Power converter intelligent control method and system with fault prediction function

Abstract

The invention relates to the technical field of power system monitoring, and particularly discloses a power converter intelligent control method and system with a fault prediction function, and the method comprises the steps: obtaining the real-time operation data of a power converter, carrying out the fault prediction according to the real-time operation data, and generating a fault prediction result which comprises a normal state and an abnormal state; if the fault result is abnormal, generating an intelligent regulation and control instruction according to the fault prediction result, and regulating and controlling the power supply variator according to the intelligent regulation and control instruction; and acquiring the adjusted real-time operation data of the power converter. According to the method, a complete system of real-time data acquisition, fault prediction, intelligent regulation and control and closed-loop feedback is constructed, a closed-loop feedback mechanism adapts to equipment aging and working condition changes through data iterative optimization after regulation and control, prediction precision and regulation and control effectiveness are continuously guaranteed, the sudden failure shutdown risk is remarkably reduced, and the system is safe and reliable. The operation stability and efficiency of the power converter are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system monitoring, in particular to a power converter intelligent control method and system with fault prediction. BACKGROUND

[0002] As the core equipment for voltage conversion and energy transmission in power electronic systems, power converters are widely used in industrial control, new energy power generation, rail transit and other fields, and their operation reliability directly determines the working stability of downstream loads.

[0003] In power systems, the protection mechanism of power converters mostly adopts fixed action logic (such as directly cutting off power when overcurrent occurs), without considering the specificity of fault types and the difference of severity levels. For example, the core requirement of electrolytic capacitor aging is to delay electrolyte evaporation through cooling to extend the service life, while MOSFET thermal failure needs to limit load current to reduce power consumption to avoid thermal breakdown, but the traditional method triggers unified power-off protection for both types of faults, which not only greatly reduces system availability (unnecessary downtime), but also fails to delay the device degradation process, making it difficult to capture such trend changes, resulting in failure to identify the evolution of faults in advance, ultimately causing sudden failure, causing unplanned downtime of equipment, and even possible single device failure diffusion forming a chain reaction, threatening the stable operation of the entire power system. SUMMARY

[0004] The present application aims to provide a power converter intelligent control method and system with fault prediction to solve the technical problems raised in the background.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] A power converter intelligent control method with fault prediction, comprising:

[0007] Obtaining real-time operation data of the power converter, and performing fault prediction according to the real-time operation data to generate a fault prediction result, wherein the fault prediction result includes normal and abnormal;

[0008] If the fault result is abnormal, generating an intelligent control instruction according to the fault prediction result, and controlling the power converter according to the intelligent control instruction;

[0009] Obtaining real-time operation data of the adjusted power converter, and returning to the step of performing fault prediction according to the real-time operation data.

[0010] Preferably, the step of obtaining real-time operation data of the power converter and performing fault prediction according to the real-time operation data to generate a fault prediction result comprises:

[0011] Obtaining real-time operation data of the power converter, wherein the real-time operation data comprises device operation parameters and electrical characteristic parameters;

[0012] Performing fault prediction according to the device operation parameters and the electrical characteristic parameters to obtain a prediction value;

[0013] Determining whether the prediction value exceeds a preset value;

[0014] If the prediction value exceeds the preset value, the fault prediction result is abnormal;

[0015] If the prediction value does not exceed the preset value, the fault prediction result is normal.

[0016] Preferably, the step of performing fault prediction according to the device operation parameters and the electrical characteristic parameters to obtain a prediction value comprises:

[0017] Extracting features according to the device operation parameters to obtain device operation fault feature information;

[0018] Obtaining a device feature vector according to the device operation fault feature information;

[0019] Obtaining electrical abnormality feature information according to the electrical characteristic parameters;

[0020] Obtaining an electrical abnormality feature vector according to the electrical abnormality feature information;

[0021] Obtaining device-electrical abnormality correlation information according to the device feature vector and the electrical abnormality feature vector;

[0022] Obtaining a full-cycle historical fault database, and extracting correlation coefficient threshold intervals corresponding to various fault types;

[0023] Establishing a fault type-correlation coefficient interval mapping table according to the correlation coefficient threshold intervals;

[0024] Obtaining a fault type interval according to the device-electrical abnormality correlation information and the fault type-correlation coefficient interval mapping table;

[0025] Obtaining a historical fault occurrence frequency from the full-cycle historical fault database according to the fault type interval;

[0026] Obtaining a prediction value according to the historical fault occurrence frequency.

[0027] Preferably, the step of generating an intelligent control instruction according to the fault prediction result comprises:

[0028] Obtaining fault type information and a fault severity level according to the fault prediction result;

[0029] Obtaining fault original operation data according to the fault type information, and extracting fault original operation features;

[0030] obtaining device fault state data according to the fault original operation characteristic;

[0031] obtaining device fault trend according to the device fault state data;

[0032] obtaining hierarchical regulation trigger condition according to the device fault trend and fault severity level, and obtaining hierarchical regulation information according to the hierarchical regulation trigger condition;

[0033] generating intelligent regulation instruction according to the hierarchical regulation information.

[0034] Preferably, the step of obtaining hierarchical regulation trigger condition according to the device fault trend and fault severity level, and obtaining hierarchical regulation information according to the hierarchical regulation trigger condition, comprises:

[0035] obtaining hierarchical regulation trigger threshold according to the device fault trend and fault severity level;

[0036] establishing threshold-regulation level mapping rule according to the hierarchical regulation trigger threshold;

[0037] obtaining device operation data, and comparing the device operation data with the hierarchical regulation trigger threshold to obtain comparison result;

[0038] obtaining corresponding regulation level according to the comparison result;

[0039] obtaining preset hierarchical regulation measure library according to the regulation level;

[0040] obtaining regulation project item according to the regulation measure library, and generating hierarchical regulation information according to the regulation project item.

[0041] Preferably, the step of obtaining regulation project item according to the regulation measure library, and generating hierarchical regulation information according to the regulation project item, comprises:

[0042] generating search tag according to fault type information and fault severity level;

[0043] calling matched regulation project item from the regulation measure library according to the search tag;

[0044] obtaining execution object information, operation instruction information and constraint condition information according to the called regulation project item;

[0045] generating hierarchical regulation information according to the execution object information, operation instruction information and constraint condition information.

[0046] Preferably, the step of generating search tag according to fault type information and fault severity level, comprises:

[0047] extracting fault associated component information and core fault features according to the fault type information;

[0048] generating fault type feature labels according to the fault associated components and the core fault features;

[0049] obtaining fault level identification labels according to the fault severity levels;

[0050] performing semantic association to obtain label associated content according to the fault level identification labels and the fault type feature labels;

[0051] generating retrieval labels according to the label associated content.

