A wind turbine generator system fault early warning method and system based on master control fault code

Abstract

The application relates to the technical field of intelligent early warning of wind power generation, and discloses a wind turbine generator system fault early warning method and system based on master control fault codes. The fault early warning method is based on a plurality of fault codes triggered by master control, uses a logic rule and a decision tree algorithm to identify faults that need to trigger early warning, and performs real-time fault early warning on the wind turbine generator system triggering the fault codes. The application solves the problems of low intelligent degree of wind turbine generator system early warning and insufficient diversification of analysis data in the prior art.

Description

Technical Field

[0001] This invention relates to the field of intelligent early warning technology for wind power generation, specifically a method and system for early warning of wind turbine generator faults based on master control fault codes. Background Technology

[0002] The early warning system for wind turbine generator sets is a powerful supplement to ensure the healthy and stable operation of the units. It can detect early potential faults and early failures of the units in advance, and proactively intervene and maintain them, ensuring that the units operate in a more ideal state.

[0003] Currently, the main methods for early warning of wind turbine generators are as follows:

[0004] I. Utilizing a Condition Monitoring System (CMS) for the generator unit, various sensors deployed on key components such as the drive train and blades are used to monitor the unit in real time. This method requires strong professional knowledge and corresponding technical expertise, involving knowledge from numerous disciplines such as mechanics and machinery, and places high demands on the personnel involved.

[0005] Second, SCADA data analysis is employed, utilizing artificial intelligence and big data methods to model and analyze relevant components. However, this approach requires a large amount of training data upfront, and the unit's operation is subject to numerous unpredictable factors, which can affect the model's results. Furthermore, it necessitates that personnel possess professional knowledge of algorithm modeling and big data theory, making development quite challenging.

[0006] Third, audio and video data are analyzed using big data technology, artificial intelligence technology, and deep learning technology to analyze and model the audio and video data of large components such as transmission chains and blades. This approach is affected by the size of sensors and the nacelle space, making the selection of placement more complex. On the other hand, it requires more professional knowledge of audio and video data analysis and processing, as well as stronger deep learning modeling capabilities. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, this invention provides a method and system for early warning of wind turbine generator faults based on master control fault codes, which solves the problems of low intelligence level of early warning of wind turbine generators and insufficient diversity of analysis data in existing technologies.

[0008] The technical solution adopted by the present invention to solve the above problems is:

[0009] A fault early warning method for wind turbine generator sets based on master control fault codes is proposed. Based on several fault codes triggered by the master control, logical rules and decision tree algorithms are used to identify faults that need to trigger early warning, and real-time fault early warning is provided for wind turbine generator sets that trigger fault codes.

[0010] As a preferred technical solution, the steps include:

[0011] S1, Data Acquisition: Based on the main control fault codes reported by the main control logic of the wind turbine, the main control fault information of the wind turbine is acquired in real time;

[0012] S2, Feature Extraction: Analyze the main control fault information using logical rules, extract the feature values ​​of the main control fault information, and construct a feature value set from the feature values;

[0013] S3, Decision Tree Analysis: Perform decision tree analysis on the feature set to identify the corresponding root causes of failures and faulty components;

[0014] S4, Information Push: Based on the root cause of the fault and the faulty component, identify the fault that needs to trigger an early warning, and push and display the information that triggers the early warning.

[0015] As a preferred technical solution, the following steps are also included:

[0016] S5, Work Order Management: Includes the pushed work order information into the maintenance work order pool for maintenance work order scheduling management.

[0017] S6, Operations and Maintenance Management: Investigate and provide feedback on operations and maintenance work orders.

[0018] As a preferred technical solution, in step S2, the IBLE decision tree algorithm is used to analyze the main control fault information, and the feature values ​​of the main control fault information are extracted according to the expert experience logic rules.

