Wind driven generator fault diagnosis method based on case reasoning

Abstract

The invention provides a wind driven generator fault diagnosis method based on case reasoning. The method comprises the steps that semantic standardization processing is carried out on obtained CMS, SCADA and PSM fault data to obtain a plurality of old tables, and the diagnosis result of each old table comprises fault parts, fault modes, fault removal methods and rectification suggestions; forming a case library by using all old tables; when a new fault is detected, performing semantic standardization processing on newly received CMS, SCADA and PSM fault data to obtain a new table; performing similarity calculation on each old table and the new table in the case library one by one, and selecting the old table with the maximum similarity with the new table as an alternative case; taking the diagnosis result in the alternative case as a new fault diagnosis result, and defining the new fault diagnosis result as a new result; updating the alternative cases by using the new table information; and checking the new result by utilizing PSM to update the case library. According to the method, fault diagnosis is refined to the fault part level of the generator, the precision of a case reasoning method is improved, similar cases are effectively distinguished, and a necessary fault removal method and rectification opinions are provided for operation and maintenance of the generator.

Description

Technical Field

[0001] This invention relates to the technical field of wind turbine generator fault diagnosis, and more specifically, to a case-based reasoning method for wind turbine generator fault diagnosis. Background Technology

[0002] Wind turbines are the core components that convert wind energy into electrical energy. Failures in these turbines directly lead to high downtime losses and maintenance costs. Therefore, intelligent fault diagnosis of wind turbines is crucial for improving the quality of wind power operation and maintenance and reducing maintenance costs.

[0003] Wind turbine operation and maintenance requires fault diagnosis methods to provide component-level fault location, as well as clear troubleshooting methods and rectification suggestions, while simultaneously enabling online and automated condition monitoring. However, currently widely used Supervisory Control and Data Acquisition (SCADA), Condition Monitoring System (CMS), and Periodic Shutdown Maintenance (PSM) systems cannot simultaneously meet these requirements. Therefore, it is necessary to obtain a novel fault diagnosis method through method fusion to overcome the shortcomings of existing methods and meet the needs of generator operation and maintenance.

[0004] Case-based reasoning is a commonly used intelligent fault diagnosis method, which has been widely used in wind power SCADA and CMS. However, existing methods have problems such as single fault feature attribute settings and similarity calculation methods not being applicable to multi-attribute fault features. This makes it difficult to distinguish similar cases, which seriously affects the accuracy of diagnosis. The more reference cases in the case library, the more serious the problem becomes. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art or related art.

[0006] Therefore, the purpose of this invention is to propose a case-based reasoning method for diagnosing wind turbine faults.

[0007] To achieve the above objectives, the technical solution of the present invention provides a case-based reasoning method for wind turbine fault diagnosis. This fault diagnosis method includes: Step S1: acquiring CMS fault data, SCADA fault data, and PSM fault data; Step S2: performing semantic normalization processing on the CMS fault data, SCADA fault data, and PSM fault data, so that the CMS fault data, SCADA fault data, and PSM fault data are merged and converted into several fault feature tables; wherein, the fault feature tables are defined as old tables, and several old tables are arranged in chronological order of fault occurrence; each old table contains detailed fault features... The old tables are converted into classification attribute features and numerical attribute features, and the diagnostic results of each old table include the corresponding faulty components, fault modes, fault troubleshooting methods, and rectification suggestions; Step S4: All the old tables are combined into a case library; Step S5: When SCADA, CMS, or PSM automatically detects a new fault, the newly received CMS fault data, SCADA fault data, and PSM fault data are semantically normalized to merge and convert them into a new fault feature table; wherein, the new fault feature table is defined as a new table, and the new table's... The fault characteristics are consistent with the refined fault characteristics of the old table described in step S2. The fault characteristics of the new table are refined into new table classification attribute characteristics and new table numerical attribute characteristics. Step S6: Based on the old table classification attribute characteristics, old table numerical attribute characteristics, and new table classification attribute characteristics and new table numerical attribute characteristics, the similarity calculation is performed on each old table in the case library and the new table one by one. Based on the calculation results, the old table with the highest similarity to the new table is selected as a candidate case. Step S8: The diagnosis results in the candidate cases are used as the diagnosis results of the new fault, and the diagnosis results of the new fault are defined as new results. Step S9: The information of the new table is used to... The candidate cases are updated, and the diagnostic results of the updated candidate cases are the new results; Step S10: The new results in step S9 are tested using PSM. If the new results in step S9 are consistent with the PSM detection results, the updated candidate cases in step S9 are used as new cases and added to the case library; if the new results in step S9 are inconsistent with the PSM detection results, the new results in step S9 are corrected using the PSM detection results to correct the updated candidate cases in step S9, and the corrected candidate cases are used as new cases and added to the case library.