[0052] The application further provides a power converter intelligent control system with fault prediction, comprising:

[0053] a real-time operation data acquisition module, configured to acquire real-time operation data of the power converter and perform fault prediction according to the real-time operation data to generate a fault prediction result, wherein the fault prediction result comprises normal and abnormal;

[0054] a control instruction generation module, configured to determine that the fault result is abnormal, generate intelligent control instructions according to the fault prediction result, and control the power converter according to the intelligent control instructions;

[0055] an adjustment data acquisition module, configured to acquire real-time operation data of the adjusted power converter and return to the step of performing fault prediction according to the real-time operation data.

[0056] Preferably, the control instruction generation module comprises:

[0057] a real-time operation data acquisition unit, configured to acquire real-time operation data of the power converter, wherein the real-time operation data comprises device operation parameters and electrical characteristic parameters;

[0058] a prediction value unit, configured to perform fault prediction according to the device operation parameters and the electrical characteristic parameters to obtain a prediction value;

[0059] a judgment acquisition unit, configured to judge whether the prediction value exceeds a preset value;

[0060] if the prediction value exceeds the preset value, the fault prediction result is abnormal;

[0061] if the prediction value does not exceed the preset value, the fault prediction result is normal.

[0062] Preferably, the control instruction generation module further comprises:

[0063] The device operation fault feature information acquisition unit is configured to extract device operation fault feature information according to the device operation parameters;

[0064] The device feature vector acquisition unit is configured to acquire a device feature vector according to the device operation fault feature information;

[0065] The electrical abnormality feature information acquisition unit is configured to acquire electrical abnormality feature information according to the electrical feature parameters;

[0066] The electrical abnormality feature vector acquisition unit is configured to acquire an electrical abnormality feature vector according to the electrical abnormality feature information;

[0067] The device-electrical abnormality correlation information acquisition unit is configured to acquire device-electrical abnormality correlation information according to the device feature vector and the electrical abnormality feature vector;

[0068] The correlation coefficient threshold interval extraction unit is configured to acquire a full-cycle historical fault database and extract correlation coefficient threshold intervals corresponding to various fault types;

[0069] The fault type-correlation coefficient interval mapping table establishment unit is configured to establish a fault type-correlation coefficient interval mapping table according to the correlation coefficient threshold intervals;

[0070] The fault type interval acquisition unit is configured to acquire a fault type interval according to the device-electrical abnormality correlation information and the fault type-correlation coefficient interval mapping table;

[0071] The historical fault occurrence frequency acquisition unit is configured to acquire a historical fault occurrence frequency from the full-cycle historical fault database according to the fault type interval;

[0072] The prediction value acquisition unit is configured to acquire a prediction value according to the historical fault occurrence frequency.

[0073] The application has the beneficial effects that: the application realizes the advanced identification and accurate intervention of power converter faults by constructing a complete system of "real-time data acquisition-fault prediction-intelligent regulation-closed-loop feedback". The multi-dimensional acquisition device operating parameters and electrical characteristic parameters are combined with feature extraction and historical fault data correlation analysis to capture early trend signals of progressive faults such as electrolytic capacitor aging and MOSFET thermal failure, identify faults earlier than traditional threshold alarms, and improve prediction accuracy. Based on the grading regulation strategy of fault type specificity and severity difference, the differentiated measures (such as cooling to delay capacitor aging, current limiting to reduce MOSFET power consumption) can be matched to balance system reliability and availability, and avoid unnecessary downtime caused by traditional fixed protection actions. The closed-loop feedback mechanism iteratively optimizes the data after regulation to adapt to device aging and working condition changes, continuously ensures the prediction accuracy and regulation effectiveness, significantly reduces the risk of sudden failure downtime, and improves the running stability and efficiency of the power converter. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 The method flowchart of an embodiment of the application is shown.

[0075] Figure 2 The system structure diagram of an embodiment of the application is shown.

[0076] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0077] It should be understood that the specific embodiments described herein are only used to explain the application and not to limit the application.

[0078] As shown in the drawings, the application provides a power converter intelligent control method with fault prediction, which comprises: Figure 1

[0079] S1, acquiring real-time running data of the power converter, and performing fault prediction according to the real-time running data to generate a fault prediction result, wherein the fault prediction result includes normal and abnormal;

[0080] S2, if the fault result is abnormal, generating an intelligent regulation instruction according to the fault prediction result, and regulating the power converter according to the intelligent regulation instruction;

[0081] S3, acquiring the real-time running data of the adjusted power converter, and returning to the step of performing fault prediction according to the real-time running data.

[0082] ​As described in steps S1-S3 above, the application realizes dynamic monitoring and precise intervention on the running state of the power converter by constructing a complete system of "real-time data acquisition-fault prediction-smart regulation-closed-loop feedback". The steps include the following: First, real-time running data of the power converter is collected by online monitoring equipment, including device running parameters (such as equivalent series resistance ESR and temperature of electrolytic capacitor, junction temperature and on-state voltage drop of MOSFET, magnetic core temperature of inductor) and electrical characteristic parameters (such as average value and ripple of output voltage, effective value of load current, input power, power factor and output voltage frequency spectrum characteristics). The collected data is preprocessed by filtering and normalization, and is used as the input of fault prediction. Second, the data is analyzed based on feature extraction and machine learning model to generate fault prediction results. Specifically, by extracting time domain features of device running parameters and frequency domain features of electrical characteristic parameters, a multi-dimensional feature vector is constructed, and a fault type-correlation coefficient interval mapping table is established by combining the correlation coefficient threshold interval in the full-cycle historical fault database, so as to determine the fault type and severity level. If the prediction result is abnormal, smart regulation instructions are generated according to the fault type and severity level, and the pre-set regulation measure library is matched by a hierarchical regulation strategy. The regulation measures include starting of redundant heat dissipation module, PWM duty cycle fine tuning, load limitation, standby module switching, etc. The instructions are transmitted to the execution mechanism through the communication protocol to complete the operation. Finally, the real-time running data is re-collected after the regulation execution, the effectiveness of the measures is evaluated and fed back to the fault prediction link to form a closed-loop optimization mechanism.

[0083] In one embodiment, the step S1 of acquiring real-time running data of the power converter and performing fault prediction according to the real-time running data to generate a fault prediction result includes:

[0084] S101, acquiring real-time running data of the power converter, wherein the real-time running data includes device running parameters and electrical characteristic parameters;

[0085] S102, performing fault prediction according to the device running parameters and the electrical characteristic parameters to obtain a prediction value;

[0086] S103, judging whether the prediction value exceeds a preset value;

[0087] If the prediction value exceeds the preset value, the fault prediction result is abnormal;

[0088] If the prediction value does not exceed the preset value, the fault prediction result is normal.