[0019] As a preferred technical solution, in step S3, for the chain fault code triggering situation analyzed by the algorithm, the corresponding feature values ​​are extracted to form a feature set u. The feature set u is analyzed by the IBLE decision tree algorithm to establish the correspondence between the feature set and the root cause of the fault and the faulty component, thereby obtaining the associated high-risk fault results, and storing the analysis results in the database.

[0020] As a preferred technical solution, step S3 includes the following steps:

[0021] S31, Initial settings: Extract the features of the main control fault codes, combine the extracted features into a feature set A, convert the feature values ​​into binary values, and each feature value is 0 or 1;

[0022] S32, Rule-building algorithm: For the current training set PE∪NE, construct the main rules using the rule-building algorithm;

[0023] Where PE represents a positive example in the test dataset, and NE represents a negative example in the test dataset;

[0024] S32 specifically includes the following steps:

[0025] S321, For each feature Am: First, transform Am into sub-features Am1, Am2, ..., Amj. Calculate the channel capacity of each sub-feature of Am, and select the sub-feature with the largest channel capacity to represent Am. The weight of Am... Then, use the maximum likelihood decoding criterion to find the values ​​of the sub-features representing Am, and replace them with the corresponding logical symbols;

[0026] S322, take the first m features with larger channel capacity to form a rule;

[0027] S323, calculate the thresholds Sp and Sn by weighting PE and NE based on the selected features and performing statistical analysis.

[0028] Where Sp represents the threshold for classifying an example as positive, and Sn represents the threshold for classifying an example as negative;

[0029] S33, use the main rule to test PE to obtain subsets PEY, PEN, and PEM, and use the main rule to test NE to obtain subsets NEY, NEN, and NEM; where PEY represents a positive example that is judged as a positive example, PEN represents a positive example that is judged as a negative example, PEM represents an example that cannot be judged as a positive example, NEY represents a negative example that is judged as a positive example, NEN represents a negative example that is judged as a negative example, and NEM represents an example that cannot be judged as a negative example;

[0030] S34, Place the main rule into node P;

[0031] S35, Let T be the space for storing the decision rule tree, and set the decision rule tree T to be empty; allocate a new node P, T:=P; where := means assignment, and T:=P means assigning the value of P to T;

[0032] If (|PEY|≠0)∪(|NEY|≠0), then PE:=PEY, NE:=NEY; allocate a new node W1, and the left pointer of P points to W1; then, construct the left division rule for the current training set PE∪NE using the rule building algorithm in step S32, and put the left division rule into node W1;

[0033] If (∧PEN∧≠0)∪(∧NEN∧≠0), then PE:=PEN, NE:=NEN; allocate a new node W2, and the right pointer of P points to W2; then, construct the right division rule for the current training set PE∪NE using the rule building algorithm in step S32, and put the right division rule into node W2;

[0034] If (|PEM|≠0)∪(|NEM|≠0), then PE:=PEM, NE:=NEM, allocate a new node W3, the middle pointer of P points to W3, P:=W3; go to step S32;

[0035] After all test samples have been allocated, proceed to step S39;

[0036] S39, End.

[0037] As a preferred technical solution, step S40 includes the following steps:

[0038] S401, Set the root node as the current node;

[0039] S402, After obtaining the decision rules, how to classify faults E with unknown root causes, and give a specific algorithm;

[0040] If the result is u1, and the left pointer of the current node is not null, set the node pointed to by the left pointer as the current node and proceed to step S402; otherwise, proceed to step S403.

[0041] If the result is u2, and the right pointer of the current node is not null, then set the node pointed to by the right pointer as the current node and proceed to step S402; otherwise, proceed to step S403.

[0042] If the decision cannot be made, and the pointer of the current node is not null, then set the node pointed to by the pointer of the current node as the current node and proceed to step S402; otherwise, proceed to step S403.

[0043] S403, output the conclusion that it is a positive example, a negative example, or cannot be determined, and proceed to step S404;

[0044] Where u1 represents a positive example of the output result and u2 represents a negative example of the output result;

[0045] S404, End.

