that may cause faults, such as excessive temperature, current fluctuations, equipment aging, etc. These risk sources help the staff to clarify the potential causes of faults;

[0134] In this embodiment, the risk level is the severity of the risk evaluated based on the real-time state and historical data of the inverter, usually divided into levels such as low, medium, and high. If the temperature and current are abnormal, it may be evaluated as a high risk, indicating that immediate inspection and intervention are required;

[0135] In this embodiment, the time and type of predicted faults are based on data analysis to predict the time point and specific type of possible faults. For example, the report may indicate that a current fault is expected to occur within the next week, and the specific type may be a short circuit or overload;

[0136] In this embodiment, the real-time state data analysis is a detailed analysis of the current operating state of the inverter, including real-time data of parameters such as temperature, current, and vibration. For example, the report may include charts or data summaries showing the operating trends of the inverter in the past 24 hours to analyze abnormal fluctuations;

[0137] In this embodiment, the preventive maintenance plan is a maintenance measure and schedule formulated for the identified risk sources and predicted faults. If it is found that the inverter temperature is too high, it is planned to clean and inspect within the next 48 hours to ensure the normal operation of the heat dissipation system.

[0138] The beneficial effects of the above technical solution are as follows: By generating a detailed fault prediction report, the accuracy and timeliness of fault identification are significantly improved. By clarifying the risk sources and levels, predicting the time and type of faults, and combining real-time status data analysis, potential problems can be effectively foreseen, reducing the occurrence of sudden faults. At the same time, the formulation of the preventive maintenance plan ensures the stability and reliability of the equipment, thereby optimizing the operation efficiency and maintenance cost of the inverter and achieving the innovative effect of intelligent management.

[0139] Embodiment 8

[0140] The inverter fault prediction system provided by the embodiment of the present invention, as Figure 2 shown, includes:

[0141] Data acquisition module: Deploy multi-modal sensors on the key components of the inverter and collect the multi-modal data of the inverter;

[0142] Data analysis module: Analyze the multi-modal data of the inverter to obtain several data features;

[0143] Coefficient determination module: Determine the comprehensive fault coefficient based on all data features and the historical data of the inverter;

[0144] Level determination module: Obtain the real-time status data of the inverter and determine the risk level of the inverter in combination with the comprehensive fault coefficient;

[0145] Report generation module: Determine the fault prediction result based on the comprehensive fault coefficient, the real-time data of the inverter, and the preset prediction network, and output the fault prediction report in combination with the risk level of the inverter.

[0146] The beneficial effects of the above technical solution: By deploying multi-modal sensors on key components, collecting and analyzing various data features in real time, the accuracy and timeliness of fault prediction are significantly improved. By combining the comprehensive fault coefficient with the real-time status data, the risk level of the inverter can be dynamically evaluated, and potential faults can be warned in advance, which not only reduces the maintenance cost and downtime, but also improves the operation reliability of the inverter and the overall efficiency of the system.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An inverter fault prediction method, characterized in that: include: Step 1: Deploy multimodal sensors on key components of the inverter and collect multimodal data of the inverter; Step 2: Analyze the multimodal data of the inverter to obtain several data features; Step 3: Determine the comprehensive fault coefficient based on all data features and inverter historical data; Step 4: Obtain the real-time status data of the inverter and determine the risk level of the inverter in combination with the comprehensive fault coefficient; Step 5: Determine the fault prediction result based on the comprehensive fault coefficient, the real-time status data of the inverter and the preset prediction network, and output the fault prediction report in combination with the risk level of the inverter.

2. The inverter fault prediction method according to claim 1, characterized in that: Analyze the multimodal data of the inverter to obtain several data features, including: Acquire a specific neural network for extracting features of each modality data based on a preset neural network-modality database; Extract data features of each modality data based on a specific neural network; The features extracted by each neural network are input into the multi-layer perceptron in the fusion layer for nonlinear mapping, thereby extracting several cross-modal comprehensive features; The evolution trend characteristics of the inverter's multimodal data in the time dimension are extracted through a preset time series neural network; The evolution trend characteristics are analyzed through clustering algorithm, and the analysis results are matched with the preset fault characteristics to obtain several abnormal pattern characteristics.