[0089] As described in steps S101-S103 above, the present application collects real-time operation data of the power converter through online monitoring equipment, covering device operation parameters (such as equivalent series resistance ESR and temperature of electrolytic capacitor, junction temperature and on-state voltage drop of MOSFET, magnetic core temperature of inductor) and electrical characteristic parameters (such as average value and ripple of output voltage, effective value of load current, input power, power factor and output voltage frequency spectrum characteristics). Device operation parameters are obtained by temperature sensors, ripple analysis circuits and Hall sensors and other tools, wherein the equivalent series resistance calculation formula is:

[0090]

[0091] Wherein K(S) represents the equivalent series resistance, ΔU represents the peak-to-peak value of ripple voltage, R represents the load resistance, and ΔI represents the peak-to-peak value of inductor ripple current (based on circuit principles, the peak-to-peak value of output ripple voltage is superimposed by inductor voltage drop and equivalent series resistance ESR voltage drop, and after simplification, the inductor impedance is ignored (high frequency equivalent series resistance ESR dominates), and the following formula is derived: The formula introduces the load resistance R to adapt to different load conditions; the electrical characteristic parameters are synchronously collected by voltage sensors, current sensors and high-speed data acquisition cards, and are preprocessed by filtering and normalization to eliminate interference and dimension difference. Secondly, based on feature extraction and prediction model, fault prediction is carried out, specifically including: extracting time domain features and correlation features from device operation parameters, extracting features such as high-frequency component energy proportion and harmonic distortion rate mutation index from electrical characteristic parameters, constructing a multi-dimensional feature vector, and establishing a fault type-correlation coefficient interval mapping table combined with the correlation coefficient threshold interval in the historical fault database, and finally determining the fault state by comparing the prediction value with the dynamic preset value. If the prediction value exceeds the preset value, it is determined as "abnormal"; if it does not exceed, it is determined as "normal". Through the design of dynamic preset value, false alarm and missed alarm of fixed threshold are avoided, multi-parameter joint determination logic reduces the risk of misjudgment caused by single parameter fluctuation, and the closed-loop feedback mechanism iteratively optimizes the prediction model by supplementing the regulated operation data, adapts to the aging characteristics and working condition changes of the equipment, for example, corrects the nonlinear growth law of ESR in high temperature environment, and ensures the prediction accuracy in long-term operation. This system solves the problem of sudden failure caused by the inability of traditional fixed threshold alarm to identify trend signals, improves the foresight of fault prediction; through the differential measures matched by the hierarchical regulation strategy, the system reliability and usability are balanced; the closed-loop feedback optimization adaptability overcomes the low utilization rate of historical data, significantly reduces the risk of sudden failure downtime, and improves the stability and efficiency of the power converter operation.

[0092] In one embodiment, the step S102 of performing fault prediction according to the device operation parameters and the electrical characteristic parameters to obtain a prediction value, comprises:

[0093] S1021, feature extraction is performed according to the device operation parameter to obtain device operation failure feature information;

[0094] S1022, a device feature vector is obtained according to the device operation failure feature information;

[0095] S1023, electrical abnormality feature information is obtained according to the electrical feature parameter;

[0096] S1024, an electrical abnormality feature vector is obtained according to the electrical abnormality feature information;

[0097] S1025, device-electrical abnormality correlation information is obtained according to the device feature vector and the electrical abnormality feature vector;

[0098] S1026, a full-cycle historical failure database is obtained, and correlation coefficient threshold intervals corresponding to various failure types are extracted;

[0099] S1027, a failure type-correlation coefficient interval mapping table is established according to the correlation coefficient threshold interval;

[0100] S1028, a failure type interval is obtained according to the device-electrical abnormality correlation information and the failure type-correlation coefficient interval mapping table;

[0101] S1029, a historical failure occurrence frequency is obtained from the full-cycle historical failure database according to the failure type interval;

[0102] S10210, a prediction value is obtained according to the historical failure occurrence frequency.

[0103] As described in steps S1021-S10210, the device operation parameter (such as the equivalent series resistance of the electrolytic capacitor, the junction temperature of the MOSFET, etc.) is preprocessed and multi-dimensional feature extraction is performed to obtain device operation failure feature information. Wavelet transform is used to remove noise and normalize data, and then time domain features (mean, peak, variance), trend features (slope, second derivative) and correlation features (multi-parameter Pearson coefficient) are extracted to form a structured failure feature description. By capturing the dynamic change trend and the correlation between multiple parameters, the limitations of traditional fixed threshold judgment focusing only on instantaneous values are broken through. The technical problems of missing early failure signals (such as parameters not exceeding the threshold but accelerating deterioration) and false alarms of transient fluctuations in traditional methods are solved. For example, when the equivalent series resistance of the electrolytic capacitor slowly increases within the threshold and the correlation with temperature increases, the trend feature and the correlation feature are extracted to identify the aging trend in advance, avoiding missing the early intervention opportunity due to relying only on threshold judgment.

[0104] The device operation failure feature information is converted into a standardized device feature vector, the extracted features are quantified, a plurality of core features are screened through a recursive feature elimination method, and a fixed dimension vector is formed by splicing in a preset order. The problems of strong subjective experience dependence and poor feature comparability in the traditional manual weighting method are solved. For example, after the failure features (such as the mean value of the equivalent series resistance and the temperature peak value) of different capacitors are quantized and spliced into vectors, whether the failure trends are similar can be objectively judged by calculating the vector similarity, avoiding the subjective deviation of manual evaluation.

[0105] Abnormal features are extracted from electrical characteristic parameters (such as output voltage ripple and current harmonic distortion rate), Fourier transform is performed on the collected voltage and current signals to extract frequency domain features (high frequency component proportion), and a cumulative sum control chart algorithm is used to detect sudden change features. The electrical abnormal feature information is formed by analyzing the physical constraint relationship between parameters to extract correlation features. The problems that the traditional instantaneous value exceeding alarm cannot distinguish between normal fluctuations and failure abnormalities and cannot locate the root cause of the abnormality are solved. For example, when the high frequency component proportion of the output voltage ripple abnormally increases, it can be identified as a decrease in the capacitor filtering capability through frequency domain feature extraction, rather than a normal fluctuation caused by load fluctuation, thereby improving the accuracy of abnormal identification.