[0046] As a preferred technical solution, in step S1, the fault record library information of the main control PLC is used to store the fault record library information in a relational or non-relational database, and the corresponding main control fault code record is retrieved at a custom time.

[0047] A wind turbine generator fault early warning method based on master control fault codes, comprising the following modules connected in sequence:

[0048] Data acquisition module: used to acquire wind turbine main control fault information in real time based on the main control fault codes reported by the wind turbine main control logic;

[0049] Feature extraction module: Used to analyze the main control fault information using logical rules, extract feature values ​​of the main control fault information, and construct a feature value set.

[0050] Decision tree analysis module: used to perform decision tree analysis on the feature set to identify the corresponding root causes and faulty components;

[0051] Information push module: Used to identify faults that require early warning based on the root cause and faulty components, and to push and display the information that triggers the early warning.

[0052] Work order management module: Used to include the pushed work order information into the operation and maintenance work order pool for operation and maintenance work order scheduling management.

[0053] Operations and maintenance management module: used to investigate and provide feedback on operations and maintenance work orders.

[0054] Compared with the prior art, the present invention has the following advantages:

[0055] (1) This invention monitors the main control fault codes of wind turbine generators in real time and uses relevant algorithms to complete chain fault code identification according to certain rules. By identifying the chain fault codes, it is determined whether the current main control triggered fault is a high-risk fault; for high-risk faults, the expert database is queried in real time, and the corresponding components of the wind turbine generator are associated with the expert database data, thereby providing early warning of component faults. At the same time, closed-loop operation and maintenance are carried out in the closed-loop system, and early warning work orders are triggered through SMS, APP, monitoring interface, etc.; and through the operation and maintenance execution process, the expert database is updated in real time, and the information in the expert database is continuously optimized and corrected to ensure that the information in the expert database is always the latest and most accurate.

[0056] (2) This invention can make full use of the main control fault triggering information and combine expert experience rules to timely warn and display potential fault hazards of the unit, providing a certain reference and basis for subsequent active operation and maintenance of the unit, thereby avoiding wind turbine failure shutdown, improving the operating efficiency of wind turbine units, and further helping to realize unmanned wind farms. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the steps of a wind turbine generator fault early warning method based on master control fault codes according to the present invention.

[0058] Figure 2 This is a schematic diagram of the fault tree described in this invention;

[0059] Figure 3 This is a schematic diagram of the expert logic triggering of the present invention;

[0060] Figure 4 This is a simplified rule diagram for Example 2;

[0061] Figure 5 This is a schematic diagram of the complex rules in Example 2;

[0062] Figure 6 This is a schematic diagram of the IBLE decision tree algorithm in Example 2. Detailed Implementation

[0063] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0064] Example 1

[0065] like Figures 1 to 6 As shown, the purpose of this invention is to analyze and model the fault codes triggered by the main control of wind turbine generators, and to propose a fault early warning method and system for wind turbine generators based on the main control fault codes, thereby supplementing the existing early warning methods.

[0066] This invention monitors wind turbine main control system fault codes in real time and uses relevant algorithms to identify chained fault codes according to certain rules. By identifying these chained fault codes, it determines whether the current main control system fault is a high-risk fault. For high-risk faults, it queries an expert database in real time, linking the expert database data to the corresponding components of the wind turbine, thereby issuing early warnings for component failures. Simultaneously, closed-loop maintenance is implemented within the closed-loop system, triggering early warning work orders via SMS, APP, monitoring interface, etc. Throughout the maintenance process, the expert database is updated in real time, continuously optimizing and correcting its information to ensure it remains up-to-date and accurate.

[0067] This invention identifies chained fault codes based on historical fault data according to certain rules. The function uses an IBLE decision tree algorithm to analyze the data, thereby obtaining the occurrence times of the fault codes specified by the first and last rules. Statistical analysis is performed on the fault codes throughout the chain to derive associated high-risk fault results. This function primarily relies on fault codes reported by the wind turbine's main control system, which are acquired in real-time using external software.