[0008] Preferably, the semantic normalization processing of the CMS fault data, SCADA fault data, and PSM fault data in step S2 specifically includes: Step S2.1: Setting a normalization statement 1, which is used to characterize the classification attribute features of the old table; the normalization statement 1 contains three fields, namely type, model, and status, and each of the three fields is represented by a variable α. i Indicates i = 1, 2, 3; where type α1 is a keyword that can characterize the excitation mode or structure of the motor; type α1 is composed of Chinese characters, including: doubly fed, permanent magnet, and squirrel cage; model α2 is a combination of letters, numbers, or other characters; state α3 is a keyword that characterizes the generator's operating state when a fault occurs; the keywords include: shutdown, idling, excitation, and generation; step S2.2: set standard statement 2, which is used to characterize the numerical attribute characteristics of the old table; the standard statement 2 contains 8 fields, which are: excitation performance, temperature performance, speed, power, wind speed, ambient temperature, rotational performance, and insulation performance, and each of these 8 fields is represented by variable β. j The expression indicates that j = 1, 2, ..., 8; where, when type α1 is permanent magnet, β1 is the magnetic attenuation degree of the permanent magnet; when type α1 is doubly fed, squirrel cage, or other types, then β1 is the asymmetry of the three-phase excitation current, in %; temperature performance variable β2 is the generator bearing temperature, in °C; speed β3 is the generator rotor speed, in rpm; power β4 is the generator power, in kW; wind speed β5 is the instantaneous wind speed at the time of the fault, in m / s; ambient temperature β6 is the nacelle temperature, in °C; rotational performance β7 is the abnormal harmonic number of the spectrum; insulation performance β8 is the shaft current, in A; step S2.3: set the standard statement 3, which is used to characterize the diagnostic results; the standard statement 3 contains 4 fields, which are: the faulty component, the fault mode, the fault troubleshooting method, and the rectification opinions, and these 4 fields are respectively represented by the variable γ. kIndicated by k = 1, 2, 3, 4; where faulty component γ1 is a Chinese character representing the mechanical structure, power conversion device, and sensor of the generator, including: rotor winding, permanent magnet, rotor core, rotor support, stator winding, cooling system, end cover, bearing, base, carbon brush, speed sensor, and temperature sensor; fault mode γ2 is a Chinese character string in the format of component + electrical fault + mechanical damage; the electrical faults include: magnetic field asymmetry, harmonics, phase loss, overheating, short circuit, and air gap unevenness; the mechanical damage includes: fracture, blockage, corrosion, crack, pulverization, and missing parts. Lubrication, poor welding, looseness, foreign objects, and wear; troubleshooting method γ3 is a keyword characterizing the tool, component, and method; each keyword is connected by the character +; rectification suggestion γ4 is a string composed of the combination and frequency of the most frequently occurring faulty component and fault mode in the case library; the characteristics of type, model, faulty component, fault mode, troubleshooting method, and rectification suggestion are all from PSM; the characteristics of state, excitation performance, temperature performance, speed, power, wind speed, and ambient temperature are all from SCADA; the rotational performance and insulation performance are all from CMS.

[0009] Preferably, the signal sources of the new fault in step S5 specifically include: SCADA alarm signals, CMS alarm signals, and abnormal results detected by PSM.

[0010] Preferably, the classification attributes of the new meter include: type, model, and status; the numerical attributes of the new meter include: excitation performance, temperature performance, rotational speed, power, wind speed, ambient temperature, rotational performance, and insulation performance.

[0011] Preferably, step S6 specifically includes: step S6.1: based on the classification attribute features and numerical attribute features of each old table, and the classification attribute features and numerical attribute features of the new table, perform local similarity calculations on each old table in the case library and the new table one by one; step S6.1 specifically includes: setting δ and θ to be the fault features of the new table and the fault features of each old table, respectively, and setting S to be the local similarity between the fault features of the new table to be calculated and the corresponding fault features of the old table; when the fault features of the new table to be calculated and the corresponding fault features of the old table belong to the classification attribute features, then method 1 or method 2 is used to calculate the local similarity S;

[0012] Method 1 is a binary method; specifically, the binary method includes: when δ = θ, then S = 1; when δ ≠ θ, then S = 0; Method 2 is the maximum membership method; the expression for S in the maximum membership method is:

[0013] S=(δ∩θ) / (δ∪θ) (1)

[0014] In equation (1), the symbol ∩ represents the intersection and ∪ represents the union;

[0015] When the fault characteristics of the new table to be calculated and the corresponding fault characteristics of the old table are numerical attribute characteristics, the local similarity S is calculated using method 3 or method 4.

[0016] Method 3 is the proportional method; the expression for the proportional method S is:

[0017] S=1-|δ-θ| / (θ max -θ min (2)

[0018] In equation (2), θ max and θ min These represent the maximum and minimum values ​​of this fault characteristic across all old tables in the case library, respectively; the symbol || represents the absolute value.

[0019] Method 4 is the distance method; the expression for the distance method S is:

[0020] S=|δ-θ| / max(δ,θ) (3)

[0021] In equation (3), max(δ,θ) represents the larger of δ and θ;

[0022] The local similarity between the calculated numerical attribute features of the new table and the corresponding numerical attribute features of the old table is directly summed to obtain the local similarity of the numerical attribute features; when a numerical attribute feature is missing in the new table or the old table, that numerical attribute feature will no longer participate in the local similarity calculation.

[0023] Step S6.2: Based on the local similarity S calculated in step S6.1, calculate the global similarity S. t ;

[0024] The global similarity S t The calculation expression is:

[0025]

[0026] In equation (4), w i The weight values ​​represent the weights of various attribute features; i represents the attribute feature number. For categorical attribute features, i = 1, and for numerical attribute features, i = 2.

[0027] When there is a situation where the candidate case is not unique, the local similarity calculation methods will be replaced until the situation where the candidate case is not unique is resolved. The replacement method is implemented in the following order: {method 1, method 3}, {method 1, method 4}, {method 2, method 3}, {method 2, method 4}.

[0028] Preferably, the weight values ​​w of each type of attribute feature are determined using the triangular fuzzy number method. i ; and the weight values ​​w of the various attribute features are determined using the triangular fuzzy number method. i Specifically, it includes the following steps: Step S6.21: r experts evaluate w i The value of w was evaluated independently, and all experts gave their respective evaluations. i The minimum, tendency, and maximum values, where the minimum value represents the expert's opinion on w. i The value will not be lower than the indicated value; the propensity score represents the expert's opinion on w. i The most likely value, the maximum value represents what experts believe w to be. i The value that will not be exceeded is represented by the variable l. i,k ,m i,k ,u i,k These represent the opinions of the kth expert on w. i The minimum, tendency, and maximum values ​​of the evaluation results are determined, where i = 1, 2, k = 1, 2, ..., r; Step S6.22: Let variable l i ,m i ,u i These represent the minimum, propensity, and maximum values ​​of all expert evaluation results, respectively, and are based on the formula... Calculate l i According to the formula Calculate m i According to the formula Calculate u i Step S6.23: According to formula w i =1-(u i -l i ) / 2m i Calculate w i Step S6.24: Set variable Ψ, which represents the number of case corrections, and the initial value of Ψ is 0; set the total number of cases in the case library to Φ; when the new result described in step S9 is inconsistent with the PSM detection result, the value of Ψ is automatically incremented by 1; when the condition Ψ / Φ>0.6 is met, return to step S1 for re-voting.