[0106] The electrical abnormal feature information is converted into an electrical abnormal feature vector, the frequency domain, sudden change and correlation features are quantized, a plurality of core features are screened, and a fixed dimension vector is formed by splicing in the order of “frequency domain → sudden change → correlation”. It is a standardized quantitative representation of electrical abnormalities, providing a unified numerical basis for correlation analysis with device feature vectors. The problems of traditional qualitative description that cannot be compared horizontally and cannot be mathematically modeled are solved. For example, after the ripple abnormalities under different operating conditions are quantized and converted into vectors, whether the abnormal patterns are consistent (such as both caused by capacitor aging) can be objectively judged by calculating the vector distance, thereby improving the feature comparability. The correlation information between the device and the electrical abnormality is obtained by vector fusion and correlation degree calculation, the device and electrical feature vectors are spliced to form a joint vector, the correlation degree is calculated by using a cosine similarity algorithm with a physical correlation weight, and the correlation information is formed by integrating the correlation degree value, the core feature pair and the intensity level.

[0107] The conduction mechanism of “component failure → system abnormality” is disclosed, and the cross-level correlation of failure features is realized. The problems of independent analysis of device and electrical performance and subjective experience-dependent fault location in traditional technologies are solved. For example, by calculating the correlation degree of the capacitor feature vector and the ripple feature vector, it can be determined that the two are strongly correlated, verifying the physical logic of “capacitor aging → ripple increase”, and avoiding misjudgment of the ripple abnormality as other component failures.

[0108] A full-cycle historical fault database is constructed, the correlation coefficient threshold interval of various faults is extracted, historical data containing fault type, characteristic vector, correlation degree and other fields are collected, and after cleaning and classification, the correlation coefficient distribution interval of each type of fault is determined through statistical analysis to establish a quantitative benchmark of historical experience, so that the fault judgment is changed from experience-driven to data-driven. The problem of lack of structured features in traditional historical data and inability to provide quantitative reference for current judgment is solved. For example, by analyzing a large number of electrolytic capacitor aging cases, the correlation coefficient threshold interval is determined to provide a clear historical basis for the matching of the current capacitor correlation degree and avoid inconsistent judgment standards.

[0109] The fault type and correlation coefficient interval are associated through a structured table, a mapping table containing fault coding, interval, core feature pair, priority and other fields is designed, the interval overlap is processed, and dynamic updating is performed to adapt to the aging characteristics of the equipment, the fast matching of fault type and interval is realized, and the uniformity of the judgment standard is ensured. This step solves the problem of traditional dependence on artificial memory or scattered documents, low query efficiency and poor judgment consistency. For example, the record of "electrolytic capacitor aging-correlation coefficient [0.7, 0.9]" in the mapping table can quickly match the current correlation degree to the corresponding fault type, improving the matching efficiency and accuracy.

[0110] The fault type interval is determined by correlation coefficient matching. The process is as follows: extract the current correlation degree, traverse the mapping table to find the interval containing the value, handle the multiple interval overlap according to the priority and feature pair matching degree, and output the unique fault type interval. Through accurate locking of the fault type, a clear direction is provided for subsequent frequency extraction. The problem of traditional subjective judgment leading to the same correlation information being judged as different fault types is solved. For example, the current correlation degree 0.85 is matched to the electrolytic capacitor aging interval through the mapping table, avoiding misjudgment caused by artificial interpretation differences and improving the consistency of fault type positioning.

[0111] The fault occurrence frequency under similar working conditions is extracted from the historical database, the current working condition parameters are collected, the similarity (Euclidean distance) with the historical samples is calculated, the frequency is counted after screening similar samples, and the time decay coefficient is introduced to correct to enhance the timeliness. The occurrence probability of the fault is quantified based on the historical data to avoid the subjectivity of experience estimation. The problem of traditional experience evaluation not considering the difference in working conditions and the large deviation between frequency and actual value is solved. For example, in the historical working conditions similar to the current load rate and temperature, the occurrence frequency of electrolytic capacitor aging is counted to provide a data-based reference for current prediction and improve the accuracy of frequency evaluation.

[0112] The generated quantitative fault prediction value is generated by combining the current fault trend to correct the historical frequency, calculating the confidence interval to reflect the uncertainty, and normalizing the corrected frequency to the [0, 100] interval for output. The quantitative and standardized fault risk is achieved, and the problems of the traditional qualitative evaluation that cannot be compared horizontally and the decision basis is fuzzy are solved. For example, the prediction value of the electrolytic capacitor aging is corrected and standardized, which can be directly compared with the preset threshold to determine whether to start the maintenance in advance, thereby improving the accuracy of the decision.

[0113] In one embodiment, the step S2 of generating intelligent control instructions according to the fault prediction result comprises:

[0114] S201, obtaining fault type information and fault severity level according to the fault prediction result;

[0115] S202, obtaining fault original operation data according to the fault type information, and extracting fault original operation features;

[0116] S203, obtaining device fault state data according to the fault original operation features;

[0117] S204, obtaining device fault trend according to the device fault state data;

[0118] S205, obtaining hierarchical control trigger conditions according to the device fault trend and the fault severity level, and obtaining hierarchical control information according to the hierarchical control trigger conditions;

[0119] S206, generating intelligent control instructions according to the hierarchical control information.

[0120] As described in the above steps S201-S206, the application analyzes the feature vector in the fault prediction result, extracts the fault type information and divides the severity level by combining the full-cycle historical fault database. The core feature vector is extracted from the prediction result, and the K nearest neighbor algorithm is used to match the historical fault template to determine the fault type; based on the prediction value and the fault influence range, the severity level is divided into 1-5 levels (from slight to critical), and the consistency of the results is ensured through cross-validation. The accurate identification of the fault type and the fine division of the severity level are achieved. The problems of low type identification accuracy and rough level division in traditional manual judgment are solved. For example, for the prediction result of “ESR growth and strong correlation with temperature”, it can be matched to the “electrolytic capacitor aging” type, and combined with the influence range on the filter module, it is divided into a medium severity level.

[0121] Collect raw operation data for specific fault types and obtain comprehensive features through multi-domain feature extraction algorithm. According to the fault type, call parameter collection list (such as ESR, temperature, etc. of capacitor aging), collect and clean data in real time through sensor network; extract time domain (mean, peak), frequency domain (harmonic proportion), trend (slope) features, and select core features through principal component analysis dimension reduction.

[0122] Comprehensively capture the original characteristics of faults and avoid missing key information. Solves the problem of insufficient data collection dimension and one-sided feature extraction in traditional data collection. For example, for electrolytic capacitor aging, collect 10-minute fluctuation data of ESR, extract features such as "ESR peak increase" and "high-frequency component proportion of ripple increase", and provide multi-dimensional basis for state assessment.