[0068] The function described in this invention uses the IBLE decision tree algorithm to analyze data, thereby obtaining the occurrence times of the fault codes specified by the first and last rules. Statistical analysis is performed on the fault codes throughout the entire chain to derive associated high-risk fault results. The analysis results are stored in a database. This function pushes the analysis results to the monitoring interface and displays the warning results through audible and visual alarms. Work order management automatically associates the warning results with the digital operation and maintenance system, which generates work orders from the warning results and pushes these work orders to the work order pool for queuing. This function proactively performs maintenance on priority work orders, recording the maintenance process and results in the digital operation and maintenance system.

[0069] The steps used in this invention are as follows:

[0070] Data acquisition is primarily based on the main control fault codes reported by the wind turbine's main control logic. The role of this data acquisition is to obtain real-time fault information from the wind turbine's main control system.

[0071] Feature extraction involves analyzing the main control fault information using logical rules, extracting feature values ​​from the fault information, and constructing a feature value set. (The IBLE decision tree algorithm is used to analyze the data, and the triggered fault information is extracted as feature values ​​according to expert experience and logical rules.)

[0072] Decision tree analysis involves performing decision tree analysis on a set of feature values ​​to identify corresponding root causes and faulty components. (After analyzing the feature values ​​using the IBLE decision tree algorithm, the results are output, thus deriving the associated high-risk fault results. The analysis results are then stored in a database.)

[0073] Information push notifications will send triggered information from the relevant database to the monitoring system interface, providing alerts and warnings via voice or audio-visual displays.

[0074] Work order management automatically links information from the early warning information database to the digital operation and maintenance system, and then puts the work order into the work order pool for queuing through the digital operation and maintenance system.

[0075] Operation and maintenance management involves proactively performing operations and maintenance on the 6 rights work orders, supplementing the corresponding operations and maintenance results and processes into the digital operations and maintenance system, and storing the results in the corresponding database.

[0076] The more specific implementation method is as follows:

[0077] The main implementation steps include the following:

[0078] (1) Obtain main control fault information in real time.

[0079] By utilizing the fault record library information of the main control PLC, the information is stored in a relational or non-relational database. The corresponding main control fault code records are retrieved at custom intervals. The specific data content includes fault name, fault code, masking status, reset status, fault start time, fault end time, fault duration, etc.

[0080] (2) Analyze the fault code triggering situation of the analysis unit using the corresponding rules and decision tree algorithm to obtain the number of triggers and the triggering time.

[0081] Before the program runs, it first checks if fault A has occurred in the historical records. If fault A has not occurred, it proceeds to the next analysis cycle. If fault A has occurred, it checks the reset and masking status. If the status is reset, it proceeds to the next analysis cycle. If the status is not reset or the masking status is unmasked, it checks if fault B has occurred. If fault B has occurred, it checks the reset and masking status. If the status is reset, it proceeds to the next analysis cycle. If the status is not reset or the masking status is unmasked, it checks if fault C has occurred. This process continues until fault D has been analyzed. The corresponding fault time and fault count are recorded.

[0082] (3) Perform decision fault tree analysis on the triggered chain faults to identify the corresponding root causes and faulty components.

[0083] For the chain fault code triggering situation analyzed by the algorithm, the corresponding feature values ​​are extracted to form a feature set u. The set mainly contains feature quantities such as {time interval, fault type, number of faults}. The set is analyzed by associating the fault tree with IBLE to establish the correspondence between the feature set and the root cause of the fault and the faulty component.

[0084] (4) After analyzing and processing the above process, trigger warnings are generated and the trigger information is pushed and displayed with sound and light using the digital operation and maintenance system.

[0085] The digital operation and maintenance system and the intelligent system regularly check whether the above processes trigger early warnings. Once an early warning is detected, the digital operation and maintenance system will issue an alarm via interface pop-up and voice prompts to indicate that an early warning has been triggered.

[0086] (5) The pushed work order information is included in the operation and maintenance work order pool for operation and maintenance work order scheduling management.