[0029] Preferably, the method for forming the new case specifically includes the following steps: Step S10.1: If the new result described in step S9 is consistent with the PSM detection result, and the new result includes the faulty component, fault mode, fault troubleshooting method, and rectification opinion after semantic normalization processing as described in claim 2, then the updated candidate case described in step S9 is directly used as the new case, and the new case is added to the case library; Step S10.2: If the new result described in step S9 is inconsistent with the PSM detection result, then the faulty component in the PSM detection result is processed according to the semantic normalization method described in claim 2. The fault mode, troubleshooting method, and rectification suggestions are processed, and the updated diagnostic results of the alternative cases described in step S9 are replaced with the processed results. The alternative cases after replacement are used as new cases, and the new cases are added to the case library. The name of the new cases added to the case library in steps S10.1 and S10.2 consists of two parts: the first part is the specific time when SCADA, CMS, or PSM detects a new fault, accurate to the second and without units, as described in step S5; the second part is the three characters "new case". The two parts are connected by the symbol +.

[0030] The beneficial effects of this invention are:

[0031] (1) The case-based reasoning method for wind turbine fault diagnosis provided by this invention integrates CMS fault data, SCADA fault data and PSM fault data through semantic normalization processing, which refines the fault diagnosis from the subsystem level of the wind turbine to the fault component level of the generator, and provides necessary fault troubleshooting methods and rectification opinions for generator operation and maintenance.

[0032] (2) The case-based reasoning method for wind turbine fault diagnosis provided by this invention performs semantic normalization processing on the acquired CMS fault data, SCADA fault data, and PSM fault data. This semantic normalization processing method is applicable to wind turbines, SCADA, CMS, and PSM of any type and manufacturer. Moreover, even if fault data is missing, the most similar alternative case will always be found from the case library as the diagnostic result.

[0033] (3) The case-based reasoning method for wind turbine fault diagnosis provided by the present invention has improved the accuracy of the case reasoning method and effectively distinguished similar cases by meticulously classifying the fault characteristics of the new and old tables, arranging different local similarity calculation methods, and using the triangular fuzzy number method to perform global similarity calculation weight allocation for various attribute characteristics.

[0034] Additional aspects and advantages of the invention will become apparent from the description which follows, or may be learned by practice of the invention. Attached Figure Description

[0035] Figure 1 A schematic flowchart of a case-based reasoning method for wind turbine fault diagnosis according to an embodiment of the present invention is shown.

[0036] Figure 2 A schematic diagram illustrating the semantic normalization process of fault features according to an embodiment of the present invention is shown.

[0037] Figure 3 A schematic block diagram of a faulty component of a generator according to an embodiment of the present invention is shown;

[0038] Figure 4 A fault mode tree model diagram of a generator according to an embodiment of the present invention is shown;

[0039] Figure 5 A bar chart showing the frequency statistics of each fault mode of each faulty component of the generator according to an embodiment of the present invention is shown.

[0040] Figure 6 A schematic diagram of a novel fault detection method according to an embodiment of the present invention is shown;

[0041] Figure 7 This diagram illustrates a schematic flowchart of an embodiment of the present invention, which calculates the similarity between each old table in the case library and the new table one by one.

[0042] Figure 8 A schematic flowchart of an embodiment of the present invention is shown, which corrects the weight values ​​of various attribute features based on the triangular fuzzy number method. Detailed Implementation

[0043] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0044] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0045] Figure 1 A flowchart illustrating a case-based reasoning-based wind turbine fault diagnosis method according to an embodiment of the present invention is shown. Figure 1 As shown, the case-based reasoning-based wind turbine fault diagnosis method includes:

[0046] Step S1: Obtain CMS fault data, SCADA fault data, and PSM fault data;

[0047] Step S2: Perform semantic normalization processing on the CMS fault data, SCADA fault data, and PSM fault data to merge and convert them into several fault feature tables. These fault feature tables are defined as old tables, and several old tables are arranged in chronological order of fault occurrence. The fault features of each old table are refined into old table classification attribute features and old table numerical attribute features. The diagnostic results of each old table include the corresponding faulty component, fault mode, fault troubleshooting method, and rectification suggestions.

[0048] Step S4: Combine all the aforementioned old tables into a case library;

[0049] Step S5: When SCADA, CMS, or PSM automatically detects a new fault, semantic normalization processing is performed on the newly received CMS fault data, SCADA fault data, and PSM fault data to merge and convert them into a new fault feature table. The new fault feature table is defined as a new table, and the fault features of the new table are consistent with the refined fault features of the old table in Step S2. The fault features of the new table are refined into new table classification attribute features and new table numerical attribute features.

[0050] Step S6: Based on the classification attribute features and numerical attribute features of each old table, as well as the classification attribute features and numerical attribute features of the new table, perform similarity calculations on each old table in the case library and the new table one by one. Based on the calculation results, select the old table with the highest similarity to the new table as the candidate case.

[0051] Step S8: Use the diagnostic results in the alternative cases as the diagnostic results of the new fault, and define the diagnostic results of the new fault as the new result;

[0052] Step S9: Update the candidate cases with the information in the new table, and the diagnosis result of the updated candidate cases is the new result;

[0053] Step S10: Use PSM to verify the new result described in step S9. If the new result described in step S9 is consistent with the PSM detection result, then the updated candidate case described in step S9 is used as a new case and added to the case library. If the new result described in step S9 is inconsistent with the PSM detection result, then the new result described in step S9 is corrected using the PSM detection result to correct the updated candidate case described in step S9, and the corrected candidate case is used as a new case and added to the case library.

[0054] In this embodiment, the case-based reasoning-based wind turbine fault diagnosis method provided by the present invention integrates CMS fault data, SCADA fault data, and PSM fault data through semantic normalization processing, which refines the fault diagnosis from the subsystem level of the wind turbine to the fault component level of the generator, and provides necessary troubleshooting methods and rectification suggestions for generator operation and maintenance.