[0123] Convert raw operation features into quantitative fault state data through multi-feature fusion. Use analytic hierarchy process to assign weights to features (such as ESR trend weight higher than temperature), calculate health index (HI) and derive aging degree, performance decay rate, and fault risk probability, etc. parameters, and verify and evaluate the results through similar device data comparison. Realize the quantitative representation of device health state and avoid the limitations of single feature judgment. Solves the problem of one-sided and lack of quantitative indicators in traditional single threshold evaluation. For example, for the features of capacitor aging, the health index is 45, corresponding to an aging degree of 55% and a performance decay rate of 30%, which objectively reflects the current aging state. Based on historical state data, construct time series and predict fault development trend through multi-factor modeling. Collect state data for the past 3 months to form a sequence, extract instantaneous deterioration rate, acceleration factor and other trend parameters; use LSTM model combined with temperature, load and other factors to predict state in the next 30 days and extract predicted failure time, risk level and other indicators. Accurately predict the fault development speed and predicted failure time to provide time dimension basis for control timing. Solves the problem of traditional linear extrapolation not considering nonlinear factors and large prediction error. For example, by analyzing the health index sequence of capacitor aging and considering the influence of recent temperature rise, it can be predicted that it will reach the serious failure threshold in 45 days, with a risk level of medium.

[0124] According to the device failure trend and failure severity level, a hierarchical regulation trigger condition is obtained, and hierarchical regulation information is obtained according to the hierarchical regulation trigger condition, and dynamic trigger conditions and hierarchical regulation information are generated in combination with the failure trend and the severity level. According to the predicted failure time, the basic threshold (such as the rapid deterioration reduction threshold) is corrected, and the early warning, intervention, and emergency three-level thresholds are divided; a mapping rule of the threshold and the regulation level (L1-L4) is established, and the regulation information including the execution subject and the steps is generated by calling the measure library. The dynamic adaptation of the trigger condition and the differentiated design of the regulation measures are realized, and the balance between failure prevention and control and operation continuity is achieved. The problems of "whether to regulate" or "excessive regulation" caused by traditional fixed thresholds and single measures are solved. For example, for the capacitor aging with medium severity level and medium deterioration trend, the intervention threshold is set to 70% of the aging degree, and the mild intervention information (such as reducing the load rate) is obtained. The hierarchical regulation information is converted into standardized instructions executable by the device. The operation object, target parameter, and other elements in the regulation information are analyzed, the instruction template (such as the Modbus protocol) corresponding to the device is matched, the instruction including the check bit is coded, and the instruction is output and the execution feedback is received through the industrial bus to ensure that the instruction is effectively implemented. The automation and cross-device compatible execution of the regulation measures are realized, and the response efficiency and accuracy are improved. The problems of low efficiency and poor compatibility of traditional manual instruction writing are solved. For example, for the regulation information of "reducing the load rate of the capacitor module", the instruction code conforming to the device protocol is generated, and after being sent through the bus, the feedback of successful execution can be received in real time.

[0125] In one embodiment, the step S205 of obtaining the hierarchical regulation trigger condition according to the device failure trend and the failure severity level, and obtaining the hierarchical regulation information according to the hierarchical regulation trigger condition comprises:

[0126] S2051, obtaining a hierarchical regulation trigger threshold according to the device failure trend and the failure severity level;

[0127] S2052, establishing a threshold-regulation level mapping rule according to the hierarchical regulation trigger threshold;

[0128] S2053, obtaining device operation data, and comparing the device operation data with the hierarchical regulation trigger threshold to obtain a comparison result;

[0129] S2054, obtaining a corresponding regulation level according to the comparison result;

[0130] S2055, obtaining a preset hierarchical regulation measure library according to the regulation level;

[0131] S2056, obtaining a regulation item entry according to the regulation measure library, and generating hierarchical regulation information according to the regulation item entry.

[0132] As described in steps S2051-S2056 above, the application calculates dynamic hierarchical regulation trigger thresholds by quantifying device failure trend and failure severity level. The failure trend is quantified as a trend index, and the failure severity level is quantified as a severity coefficient; after determining the base threshold based on historical failure data, the dynamic threshold is obtained by combining the trend index and the severity coefficient, and is divided into multiple levels of thresholds such as early warning, mild intervention, severe intervention, etc. Through adaptive adjustment of the trigger threshold, the regulation timing is matched with the failure development speed and impact degree. The problem of traditional fixed threshold being unable to adapt to dynamic changes in failure, leading to excessive regulation or insufficient regulation, is solved. For example, for capacitor aging failure, when it deteriorates rapidly and has a serious impact, the threshold will be appropriately lowered to trigger regulation in advance; when it deteriorates slowly and has a slight impact, the threshold will be close to the base value to reduce unnecessary intervention.

[0133] By defining regulation levels and mapping conditions, the association rules between threshold intervals and regulation levels are established. Four levels of regulation levels, early warning, mild intervention, severe intervention, and emergency shutdown, are divided, and the targets of each level are clear. The basic correspondence between hierarchical thresholds and regulation levels is established, and auxiliary conditions such as failure propagation risk and device operating mode are introduced to modify the mapping results. The rules are stored in a structured decision table and dynamically updated. The accurate matching of thresholds and regulation levels is achieved, ensuring that the regulation intensity is adapted to the failure degree. The problem of traditional "single threshold-single measure" strategy lacking flexibility, leading to mismatch between measures and failures, is solved. For example, when the parameter is in the second threshold interval, the basic mapping is mild intervention, but if there is a high risk of propagation (such as causing adjacent device failure), it is modified to severe intervention to avoid failure expansion. Real-time device operating data is collected and automatically compared with hierarchical regulation trigger thresholds. Device parameters and environmental parameters are collected in real time by a sensor network and transmitted to the decision unit after preprocessing; the hierarchical thresholds are called, and the data is determined to be in the threshold interval by parameter-by-parameter comparison and comprehensive scoring, and the comparison results including single parameter interval, comprehensive interval, and over-standard duration are output. The real-time and objectivity of data comparison are achieved, providing accurate basis for regulation level judgment. The problem of traditional manual comparison being real-time and subjective is solved. For example, after real-time monitoring of the equivalent series resistance value of the capacitor, it is automatically compared with the third threshold, and it is quickly determined that it is in the mild intervention interval, and the over-standard duration has reached 5 minutes, providing quantitative data for subsequent level judgment.

[0134] Based on the comparison result and the mapping rule, the regulation level is automatically determined. According to the comprehensive interval matching of the comparison result, the basic regulation level is matched, combined with the single parameter interval, the duration of over-standard and the real-time state of the equipment to adapt to the auxiliary conditions; for the conflicting adjustment requirements, the priority rule of safety first and operation continuity second is processed, and the final regulation level and the judgment basis are output. It is ensured that the regulation level is consistent with the actual severity of the fault, and the subjective deviation of manual judgment is avoided. The problem of inconsistent levels caused by traditional experience-dependent judgment is solved. For example, if the comparison result shows that the parameter is in the heavy intervention interval, and there are multiple parameter over-standard (high diffusion risk), the final determination is emergency shutdown level, which ensures timely prevention and control of safety risk.