[0087] The digital operation and maintenance system generates work orders for triggered warnings and manages the scheduling of generated operation and maintenance work orders.

[0088] The digital operation and maintenance system incorporates triggered early warning work orders into a unified work order schedule, and performs automated scheduling by combining comprehensive information such as weather, operation and maintenance cycle, on-site personnel, and vehicles.

[0089] (6) Investigate and provide feedback on maintenance work orders.

[0090] When an alert work order enters the execution phase, maintenance personnel record the relevant inspection process and results data in the system and evaluate the accuracy of the alert results. The alert algorithm designers then combine this evaluation information to further optimize the alert method in a closed-loop manner, thereby continuously improving the accuracy of the alert results.

[0091] This invention employs an IBLE fault tree early warning method based on master control fault codes. It can fully utilize master control fault triggering information and combine it with expert experience rules to promptly display potential fault hazards of the unit, providing a certain reference and basis for subsequent proactive operation and maintenance of the unit. This helps to avoid wind turbine failure shutdowns, improve the operating efficiency of wind turbine units, and further contribute to the realization of unmanned wind farms.

[0092] Example 2

[0093] like Figures 1 to 6 As shown, as a further optimization of Embodiment 1, this embodiment also includes the following technical features based on Embodiment 1:

[0094] This invention relates to the use of master control fault codes, fault triggering information, and decision tree algorithms to achieve early warning and fault location for generator units.

[0095] The following example illustrates this:

[0096] Direct-drive generator stator circuit to ground short-circuit model

[0097] 1. If a "Class A fault of the frequency converter" is triggered and occurs three times within 72 hours (i.e., after a reset, the state reappears three times), then a warning "generator stator circuit to ground short circuit" will be triggered.

[0098] 2. When the unit's "start-up timeout" fault is triggered, immediately monitor the wind speed at 10 min, 60 s, and 1 s. If all of these values ​​are greater than 5 m / s, then trigger the warning "generator stator circuit to ground short circuit".

[0099] 3. Real-time monitoring of wind speed. If the wind speed is >4m / s for 30 seconds and continues for 30 seconds, and the blade angle is <20° and the wind turbine speed is less than 0.5rpm for 30 seconds, then an early warning "generator stator circuit to ground short circuit" will be triggered.

[0100] 4. Early warning fault reset

[0101] Fault tree analysis is used to locate the cause and corresponding components, and a troubleshooting plan is specified. Manual inspection is performed to ensure the stator circuit is normal. Once it is confirmed that the fault can be reset, a manual reset is performed to eliminate the warning status.

[0102] Algorithm implementation process:

[0103] Based on expert experience, the decision rule tree method is used to reflect the current status of the unit in a timely manner. Once the unit triggers a rule, the triggered warning information is promptly looped into the operation and maintenance system, and the relevant parts and spare parts are quickly located through the expert database.

[0104] IBLE Decision Fault Tree Algorithm:

[0105] IBLE is an example-based learning method based on information theory. In the actual fault warning determination of wind turbines, a specified fault is classified as u1 if it meets the conditions and u2 if it does not. Starting from analyzing the characteristic attributes of the fault, rule analysis will yield three possible conclusions: the specified fault belongs to class u1, class u2, or no judgment can be made, requiring further analysis before a conclusion is reached. During further analysis, the above three scenarios will again occur. For a specific fault analysis, this process continues until a specific category is determined. See details... Figure 4 For more complex problems, in addition to the main rule, sub-rules are added; see details below. Figure 5 .

[0106] The IBLE algorithm generally consists of four parts: initial setup, rule building algorithm, decision tree building algorithm, and category determination algorithm.

[0107] (1) Initial settings;

[0108] Extract features from the object being analyzed, and combine the extracted features into a feature set, with each feature taking a value of {0,1}.