[0055] Furthermore, the case-based reasoning-based wind turbine fault diagnosis method provided by this invention performs semantic normalization processing on the acquired CMS fault data, SCADA fault data, and PSM fault data. This semantic normalization processing method is applicable to wind turbines, SCADA, CMS, and PSM systems of any type and manufacturer. Moreover, even if fault data is missing, the most similar alternative case will always be found from the case library as the diagnostic result.

[0056] Furthermore, the case-based reasoning method for wind turbine fault diagnosis provided by this invention meticulously classifies the fault characteristics of the new and old tables. Then, based on the classification and numerical attributes of each old table, as well as the classification and numerical attributes of the new table, it calculates the similarity between each old table and the new table in the case library. Based on the calculation results, the old table with the highest similarity to the new table is selected as a candidate case. Then, PSM is used to verify and correct the diagnosis results of the new fault, thereby improving the accuracy of the case-based reasoning method and effectively distinguishing similar cases.

[0057] In one embodiment of the present invention, such as Figure 2 As shown, the semantic normalization processing of the CMS fault data, SCADA fault data, and PSM fault data in step S2 specifically includes: Step S2.1: Setting a normalization statement 1, which is used to characterize the classification attribute features of the old table; the normalization statement 1 contains 3 fields, namely type, model, and status, and each of these 3 fields is represented by a variable α. iIndicated by i = 1, 2, 3; where type α1 is a keyword that can characterize the excitation mode or structure of the motor; type α1 is composed of Chinese characters, including: doubly fed, permanent magnet, and squirrel cage; model α2 is a combination of letters, numbers, or other characters; state α3 is a keyword that characterizes the generator's operating state when a fault occurs; the keywords include: shutdown, idling, excitation, and power generation;

[0058] Step S2.2: Set up specification statement 2, which is used to characterize the numerical attribute features of the old table; specification statement 2 contains 8 fields, namely: excitation performance, temperature performance, rotational speed, power, wind speed, ambient temperature, rotational performance, and insulation performance, and each of these 8 fields is represented by a variable β. j Let j = 1, 2, ..., 8; where β1 is the magnetic attenuation degree of the permanent magnet when type α1 is permanent magnet; when type α1 is doubly fed, squirrel cage, or other types, β1 is the asymmetry of the three-phase excitation current, in %; temperature performance variable β2 is the generator bearing temperature, in °C; speed β3 is the generator rotor speed, in revolutions per minute; power β4 is the generator power, in kW; wind speed β5 is the instantaneous wind speed at the time of the fault, in m / s; ambient temperature β6 is the nacelle temperature, in °C; rotational performance β7 is the abnormal harmonic number of the spectrum; insulation performance β8 is the shaft current, in A.

[0059] Step S2.3: Set up standard statement 3, which is used to characterize the diagnostic result; the standard statement 3 contains 4 fields, namely: the faulty component, the fault mode, the fault troubleshooting method, and the rectification suggestions, and each of the 4 fields is represented by a variable γ. k Let k = 1, 2, 3, 4; where, for example... Figure 3 As shown, the faulty component γ1 is represented by Chinese characters indicating the mechanical structure, power conversion components, and sensors of the generator. These characters include: rotor winding, permanent magnet, rotor core, rotor support, stator winding, cooling system, end cover, bearing, base, carbon brush, speed sensor, and temperature sensor; for example... Figure 4 As shown, fault mode γ2 is a string of Chinese characters in the format of "component + electrical fault + mechanical damage"; the electrical faults include: magnetic field asymmetry, harmonics, phase loss, overheating, short circuit, and uneven air gap; the mechanical damage includes: fracture, blockage, corrosion, cracks, pulverization, lack of lubrication, poor welding, loosening, foreign objects, and wear; fault troubleshooting method γ3 is keywords characterizing tools, components, and methods; each keyword is connected by the character "+"; for example... Figure 5As shown, rotor windings, stator windings, and cooling devices are all considered faulty components, while magnetic circuit asymmetry, short circuits, and cracks are all fault modes. The vertical axis represents the total number of occurrences of various fault modes for different faulty components. The higher the number, the more attention needs to be paid to production and maintenance, and the more relevant preventive measures need to be taken. Therefore, this invention uses this number as a rectification suggestion. Rectification suggestion γ4 is a string composed of the combination of the faulty component and fault mode with the highest frequency in the case library, along with the number of occurrences. The characteristics of type, model, faulty component, fault mode, troubleshooting method, and rectification suggestion are all from PSM; the characteristics of state, excitation performance, temperature performance, speed, power, wind speed, and ambient temperature are all from SCADA; and the rotational performance and insulation performance are all from CMS.

[0060] In this embodiment, model α2 is a combination of letters, numbers, or other characters, such as "YJ001-23"; fault mode γ2 is a string of Chinese characters in the format of component + electrical fault + mechanical damage, such as "bearing + overheating + lack of lubrication"; troubleshooting method γ3 is a keyword representing the tool, component, and method, with each keyword connected by the character +, such as "welding gun + rotor winding + welding"; rectification suggestion γ4 is a string composed of the combination and frequency of the faulty component and fault mode that appears most frequently in the case library, such as "bearing overheated 24 times".

[0061] In one embodiment of the present invention, such as Figure 6 As shown, the signal sources of the new fault in step S5 specifically include: SCADA alarm signals, CMS alarm signals, and abnormal results detected by PSM.

[0062] In one embodiment of the present invention, the classification attribute features of the new meter include: type, model, and status; the numerical attribute features of the new meter include: excitation performance, temperature performance, rotational speed, power, wind speed, ambient temperature, rotational performance, and insulation performance.

[0063] In one embodiment of the present invention, such as Figure 7 As shown, step S6 specifically includes: Step S6.1: Based on the classification attribute features and numerical attribute features of each old table, and the classification attribute features and numerical attribute features of the new table, perform local similarity calculations on each old table in the case library and the new table one by one; Step S6.1 specifically includes:

[0064] Let δ and θ be the fault characteristics of the new table and the fault characteristics of each old table, respectively, and let S be the local similarity between the fault characteristics of the new table and the corresponding fault characteristics of the old tables that need to be calculated.

[0065] When the fault features of the new table to be calculated and the corresponding fault features of the old table belong to the classification attribute features, then method 1 or method 2 is used to calculate the local similarity S.