[0135] By constructing a structured measure library, the association between regulation level and specific measures is realized. A multi-level measure library is designed according to the regulation level and fault type, and measure items containing execution subject, steps and expected effect are stored; based on historical experience, simulation verification and expert review, the items are generated and associated to the corresponding level according to the regulation intensity (such as notification measures for early warning and safety protection measures for emergency shutdown), and dynamically updated to include new measures. The quick calling and adaptation of measures are realized, and the problem of scattered and low-efficiency calling of traditional measures is solved. For example, when the regulation level is light intervention, the adaptive measures such as "turn on auxiliary cooling fan" and "reduce module load rate" can be quickly obtained from the measure library, ensuring timely and effective intervention.

[0136] Convert the regulation item into executable hierarchical regulation information. According to the regulation level and fault type, the optimal measure item is selected; for the automatic execution item, the device recognizable instruction code is generated, and for the manual execution item, the step-by-step operation guide is generated; adapt to the output of multiple platforms such as automatic system and operation terminal, and establish feedback mechanism to track the execution result. It is ensured that the regulation measures are accurately landed, and the execution efficiency and consistency are improved. The problem of fuzzy description and large execution deviation of traditional measures is solved. For example, for the load adjustment item of light intervention, the load rate adjustment instruction conforming to the device protocol is automatically generated, which is pushed to the control system for execution, and the operation progress is synchronized to the operation terminal, ensuring that the measures take effect as expected.

[0137] In one embodiment, the step S2056 of obtaining a regulation item from the regulation measure library and generating hierarchical regulation information according to the regulation item includes:

[0138] S20561, generating a search tag according to the fault type information and the fault severity level;

[0139] S20562, calling the matched regulation item from the regulation measure library according to the search tag;

[0140] S20563, obtaining execution object information, operation instruction information, and constraint condition information according to the retrieved regulation item entry;

[0141] S20564, generating hierarchical regulation information according to the execution object information, operation instruction information, and constraint condition information.

[0142] As described in steps S20561-S20564, the application generates a structured search label containing fault specificity and severity by analyzing fault type information and fault severity level. The associated components (such as "C phase #2 electrolytic capacitor") and core fault characteristics (such as "ESR increase") are extracted from the fault type, and the severity level is converted into "L1-L5" coding and risk description. Through semantic association algorithm, the above information is fused to form a three-part label of "component code-core feature code-level code", and the label compliance is verified. The search label accurately covers the fault characteristics, providing a clear basis for measure library search. It solves the problem of traditional single keyword search information ambiguity, which easily leads to measure mismatch. For example, for "C phase #2 electrolytic capacitor aging and severity level 3", the label "EC-C2-ESR-INC-L3" is generated, which can accurately point to the adaptive measures for the fault.

[0143] Based on the search label, the matching regulation item entry is accurately retrieved from the regulation measure library. The "component-feature-level" three-dimensional index structure of the measure library is used to split the search label and assign weights (component 0.4, feature 0.4, level 0.2). The matching degree of the label and the associated label in the library entry is calculated by weighted cosine similarity, and the candidate entries are sorted by similarity. After effectiveness (success rate ≥80%) and scene adaptability screening, the optimal entry is determined. It improves the accuracy and efficiency of measure retrieval, and ensures that the measures match the fault characteristics. It solves the problem of low efficiency and poor matching accuracy of traditional manual search. For example, for the "MOS-F3-Tj-OV-L4" label, the adaptive entry of "switching to standby MOSFET module" can be quickly retrieved, avoiding insufficient measure strength.

[0144] Extract the execution object, operation instruction and constraint condition core elements from the retrieved regulation item entry. Identify the operation subject (such as "automatic control system") and the action object (such as "F3 phase MOSFET module"); quantify the operation instruction (such as "switch module + limit current" and target parameter) and clarify the execution steps; define time constraints (such as "execute within 3 seconds"), safety constraints (such as "voltage fluctuation ≤ ± 5%"), etc.; check the completeness and consistency of the elements, and complete the missing ones from the default template. Clarify the execution details of the regulation measures to avoid execution deviation caused by ambiguous information. The problems of traditional entry description being general and key elements being missing are solved. For example, from the "switch standby MOSFET module" entry, the execution object "automatic control system and F3 phase module", the operation instruction "first disconnect the main module and then put in the standby module", and the constraint condition "switching time ≤ 3 seconds" can be extracted.

[0145] Integrate the extracted elements to generate structured and standardized hierarchical regulation information. Use the "header-body-tail" architecture, the header contains regulation ID and fault label, the body integrates elements according to "execution object-operation instruction-constraint condition" and standardizes coding (such as role code "AC-01" represents automatic control system), and the tail adds check code and feedback requirements; adapt to automatic system (JSON format), operation and maintenance terminal (graphical format) and other platforms, listen to execution feedback after distribution. Ensure that the regulation information can be accurately parsed and executed to realize the consistency of measure landing. The problems of traditional information being scattered and having large interpretation differences are solved. For example, for the elements of capacitor load adjustment, the generated JSON format information can be directly parsed by the automatic control system to ensure accurate execution of the instruction "reduce to 80% load rate within 10 minutes".

[0146] In one embodiment, the step S20561 of generating a retrieval label according to the fault type information and the fault severity level includes:

[0147] S205611, extracting fault associated component information and core fault characteristics according to the fault type information;

[0148] S205612, generating a fault type feature label according to the fault associated component and the core fault characteristics;

[0149] S205613, obtaining a fault level identification label according to the fault severity level;

[0150] S205614, performing semantic association according to the fault level identification label and the fault type feature label to obtain label associated content;

[0151] S205615, generating a retrieval label according to the label associated content.

[0152] As described in the above steps S205611-S205615, the application extracts the fault associated components and core fault features by analyzing the fault type information. The power field segmentation tool is used to analyze the fault type text, and the BiLSTM entity recognition model is used to locate the associated components (including type, location and unique code). Combined with the fault mechanism library, the core features (the top 3 high correlation features are retained) strongly related to the fault are screened through mutual information entropy, and the matching of components and features is verified. The accurate positioning of associated components and the comprehensive extraction of core features are realized. The problems of ambiguous component positioning and one-sided features in traditional manual extraction are solved. For example, from "B phase #1 MOSFET junction temperature too high", the associated component "B phase #1 MOSFET" and the core features "junction temperature too high" and "turn-on voltage drop increase" can be accurately extracted, avoiding misjudgment of the features as irrelevant parameters.