[0109] (2) Rule-building algorithm;

[0110] Calculate the channel capacity CK for each feature AK. When a feature has sub-features, take the sub-feature with the largest C value to represent that feature. The formula for calculating (rounding down) the weight is: WK = [CK * 1000]. Define the decoding functions F(1) and F(0) for the feature AK using the maximum a posteriori criterion. First, calculate the weighted sum of each positive and negative example and fill it into the array A(m,n). Then, sort the array A(m,n) in ascending order of the weighted sum. For positive and negative examples with the same weighted sum but different weighted sums, merge them into a column with the same weighted sum and accumulate the number of positive and negative examples. Calculate the weighted sum tree of all positive and negative examples and obtain the Sp and Sn thresholds from their distribution patterns. These thresholds are used to distinguish between positive and negative examples.

[0111] (3) Decision tree construction algorithm;

[0112] Let T be the space for storing the number of decision rules.

[0113] S31, Initial settings: Extract the features of the main control fault codes, combine the extracted features into a feature set A, convert the feature values ​​into binary values, and each feature value is 0 or 1;

[0114] S32, Rule-building algorithm: For the current training set PE∪NE, construct the main rules using the rule-building algorithm;

[0115] Where PE represents a positive example in the test dataset, and NE represents a negative example in the test dataset;

[0116] S32 specifically includes the following steps:

[0117] S321, For each feature Am: First, transform Am into sub-features Am1, Am2, ..., Amj. Calculate the channel capacity of each sub-feature of Am, and select the sub-feature with the largest channel capacity to represent Am. The weight of Am... Then, use the maximum likelihood decoding criterion to find the values ​​of the sub-features representing Am, and replace them with the corresponding logical symbols;

[0118] S322, take the first m features with larger channel capacity to form a rule;

[0119] S323, calculate the thresholds Sp and Sn by weighting PE and NE based on the selected features and performing statistical analysis.

[0120] Where Sp represents the threshold for classifying an example as positive, and Sn represents the threshold for classifying an example as negative;

[0121] S33, use the main rule to test PE to obtain subsets PEY, PEN, and PEM, and use the main rule to test NE to obtain subsets NEY, NEN, and NEM; where PEY represents a positive example that is judged as a positive example, PEN represents a positive example that is judged as a negative example, PEM represents an example that cannot be judged as a positive example, NEY represents a negative example that is judged as a positive example, NEN represents a negative example that is judged as a negative example, and NEM represents an example that cannot be judged as a negative example;

[0122] S34, Place the main rule into node P;

[0123] S35, Let T be the space for storing the decision rule tree, and set the decision rule tree T to be empty; allocate a new node P, T:=P; where := means assignment, and T:=P means assigning the value of P to T;

[0124] If (|PEY|≠0)∪(|NEY|≠0), then PE:=PEY, NE:=NEY; allocate a new node W1, and the left pointer of P points to W1; then, construct the left division rule for the current training set PE∪NE using the rule building algorithm in step S32, and put the left division rule into node W1;

[0125] If (∧PEN∧≠0)∪(∧NEN∧≠0), then PE:=PEN, NE:=NEN; allocate a new node W2, and the right pointer of P points to W2; then, construct the right division rule for the current training set PE∪NE using the rule building algorithm in step S32, and put the right division rule into node W2;

[0126] If (|PEM|≠0)∪(|NEM|≠0), then PE:=PEM, NE:=NEM, allocate a new node W3, the middle pointer of P points to W3, P:=W3; go to step S32;

[0127] After all test samples have been allocated, proceed to step S39;

[0128] S39, End.

[0129] See the schematic diagram of the decision tree algorithm. Figure 6 .

[0130] (4) Category determination algorithm:

[0131] After obtaining the decision rules, a specific algorithm S40 is given for classifying the unknown entity E.

[0132] S401, Set the root node as the current node;

[0133] S402, Use the rule in the current node to judge E:

[0134] S4021, Judgment rule:

[0135] Features: A1, A2, …, Am;

[0136] Feature values: V1, V2, …, Vm;

[0137] Weights: W1, W2, …, Wm;

[0138] Thresholds: Sp, Sn;

[0139] Logical operators: #1, #2, …, #m;

[0140] Where the logical operator #i = {=, ≠}, i = 1, 2, …, m.