[0066] Method 1 is a binary method; specifically, the binary method includes: when δ = θ, then S = 1; when δ ≠ θ, then S = 0; Method 2 is the maximum membership method; the expression for S in the maximum membership method is:

[0067] S=(δ∩θ) / (δ∪θ) (1)

[0068] In equation (1), the symbol ∩ represents the intersection and ∪ represents the union;

[0069] When the fault characteristics of the new table to be calculated and the corresponding fault characteristics of the old table are numerical attribute characteristics, the local similarity S is calculated using method 3 or method 4.

[0070] Method 3 is the proportional method; the expression for the proportional method S is:

[0071] S=1-|δ-θ| / (θ max -θ min (2)

[0072] In equation (2), θ max and θ min These represent the maximum and minimum values ​​of this fault characteristic across all old tables in the case library, respectively; the symbol || represents the absolute value.

[0073] Method 4 is the distance method; the expression for the distance method S is:

[0074] S=|δ-θ| / max(δ,θ) (3)

[0075] In equation (3), max(δ,θ) represents the larger of δ and θ;

[0076] The local similarity between the calculated numerical attribute features of the new table and the corresponding numerical attribute features of the old table is directly summed to obtain the local similarity of the numerical attribute features; when a numerical attribute feature is missing in the new table or the old table, that numerical attribute feature will no longer participate in the local similarity calculation.

[0077] Step S6.2: Based on the local similarity S calculated in step S6.1, calculate the global similarity S. t ;

[0078] The global similarity S t The calculation expression is:

[0079]

[0080] In equation (4), w i The weight values ​​represent the weights of various attribute features; i represents the attribute feature number. For categorical attribute features, i = 1, and for numerical attribute features, i = 2.

[0081] When there is a situation where the candidate case is not unique, the local similarity calculation methods will be replaced until the situation where the candidate case is not unique is resolved. The replacement method is implemented in the following order: {method 1, method 3}, {method 1, method 4}, {method 2, method 3}, {method 2, method 4}.

[0082] In one embodiment of the present invention, such as Figure 7 and Figure 8 As shown, the weight values ​​w of the various attribute features are determined using the triangular fuzzy number method. i ; and the weight values ​​w of the various attribute features are determined using the triangular fuzzy number method. i Specifically, it includes the following steps:

[0083] Step S6.21: r experts evaluate w i The value of w was evaluated independently, and all experts gave their respective evaluations. i The minimum, tendency, and maximum values, where the minimum value represents the expert's opinion on w. i The value will not be lower than the indicated value; the propensity score represents the expert's opinion on w. i The most likely value, the maximum value represents what experts believe w to be. i The value that will not be exceeded is represented by the variable l. i,k ,m i,k ,u i,k These represent the opinions of the kth expert on w. i The minimum, tendency, and maximum values ​​of the evaluation results, where i = 1, 2, k = 1, 2, ..., r;

[0084] Step S6.22: Let variable l i ,m i ,u i These represent the minimum, propensity, and maximum values ​​of all expert evaluation results, respectively, and are based on the formula... Calculate l i According to the formula Calculate m i According to the formula Calculate u i ;

[0085] Step S6.23: According to formula w i =1-(u i -l i ) / 2m i Calculate w i ;

[0086] Step S6.24: Set variable Ψ, which represents the number of case corrections, and the initial value of Ψ is 0; set the total number of cases in the case library to Φ; when the new result described in step S9 is inconsistent with the PSM detection result, the value of Ψ is automatically incremented by 1; when the condition Ψ / Φ>0.6 is met, return to step S1 for re-voting.

[0087] In one embodiment of the present invention, the method for forming the new case specifically includes the following steps: Step S10.1: If the new result described in step S9 is consistent with the PSM detection result, and the new result includes the faulty component, fault mode, fault troubleshooting method, and rectification opinion after semantic normalization processing as described in claim 2, then the updated candidate case described in step S9 is directly used as the new case, and the new case is added to the case library; Step S10.2: If the new result described in step S9 is inconsistent with the PSM detection result, then the faulty component, fault mode, fault troubleshooting method, and rectification opinion after semantic normalization processing as described in claim 2 are processed according to the semantic normalization method described in claim 2. The faulty components, fault modes, troubleshooting methods, and rectification suggestions are processed, and the diagnostic results of the updated alternative cases described in step S9 are replaced with the processed results. The alternative cases that have been replaced are then used as new cases, and the new cases are added to the case library. The name of the new case added to the case library in steps S10.1 and S10.2 consists of two parts: the first part is the specific time when SCADA, CMS, or PSM detects a new fault, accurate to the second and excluding units, as described in step S5; the second part is the three characters "new case". The two parts are connected by the symbol +.

[0088] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0089] The invention can achieve component-level fault location and pattern recognition by using wind turbine failure cases, and provides fault troubleshooting methods and rectification suggestions in the diagnostic results; in particular, it can accurately distinguish between cases with high similarity; the method is applicable to wind turbines, SCADA, CMS and PSM of any type and manufacturer; even if fault data is missing, the most similar alternative case will always be found from the case library and the diagnostic results of the alternative case will be obtained.

[0090] At 21:21:43 on April 22, 2023, the CMS of a wind turbine generator issued an alarm signal. Upon investigation, the CMS fault information indicated an abnormal second harmonic component in the frequency spectrum and a shaft current of 1A. Further investigation of SCADA data confirmed that the main control unit was in generator mode, with a three-phase excitation current asymmetry of 300%, a generator bearing temperature of 85℃, a generator speed of 1300 rpm, a power output of 150kW, an instantaneous wind speed of 3.8 m / s, and a nacelle temperature of 15℃. The SCADA fault information indicated excessive inverter rotor current asymmetry. After performing PSM (Power Supply Management) troubleshooting, maintenance personnel determined that the generator type was doubly fed, model YJ93A, and that a short circuit had occurred in the generator rotor winding. Using an insulation tester, they located the short circuit point and repaired the armature winding on the tower using repair equipment. The fault was resolved at 11:00:22 on April 24, 2023.