[0153] The fault associated components and core fault features are converted into standardized fault type feature labels. According to the preset coding rules, the component type and location are converted into the "type abbreviation-location identification" format (such as "electrolytic capacitor C phase #2" → "EC-C2"); the core features are converted into the "parameter abbreviation-exception mode abbreviation" format (such as "ESR increase" → "ESR-INC"); according to the "component code-feature code" structure, the fixed format label is spliced and the compliance is verified. The standardized coding of fault features is realized, and the retrieval deviation caused by description differences is eliminated. The problems of chaotic natural language label format and information redundancy in traditional methods are solved. For example, for "C phase #2 electrolytic capacitor ESR increases, ripple high frequency component increases", the label "EC-C2-ESR-INC-RIPPLE-HF" is generated, ensuring the consistency of labels for similar faults in different scenarios.

[0154] The fault severity level is converted into a structured fault level identification label. The 1-5 level severity level output by S201 is analyzed to generate an "L+ level value" code (such as level 3 → "L3"); the standardized risk description is matched from the level definition library (such as level 3 → "system local influence"); the label is combined in the format of "level code|risk description", and the matching of the code and the description is verified. The urgency and impact range of the fault are clear, which provides a basis for the matching of the measure intensity. The problem of semantic ambiguity in traditional text description level is solved. For example, the 4-level severity level corresponds to the label "L4|system stable influence", which clearly conveys the urgency of "threatening system stable operation", avoiding the interpretation difference of words such as "serious".

[0155] The logical association of the fault type feature label and the level identification label is realized through semantic analysis. The two types of labels are converted into semantic vectors, and the cosine similarity is calculated (≥ 0.5 is considered as matching); the “fault type-level association rule base” is called for secondary verification (such as prohibiting the association of “electrolytic capacitor aging” with L5); for the combinations that pass the verification, the associated content in the format of “type feature label | level identification label” is generated, with the similarity confidence added. The logical consistency of the label combination is ensured, and contradictory associations (such as associating a minor fault with a high level) are avoided. The problem of contradictory combinations easily occurring in traditional label simple splicing is solved. For example,

[0156] “MOS-B1-Tj-OV-VDS-INC” (MOSFET overheating) and “L4 | system stability impact” have a semantic similarity of 0.72 and meet the rules, and a reasonable associated content is generated, excluding the contradictory combination of overheating and low level.

[0157] The label association content is optimized to a concise retrieval label. The associated content is parsed, the redundant risk description is removed, and the fault type feature label and the level code are retained; the labels are spliced in the format of “type feature label-level code” (such as “EC-C2-ESR-INC-RIPPLE-HF-L3”); the coding compliance and the adaptability to the measure library index (such as the prefix matching component index and the suffix matching level index) are verified. The standardized label for efficient and adaptive retrieval is generated, and the measure library matching efficiency is improved. This step solves the problem of low retrieval efficiency caused by redundant and disordered structure of traditional labels. For example, the retrieval label extracted from the associated content can be quickly located by the measure library “EC→ESR-INC→L3” index path, ensuring accurate retrieval of adaptive measures.

[0158] As shown in Figure 2 The present application also provides a power converter intelligent control system with fault prediction, comprising:

[0159] A real-time operation data acquisition module 1 is configured to acquire real-time operation data of the power converter and perform fault prediction based on the real-time operation data to generate a fault prediction result, wherein the fault prediction result includes normal and abnormal;

[0160] A regulation and control instruction generation module 2 is configured to determine that the fault result is abnormal, generate an intelligent regulation and control instruction based on the fault prediction result, and regulate and control the power converter based on the intelligent regulation and control instruction;

[0161] An adjusted data acquisition module 3 is configured to acquire real-time operation data of the adjusted power converter and return to the step of performing fault prediction based on the real-time operation data.

[0162] In one embodiment, the regulation and control instruction generation module 2 comprises:

[0163] The real-time operation data acquisition unit is configured to acquire real-time operation data of the power converter, wherein the real-time operation data comprises device operation parameters and electrical characteristic parameters.

[0164] The prediction value unit is configured to perform fault prediction according to the device operation parameters and the electrical characteristic parameters to obtain a prediction value.

[0165] The judgment acquisition unit is configured to judge whether the prediction value exceeds a preset value.

[0166] If the prediction value exceeds the preset value, the fault prediction result is abnormal.

[0167] If the prediction value does not exceed the preset value, the fault prediction result is normal.

[0168] In one embodiment, the regulation instruction generation module 2 further comprises:

[0169] The device operation fault characteristic information acquisition unit is configured to perform feature extraction according to the device operation parameters to obtain device operation fault characteristic information.

[0170] The device feature vector acquisition unit is configured to acquire a device feature vector according to the device operation fault characteristic information.

[0171] The electrical abnormality characteristic information acquisition unit is configured to acquire electrical abnormality characteristic information according to the electrical characteristic parameters.

[0172] The electrical abnormality feature vector acquisition unit is configured to acquire an electrical abnormality feature vector according to the electrical abnormality characteristic information.

[0173] The device-electrical abnormality correlation information acquisition unit is configured to acquire device-electrical abnormality correlation information according to the device feature vector and the electrical abnormality feature vector.

[0174] The correlation coefficient threshold interval extraction unit is configured to acquire a full-cycle historical fault database and extract correlation coefficient threshold intervals corresponding to various fault types.

[0175] The fault type-correlation coefficient interval mapping table establishment unit is configured to establish a fault type-correlation coefficient interval mapping table according to the correlation coefficient threshold intervals.

[0176] The fault type interval acquisition unit is configured to acquire a fault type interval according to the device-electrical abnormality correlation information and the fault type-correlation coefficient interval mapping table.

[0177] The historical fault occurrence frequency acquisition unit is configured to acquire a historical fault occurrence frequency from the full-cycle historical fault database according to the fault type interval.

[0178] The prediction value acquisition unit is configured to acquire a prediction value according to the historical fault occurrence frequency.

[0179] Those of ordinary skill in the art understand that all or part of the processes in the above-mentioned embodiments can be implemented by a computer program instructing relevant hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, storage, value library or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0180] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, device, article or method including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such a process, device, article or method. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0181] The above description is only the preferred embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent results or equivalent process transformations, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for intelligent control of a power converter with failure prediction, characterized by, The method comprises the following steps: acquiring real-time operation data of a power converter, and performing fault prediction according to the real-time operation data to generate a fault prediction result, wherein the fault prediction result comprises normal and abnormal; if the fault result is abnormal, generating an intelligent control instruction according to the fault prediction result, and controlling the power converter according to the intelligent control instruction; acquiring real-time operation data of the adjusted power converter, and returning to the step of performing fault prediction according to the real-time operation data.