[0141] Let sum = 0;

[0142] if (A1 = V1) then sum := sum + W1;

[0143] if (A2 = V2) then sum := sum + W2;

[0144] if (Am = Vm) then sum := sum + Wm;

[0145] if (sum >= Sp) then the example is a positive example u1;

[0146] if (sum <= Sn) then the example is a negative example u2;

[0147] if (Sn < sum < Sp) then the example cannot be judged;

[0148] S4022, When judged as u1, if the left pointer of the current node is not null (i.e., the left rule exists), set the node pointed to by the left pointer as the current node and go to step S402, otherwise go to step S403;

[0149] When judged as u2, if the right pointer of the current node is not null (i.e., the right rule exists), then set the node pointed to by the right pointer as the current node and go to step S402, otherwise go to step S403;

[0150] When it cannot be judged, if the middle pointer of the current node is not null, then set the node pointed to by the middle node pointer of the current node as the current node and go to step S402, otherwise go to step S403;

[0151] S403, Output the conclusion of being judged as a positive example, a negative example or cannot be judged, and go to step S404;

[0152] Where u1 represents a positive example of the output result and u2 represents a negative example of the output result;

[0153] S404, End.

[0154] As described above, the present invention can be implemented well.

[0155] All features disclosed in all embodiments of this specification, or steps in all methods or processes implied in the disclosure, may be combined and / or extended or replaced in any way, except for mutually exclusive features and / or steps.