[0091] In this case, SCADA could only locate the fault at the generator rotor wiring connection, but it couldn't further distinguish whether the fault was in the generator or the frequency converter with which it was electrically connected, nor could it clearly identify the internal components of the generator that failed or the specific fault mode. CMS could identify the fault mode as an abnormal second harmonic frequency in the vibration signal spectrum, but the causes of this fault mode are not unique, thus failing to pinpoint the cause of the fault. Furthermore, neither SCADA nor CMS provided any clear suggestions for troubleshooting or corrective measures, failing to guide maintenance operations.

[0092] Acquire CMS fault data, SCADA fault data, and PSM fault data, and perform semantic normalization processing on the CMS fault data, SCADA fault data, and PSM fault data so that the CMS fault data, SCADA fault data, and PSM fault data are merged and converted into several fault feature tables (referred to as old tables).

[0093] Semantic normalization processing is performed on CMS fault data, SCADA fault data, and PSM fault data, such as... Figure 2 As shown, the specific steps include: Step S2.1: Setting a specification statement 1, which is used to characterize the classification attribute features of the old table; the specification statement 1 contains three fields, namely type, model, and status, and each of these three fields is represented by a variable α. i Indicated by i = 1, 2, 3; where type α1 is a keyword that can characterize the excitation mode or structure of the motor; type α1 is composed of Chinese characters, including: doubly fed, permanent magnet, and squirrel cage; model α2 is a combination of letters, numbers, or other characters; state α3 is a keyword that characterizes the generator's operating state when a fault occurs; the keywords include: shutdown, idling, excitation, and power generation;

[0094] Step S2.2: Set up specification statement 2, which is used to characterize the numerical attribute features of the old table; specification statement 2 contains 8 fields, namely: excitation performance, temperature performance, rotational speed, power, wind speed, ambient temperature, rotational performance, and insulation performance, and each of these 8 fields is represented by a variable β. j Let j = 1, 2, ..., 8; where β1 is the magnetic attenuation degree of the permanent magnet when type α1 is permanent magnet; when type α1 is doubly fed, squirrel cage, or other types, β1 is the asymmetry of the three-phase excitation current, in %; temperature performance variable β2 is the generator bearing temperature, in °C; speed β3 is the generator rotor speed, in revolutions per minute; power β4 is the generator power, in kW; wind speed β5 is the instantaneous wind speed at the time of the fault, in m / s; ambient temperature β6 is the nacelle temperature, in °C; rotational performance β7 is the abnormal harmonic number of the spectrum; insulation performance β8 is the shaft current, in A.

[0095] Step S2.3: Set up standard statement 3, which is used to characterize the diagnostic result; the standard statement 3 contains 4 fields, namely: the faulty component, the fault mode, the fault troubleshooting method, and the rectification suggestions, and each of the 4 fields is represented by a variable γ. k Indicated by k = 1, 2, 3, 4; where faulty component γ1 is a Chinese character representing the mechanical structure, power conversion device, and sensor of the generator, including: rotor winding, permanent magnet, rotor core, rotor support, stator winding, cooling system, end cover, bearing, base, carbon brush, speed sensor, and temperature sensor; fault mode γ2 is a Chinese character string in the format of component + electrical fault + mechanical damage; the electrical faults include: magnetic field asymmetry, harmonics, phase loss, overheating, short circuit, and air gap unevenness; the mechanical damage includes: fracture, blockage, corrosion, crack, pulverization, and missing parts. Lubrication, poor welding, looseness, foreign objects, and wear; troubleshooting method γ3 is a keyword characterizing the tool, component, and method; each keyword is connected by the character +; rectification suggestion γ4 is a string composed of the combination and frequency of the most frequently occurring faulty component and fault mode in the case library; the characteristics of type, model, faulty component, fault mode, troubleshooting method, and rectification suggestion are all from PSM; the characteristics of state, excitation performance, temperature performance, speed, power, wind speed, and ambient temperature are all from SCADA; the rotational performance and insulation performance are all from CMS.

[0096] Furthermore, the fault characteristics are shown in Table 1 (old Table 1).

[0097] Table 1 Fault Characteristics Table 1 (Old Table 1)

[0098]

[0099] As shown in Table 1, by semantically normalizing and fusing fault information from SCADA, CMS, and PSM, the fault can be further refined down to the rotor winding component. The specific fault mode of this component is identified as short circuit, and the values ​​of various fault indicators obtained from CMS and SCADA are also provided. More importantly, Table 1 provides clear fault troubleshooting tools and methods, and offers quantitative rectification suggestions based on the fault statistics, which is crucial for guiding operation and maintenance.

[0100] Table 1 uses a standardized semantic representation method, so it is not affected by differences in wind turbine, generator, SCADA and CMS, and PSM types, and has wide applicability.

[0101] The tables are arranged in chronological order of occurrence and named according to the start and end times of the faults, such as Table 1 being named "2023.04.22.21:21:43-2023.04.24.11:00:22". This process is repeated to compile all the old tables into a case library.

[0102] Furthermore, such as Figure 6 As shown, while waiting for alarm information from SCADA and CMS, and for PSM to detect abnormal results, at 11:24:07 on August 15, 2023, the SCADA of a certain wind turbine issued an alarm signal. After querying and repeating the above semantic normalization processing steps, a new table was obtained. The fault characteristics in the new table are consistent with the detailed fault characteristics in the old table. The new table is as follows:

[0103] Table 2 Fault Characteristics Table 2 (New Table)

[0104]

[0105] Furthermore, based on the classification and numerical attributes of each old table, as well as the classification and numerical attributes of the new table, local similarity calculations are performed on each old table and the new table in the case library.

[0106] The step involves calculating the local similarity between each old table and the new table in the case library based on the classification and numerical attributes of each old table, as well as the classification and numerical attributes of the new table. Specifically, this includes:

[0107] Let δ and θ be the fault characteristics of the new table and the fault characteristics of each old table, respectively, and let S be the local similarity between the fault characteristics of the new table and the corresponding fault characteristics of the old tables that need to be calculated.

[0108] When the fault features of the new table to be calculated and the corresponding fault features of the old table belong to the classification attribute features, then method 1 or method 2 is used to calculate the local similarity S.