2. The power converter intelligent control method with failure prediction according to claim 1, characterized in that, The step of acquiring real-time operation data of a power converter, and performing fault prediction according to the real-time operation data to generate a fault prediction result, comprises the following steps: acquiring real-time operation data of the power converter, wherein the real-time operation data comprises device operation parameters and electrical characteristic parameters; performing fault prediction according to the device operation parameters and the electrical characteristic parameters to obtain a prediction value; judging whether the prediction value exceeds a preset value; if the prediction value exceeds the preset value, the fault prediction result is abnormal; if the prediction value does not exceed the preset value, the fault prediction result is normal.

3. The method of claim 2, wherein the method further comprises: The step of performing fault prediction according to the device operation parameters and the electrical characteristic parameters to obtain a prediction value comprises the following steps: extracting features according to the device operation parameters to obtain device operation fault feature information; acquiring a device feature vector according to the device operation fault feature information; acquiring electrical abnormal feature information according to the electrical characteristic parameters; acquiring an electrical abnormal feature vector according to the electrical abnormal feature information; acquiring device-electrical abnormality correlation information according to the device feature vector and the electrical abnormal feature vector; acquiring a full-cycle historical fault database, and extracting correlation coefficient threshold intervals corresponding to various fault types; establishing a fault type-correlation coefficient interval mapping table according to the correlation coefficient threshold intervals; acquiring a fault type interval according to the device-electrical abnormality correlation information and the fault type-correlation coefficient interval mapping table; acquiring a historical fault occurrence frequency from the full-cycle historical fault database according to the fault type interval; acquiring the prediction value according to the historical fault occurrence frequency.

4. The power converter intelligent control method with failure prediction according to claim 1, characterized in that, The step of generating an intelligent control instruction according to the fault prediction result comprises the following steps: acquiring fault type information and a fault severity level according to the fault prediction result; acquiring fault original operation data according to the fault type information, and extracting fault original operation features; acquiring device fault state data according to the fault original operation features; acquiring a device fault trend according to the device fault state data; acquiring a hierarchical control trigger condition according to the device fault trend and the fault severity level, and acquiring hierarchical control information according to the hierarchical control trigger condition; generating an intelligent control instruction according to the hierarchical control information.

5. The method of claim 4, wherein the method further comprises: The step of acquiring a hierarchical control trigger condition according to the device fault trend and the fault severity level, and acquiring hierarchical control information according to the hierarchical control trigger condition, comprises the following steps: acquiring a hierarchical control trigger threshold according to the device fault trend and the fault severity level; establishing a threshold-control level mapping rule according to the hierarchical control trigger threshold; acquiring device operation data, and comparing the device operation data with the hierarchical control trigger threshold to obtain a comparison result; acquiring a corresponding control level according to the comparison result; According to the regulation level, a preset hierarchical regulation measure library is obtained; According to the regulation measure library, a regulation item entry is obtained, and hierarchical regulation information is generated according to the regulation item entry.

6. The method of claim 5, wherein the method further comprises: According to the regulation measure library, a regulation item entry is obtained, and hierarchical regulation information is generated according to the regulation item entry. According to the fault type information and the fault severity level, a search tag is generated; According to the search tag, a matched regulation item entry is retrieved from the regulation measure library; According to the retrieved regulation item entry, execution object information, operation instruction information, and constraint condition information are obtained; According to the execution object information, operation instruction information, and constraint condition information, hierarchical regulation information is generated.

7. The method of intelligent control of a power converter with failure prediction according to claim 6, characterized in that, The step of generating a search tag according to the fault type information and the fault severity level includes: According to the fault type information, fault associated component information and core fault characteristics are extracted; According to the fault associated component and the core fault characteristics, a fault type feature tag is generated; According to the fault severity level, a fault level identification tag is obtained; According to the fault level identification tag and the fault type feature tag, semantic association is performed to obtain tag association content; According to the tag association content, a search tag is generated.

8. A power converter intelligent control system with failure prediction, characterized by, It includes: A real-time running data acquisition module is configured to acquire real-time running data of the power converter, and perform fault prediction according to the real-time running data to generate a fault prediction result, wherein the fault prediction result includes normal and abnormal; A regulation instruction generation module is configured to determine that the fault result is abnormal, and generate an intelligent regulation instruction according to the fault prediction result, and regulate the power converter according to the intelligent regulation instruction; An adjusted data acquisition module is configured to acquire real-time running data of the adjusted power converter, and return to the step of performing fault prediction according to the real-time running data.

9. The power converter intelligent control system with fault prediction of claim 8, wherein, The regulation instruction generation module includes: A real-time running data acquisition unit is configured to acquire real-time running data of the power converter, wherein the real-time running data includes device running parameters and electrical characteristic parameters; A prediction value unit is configured to perform fault prediction according to the device running parameters and the electrical characteristic parameters to obtain a prediction value; A judgment acquisition unit is configured to determine whether the prediction value exceeds a preset value; If the prediction value exceeds the preset value, the fault prediction result is abnormal; If the prediction value does not exceed the preset value, the fault prediction result is normal.

10. The power converter intelligent control system with fault prediction of claim 9, wherein, The regulation instruction generation module further includes: A device running fault feature information acquisition unit is configured to extract device running fault feature information according to the device running parameters; A device feature vector acquisition unit is configured to acquire a device feature vector according to the device running fault feature information; An electrical abnormal feature information acquisition unit is configured to acquire electrical abnormal feature information according to the electrical characteristic parameters; An electrical abnormal feature vector acquisition unit is configured to acquire an electrical abnormal feature vector according to the electrical abnormal feature information; A device-electrical abnormality association information acquisition unit is configured to acquire device-electrical abnormality association information according to the device feature vector and the electrical abnormal feature vector; A correlation coefficient threshold interval extraction unit is configured to acquire a full-cycle historical fault database, and extract correlation coefficient threshold intervals corresponding to various fault types; The fault type-correlation coefficient interval mapping table establishing unit is configured to establish a fault type-correlation coefficient interval mapping table according to the correlation coefficient threshold interval; The fault type interval obtaining unit is configured to obtain a fault type interval according to the device-electricity anomaly correlation information and the fault type-correlation coefficient interval mapping table; The historical fault occurrence frequency obtaining unit is configured to obtain a historical fault occurrence frequency from a full-cycle historical fault database according to the fault type interval; The prediction value obtaining unit is configured to obtain a prediction value according to the historical fault occurrence frequency.

Images