3. The inverter fault prediction method according to claim 2, characterized in that: The extracted features of each neural network are input into the multi-layer perceptron in the fusion layer for nonlinear mapping, including: Connect the data features of each modality into a high-dimensional input vector based on a preset connection method; The high-dimensional input vector is nonlinearly mapped through a multi-layer perceptron, and a number of high-order features including the features of each modality and the correlation information between the features of each modality are obtained based on a number of preset activation functions; Based on the preset method, all high-order features are converted into several cross-modal comprehensive features.

4. The inverter fault prediction method according to claim 2, characterized in that: Determine the comprehensive fault coefficient based on all data features and inverter historical data, including: By presetting an adaptive weighting mechanism, the fusion weight corresponding to each neural network output is obtained according to the real-time deviation degree of each type of data, and then the weight of each cross-modal comprehensive feature is determined; The comprehensive fault coefficient is determined based on the comprehensive features of each cross-modality, the weight of each cross-modality comprehensive feature, the abnormal mode features, and the inverter historical data: Among them, s i is the i-th cross-modal comprehensive feature, n1 is the number of cross-modal comprehensive features, is the weight of the comprehensive feature of the i-th cross-modality, m u is the u-th abnormal pattern feature, h j is the jth historical data feature, n2 is the number of abnormal pattern features, t is the time difference between the current time and the historical data, and γ is the preset time attenuation coefficient.

5. The inverter fault prediction method according to claim 1, characterized in that: Obtain the real-time status data of the inverter and determine the risk level of the inverter based on the comprehensive fault coefficient, including: Acquire real-time status data of the inverter based on a preset monitoring device, analyze the real-time status data of the inverter, and acquire a number of features; Determine the value range of each feature under normal working conditions based on inverter historical data; Determine the range of each feature under fault conditions based on historical fault data; Determine several levels of risk level thresholds based on the value range of the inverter under normal working conditions and the range under fault conditions; Determine the real-time risk factor of the inverter based on the characteristics of the inverter and the comprehensive fault factor; The risk level of the inverter is determined based on the real-time risk coefficient of the inverter and the risk level threshold.

6. The inverter fault prediction method according to claim 4, characterized in that: The real-time risk factor of the inverter is determined based on the characteristics of the inverter and the comprehensive fault factor, including: Analyze the characteristics of the inverter to obtain several real-time status characteristics of the inverter; Determine the real-time risk factor of the inverter based on the real-time status characteristics of the inverter and the comprehensive fault coefficient: Among them, R1 is the real-time risk factor of the inverter, d p is the real-time state feature of the pth inverter, w p is the weight coefficient corresponding to the real-time state feature of the pth inverter, d q is the real-time state feature of the qth inverter, n3 is the number of real-time state features of the inverter, ∝d q is the preset conversion coefficient of the real-time state characteristic of the qth inverter, α1 is the preset weight coefficient corresponding to the comprehensive fault coefficient, is the preset interaction coefficient of the real-time state characteristic of the pth inverter and the real-time state characteristic of the qth inverter, θ is the preset parameter for controlling the smoothness of the exponential function, and ε is the preset weight coefficient of the comprehensive fault coefficient and the real-time state characteristic deviation.

7. The inverter fault prediction method according to claim 1, characterized in that: Failure prediction report, including: risk source, risk level, predicted time and type of failure, real-time status data analysis and preventive maintenance plan.

8. An inverter fault prediction system, characterized in that: include: Data acquisition module: deploy multimodal sensors in key components of the inverter and collect multimodal data of the inverter; Data analysis module: analyzes the multi-modal data of the inverter and obtains several data features; Coefficient determination module: determines the comprehensive fault coefficient based on all data features and inverter historical data; Level determination module: obtains the real-time status data of the inverter and determines the risk level of the inverter in combination with the comprehensive fault coefficient; Report generation module: Determines the fault prediction result based on the comprehensive fault coefficient, inverter real-time data and the preset prediction network, and outputs the fault prediction report in combination with the risk level of the inverter.

Images