[0109] Method 1 is a binary method; specifically, the binary method includes: when δ = θ, then S = 1; when δ ≠ θ, then S = 0; Method 2 is the maximum membership method; the expression for S in the maximum membership method is:

[0110] S=(δ∩θ) / (δ∪θ) (1)

[0111] In equation (1), the symbol ∩ represents the intersection and ∪ represents the union;

[0112] When the fault characteristics of the new table to be calculated and the corresponding fault characteristics of the old table are numerical attribute characteristics, the local similarity S is calculated using method 3 or method 4.

[0113] Method 3 is the proportional method; the expression for the proportional method S is:

[0114] S=1-|δ-θ| / (θ max -θ min (2)

[0115] In equation (2), θ max and θ min These represent the maximum and minimum values ​​of this fault characteristic across all old tables in the case library, respectively; the symbol || represents the absolute value.

[0116] Method 4 is the distance method; the expression for the distance method S is:

[0117] S=|δ-θ| / max(δ,θ) (3)

[0118] In equation (3), max(δ,θ) represents the larger of δ and θ;

[0119] The local similarity between the calculated numerical attribute features of the new table and the corresponding numerical attribute features of the old table is directly summed to obtain the local similarity of the numerical attribute features; when a certain numerical attribute feature is missing in the new table or the old table, that numerical attribute feature will no longer participate in the local similarity calculation.

[0120] When a candidate case is not unique, the local similarity calculation methods will be replaced until the case is resolved. The replacement methods are implemented sequentially in the form of combinations of {method 1, method 3}, {method 1, method 4}, {method 2, method 3}, and {method 2, method 4}.

[0121] Furthermore, based on the calculated local similarity S, the global similarity S is calculated.t The global similarity S t The calculation expression is:

[0122]

[0123] In equation (4), w i The weight values ​​represent the weights of various attribute features; i represents the attribute feature number. For categorical attribute features, i = 1, and for numerical attribute features, i = 2.

[0124] In this case, w1 = 0.25, w2 = 0.75, and the final global similarity between the old table 1 and the new table is 2.5. Similarly, the global similarity between each old table and the new table in the case library is calculated sequentially, and the old table with the highest global similarity value is selected as the candidate case. Assuming Table 1 is the candidate case, the fault diagnosis result is the components, fault mode, troubleshooting measures, and rectification suggestions in Table 1, namely, "Rotor winding short circuit; it is recommended to use an insulation tester to detect the rotor short circuit point and to tighten the rotor bars using a tower tool kit. The rotor winding short circuit has occurred 5 times; rectification is recommended."

[0125] Finally, maintenance personnel were assigned to perform PSM (Power Supply Management), which revealed that the fault mode should be "rotor winding + foreign object". Based on the test results, the new table was calibrated, and the components, fault modes, troubleshooting methods and rectification suggestions were updated to obtain the old Table 2 in Table 3, which was then added to the case library.

[0126] Table 3 Fault Characteristics (Old Table 2)

[0127]

[0128] In the above calculation process, even if some fault characteristics of some attributes are missing, such as the missing β1 value, they will not participate in the local similarity calculation, but will not affect the global similarity calculation. In the end, the most similar candidate case will always be found from the case library as the diagnosis result.

[0129] As shown in Table 4, assuming an old Table 3 exists that has the same global similarity as old Table 1, the local similarity calculation scheme is changed. It is found that old Table 3 has a higher global similarity, so it is used as a candidate case. The diagnostic conclusion and old Table 2 will also be modified accordingly. This effectively distinguishes similar cases and solves the problem of uncertain diagnostic results caused by non-unique candidate cases.

[0130] Table 4 Fault Characteristics (Old Table 3)

[0131]

[0132] Furthermore, such as Figure 8 As shown, the triangular fuzzy number method is used to calculate the weight values ​​w of each attribute feature. iSpecifically, it includes the following steps: Step S6.21: r experts evaluate w i The value of w was evaluated independently, and all experts gave their respective evaluations. i The minimum, tendency, and maximum values, where the minimum value represents the expert's opinion on w. i The value will not be lower than the indicated value; the propensity score represents the expert's opinion on w. i The most likely value, the maximum value represents what experts believe w to be. i The value that will not be exceeded is represented by the variable l. i,k ,m i,k ,u i,k These represent the opinions of the kth expert on w. i The minimum, tendency, and maximum values ​​of the evaluation results, where i = 1, 2, k = 1, 2, ..., r;

[0133] Step S6.22: Let variable l i ,m i ,u i These represent the minimum, propensity, and maximum values ​​of all expert evaluation results, respectively, and are based on the formula... Calculate l i According to the formula Calculate m i According to the formula Calculate u i ;

[0134] Step S6.23: According to formula w i =1-(u i -l i ) / 2m i Calculate w i ;

[0135] Step S6.24: Set variable Ψ, which represents the number of case corrections, and the initial value of Ψ is 0; set the total number of cases in the case library to Φ; when the new result described in step S9 is inconsistent with the PSM detection result, the value of Ψ is automatically incremented by 1; when the condition Ψ / Φ>0.6 is met, return to step S1 for re-voting.

[0136] In this case, the final calculation results are w1 = 0.25 and w2 = 0.75. The false diagnosis rate of the case-based reasoning wind turbine fault diagnosis method provided by this invention is statistically analyzed using PSM. When the false diagnosis rate exceeds 60%, a revote is conducted.

[0137] In summary, the case-based reasoning-based wind turbine fault diagnosis method provided by this invention achieves the fusion of SCADA fault data, CMS fault data, and PSM fault data through semantic normalization processing. It also meticulously classifies the fault characteristics of the new and old tables, arranges different local similarity calculation methods, and uses the triangular fuzzy number method to perform global similarity calculation weight allocation for various attribute features. This enables component-level fault diagnosis, improves the accuracy of the case-based reasoning method, effectively distinguishes similar cases, and provides necessary fault troubleshooting methods and rectification suggestions for generator operation and maintenance.