[0156] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for early warning of wind turbine generator faults based on master control fault codes, characterized in that, Based on several fault codes triggered by the main control unit, logical rules and decision tree algorithms are used to identify faults that require triggering early warnings. Real-time fault warnings are then provided for wind turbine generator sets that have triggered fault codes, including the following steps: S1, Data Acquisition: Based on the main control fault codes reported by the main control logic of the wind turbine, the main control fault information of the wind turbine is acquired in real time; S2, Feature Extraction: Analyze the main control fault information using logical rules, extract the feature values ​​of the main control fault information, and construct a feature value set from the feature values; S3, Decision Tree Analysis: Perform decision tree analysis on the feature set to identify the corresponding root causes of failures and faulty components; S4, Information Push: Based on the root cause of the fault and the faulty component, identify the fault that needs to trigger an early warning, and push the information that triggers the early warning and provide audible and visual alarm prompts; S5, Work Order Management: Includes the pushed work order information into the maintenance work order pool for maintenance work order scheduling management; S6, Operations and Maintenance Management: Investigate and provide feedback on operations and maintenance work orders; In step S3, for the chain fault code triggering situation analyzed by the algorithm, the corresponding feature values ​​are extracted to form a feature set u. The feature set u is analyzed by the IBLE decision tree algorithm to establish the correspondence between the feature set and the root cause of the fault and the faulty component, thereby obtaining the associated high-risk fault results and storing the analysis results in the database. Step S3 includes the following steps: S31, Initial settings: Extract the features of the main control fault codes, combine the extracted features into a feature set A, convert the feature values ​​into binary values, and each feature value is 0 or 1; S32, Rule-building algorithm: For the current training set PE∪NE, construct the main rules using the rule-building algorithm; Where PE represents a positive example in the test dataset, and NE represents a negative example in the test dataset; S32 specifically includes the following steps: S321, For each feature Am: First, transform Am into sub-features Am1, Am2, ..., Amj. Calculate the channel capacity of each sub-feature of Am, select the sub-feature with the largest channel capacity to represent Am, and assign a weight Wm to Am. Channel capacity representing the sub-characteristics of Am 1000 ,in" " indicates rounding down; then the maximum likelihood decoding criterion is used to find the value of the sub-feature representing Am, and it is replaced with the corresponding logical symbol; S322, take the first m features with larger channel capacity to form a rule; S323, calculate the thresholds Sp and Sn by weighting PE and NE based on the selected features and performing statistical analysis. Where Sp represents the threshold for classifying an example as positive, and Sn represents the threshold for classifying an example as negative; S33, use the main rule to test PE to obtain subsets PEY, PEN, and PEM, and use the main rule to test NE to obtain subsets NEY, NEN, and NEM; where PEY represents a positive example that is judged as a positive example, PEN represents a positive example that is judged as a negative example, PEM represents an example that cannot be judged as a positive example, NEY represents a negative example that is judged as a positive example, NEN represents a negative example that is judged as a negative example, and NEM represents an example that cannot be judged as a negative example; S34, Place the main rule into node P; S35, Let T be the space for storing the decision rule tree, and set the decision rule tree T to empty; allocate a new node P, T:=P; where := means assignment, and T:=P means assigning the value of P to T; If (|PEY|≠0)∪(|NEY|≠0), then PE:=PEY, NE:=NEY; allocate a new node W1, and point the left pointer of P to W1; then, construct the left division rule for the current training set PE∪NE using the rule building algorithm in step S32, and put the left division rule into node W1; If (∧PEN∧≠0)∪(∧NEN∧≠0), then PE:=PEN, NE:=NEN; allocate a new node W2, and the right pointer of P points to W2; then, construct the right division rule for the current training set PE∪NE using the rule building algorithm in step S32, and put the right division rule into node W2; If (|PEM|≠0)∪(|NEM|≠0), then PE:=PEM, NE:=NEM, allocate a new node W3, the middle pointer of P points to W3, P:=W3; go to step S32; After all test samples have been allocated, proceed to step S39; S39, End; Step S40 includes the following steps: S401, Set the root node as the current node; S402, After obtaining the decision rules, how to classify faults E with unknown root causes, and give a specific algorithm; If the result is u1, and the left pointer of the current node is not null, set the node pointed to by the left pointer as the current node and proceed to step S402; otherwise, proceed to step S403. If the result is u2, and the right pointer of the current node is not null, then set the node pointed to by the right pointer as the current node and proceed to step S402; otherwise, proceed to step S403. If the decision cannot be made, and the pointer of the current node is not null, then set the node pointed to by the pointer of the current node as the current node and proceed to step S402; otherwise, proceed to step S403. S403, output the conclusion that it is a positive example, a negative example, or cannot be determined, and proceed to step S404; Here, determining whether a specified fault meets the condition is u1, and not u2; u1 represents a positive example of the output result, and u2 represents a negative example of the output result; S404, End.

2. The method for early warning of wind turbine generator set faults based on master control fault codes according to claim 1, characterized in that, In step S2, the IBLE decision tree algorithm is used to analyze the main control fault information, and the feature values ​​of the main control fault information are extracted according to the expert experience logic rules.

3. A method for early warning of wind turbine generator faults based on master control fault codes according to claim 1 or 2, characterized in that, In step S1, the fault record information of the main control PLC is used to save the fault record information in a relational or non-relational database, and the corresponding main control fault code record is retrieved at a custom time.

4. A wind turbine generator fault early warning system based on master control fault codes, based on the wind turbine generator fault early warning method based on master control fault codes as described in any one of claims 1 to 3, characterized in that, Includes the following modules connected in sequence: Data acquisition module: used to acquire main control fault information of wind turbine in real time based on the main control fault codes reported by the main control logic of the wind turbine. Feature extraction module: Used to analyze the main control fault information using logical rules, extract the feature values ​​of the main control fault information, and construct a feature value set from the feature values; Decision tree analysis module: Used to perform decision tree analysis on the feature set to identify the corresponding root causes and faulty components; Information push module: Used to identify faults that require early warning based on the root cause and faulty components, and push the information that triggers the early warning and provide audible and visual alarm prompts; Work order management module: Used to include the pushed work order information into the operation and maintenance work order pool for operation and maintenance work order scheduling management; Operations and maintenance management module: used to investigate and provide feedback on operations and maintenance work orders.

Images