[0138] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A case-based reasoning-based method for diagnosing wind turbine faults, characterized in that, include: Step S1: Obtain CMS fault data, SCADA fault data, and PSM fault data; Step S2: Perform semantic normalization processing on the CMS fault data, SCADA fault data, and PSM fault data to merge and convert them into several fault feature tables. These fault feature tables are defined as old tables, and several old tables are arranged in chronological order of fault occurrence. The fault features of each old table are refined into old table classification attribute features and old table numerical attribute features. The diagnostic results of each old table include the corresponding faulty component, fault mode, fault troubleshooting method, and rectification suggestions. Step S4: Combine all the aforementioned old tables into a case library; Step S5: When SCADA, CMS, or PSM automatically detects a new fault, semantic normalization processing is performed on the newly received CMS fault data, SCADA fault data, and PSM fault data to merge and convert them into a new fault feature table. The new fault feature table is defined as a new table, and the fault features of the new table are consistent with the refined fault features of the old table in Step S2. The fault features of the new table are refined into new table classification attribute features and new table numerical attribute features. Step S6: Based on the classification attribute features and numerical attribute features of each old table, as well as the classification attribute features and numerical attribute features of the new table, perform similarity calculations on each old table in the case library and the new table one by one. Based on the calculation results, select the old table with the highest similarity to the new table as the candidate case. Step S8: Use the diagnostic results in the alternative cases as the diagnostic results of the new fault, and define the diagnostic results of the new fault as the new result; Step S9: Update the candidate cases with the information in the new table, and the diagnosis result of the updated candidate cases is the new result; Step S10: Use PSM to verify the new result described in step S9. If the new result described in step S9 is consistent with the PSM detection result, then the updated candidate case described in step S9 is used as a new case and added to the case library. If the new result described in step S9 is inconsistent with the PSM detection result, then the new result described in step S9 is corrected using the PSM detection result to correct the updated candidate case described in step S9, and the corrected candidate case is used as a new case and added to the case library.

2. The wind turbine generator fault diagnosis method based on case-based reasoning according to claim 1, characterized in that, The semantic normalization processing of the CMS fault data, SCADA fault data, and PSM fault data in step S2 specifically includes: Step S2.1: Set the specification statement 1, which is used to characterize the classification attribute features of the old table; the specification statement 1 contains 3 fields, namely type, model, and status, and each of the 3 fields is represented by a variable α. i This means that i = 1, 2, 3; Among them, type α1 is a keyword that can characterize the excitation mode or structure of the motor; the type α1 is composed of Chinese characters, including: doubly fed, permanent magnet and squirrel cage; model α2 is a combination of letters, numbers or other characters; state α3 is a keyword that characterizes the generator's operating state when a fault occurs; the keywords include: shutdown, idling, excitation and power generation; Step S2.2: Set up specification statement 2, which is used to characterize the numerical attribute features of the old table; specification statement 2 contains 8 fields, namely: excitation performance, temperature performance, rotational speed, power, wind speed, ambient temperature, rotational performance, and insulation performance, and each of these 8 fields is represented by a variable β. j Indicates that j = 1, 2, ..., 8; Wherein, when type α1 is permanent magnet, β1 is the magnetic attenuation degree of the permanent magnet; when type α1 is doubly fed, squirrel cage, or other types, β1 is the asymmetry of the three-phase excitation current, in %; temperature performance variable β2 is the generator bearing temperature, in °C; speed β3 is the generator rotor speed, in rpm; power β4 is the generator power, in kW; wind speed β5 is the instantaneous wind speed at the moment of fault occurrence, in m / s; ambient temperature β6 is the nacelle temperature, in °C; rotational performance β7 is the abnormal harmonic number of the spectrum; insulation performance β8 is the shaft current, in A. Step S2.3: Set up standard statement 3, which is used to characterize the diagnostic result; the standard statement 3 contains 4 fields, namely: the faulty component, the fault mode, the fault troubleshooting method, and the rectification suggestions, and each of the 4 fields is represented by a variable γ. k Indicates that k = 1, 2, 3, 4; Among them, the faulty component γ1 is a Chinese character representing the mechanical structure, power conversion device and sensor of the generator. The Chinese character includes: rotor winding, permanent magnet, rotor core, rotor support, stator winding, cooling system, end cover, bearing, base, carbon brush, speed sensor and temperature sensor. Fault mode γ2 is a Chinese character string in the format of component + electrical fault + mechanical damage; the electrical faults include: magnetic field asymmetry, harmonics, phase loss, overheating, short circuit, and uneven air gap; the mechanical damage includes: fracture, blockage, corrosion, crack, powdering, lack of lubrication, poor welding, loosening, foreign objects, and wear; troubleshooting method γ3 is a keyword representing the tool, component, and method; the keywords are connected by the character +; rectification suggestion γ4 is a string composed of the combination and frequency of the most frequently occurring faulty component and fault mode in the case library; The characteristics of type, model, faulty component, fault mode, troubleshooting method and rectification suggestion are all from PSM; the characteristics of state, excitation performance, temperature performance, speed, power, wind speed and ambient temperature are all from SCADA; the rotation performance and insulation performance are all from CMS.

3. The wind turbine generator fault diagnosis method based on case-based reasoning according to claim 1, characterized in that, The signal sources for the new fault in step S5 specifically include: SCADA alarm signals, CMS alarm signals, and abnormal results detected by PSM.

4. The wind turbine generator fault diagnosis method based on case-based reasoning according to claim 2, characterized in that, The new table's classification attributes include: type, model, and status; the new table's numerical attributes include: excitation performance, temperature performance, rotational speed, power, wind speed, ambient temperature, rotational performance, and insulation performance.

5. The wind turbine generator fault diagnosis method based on case-based reasoning according to claim 1, characterized in that, Step S6 specifically includes: Step S6.1: Based on the classification attribute features and numerical attribute features of each old table, as well as the classification attribute features and numerical attribute features of the new table, perform local similarity calculations on each old table in the case library and the new table one by one. Step S6.1 specifically includes: Let δ and θ be the fault characteristics of the new table and the fault characteristics of each old table, respectively, and let S be the local similarity between the fault characteristics of the new table and the corresponding fault characteristics of the old tables that need to be calculated. When the fault features of the new table to be calculated and the corresponding fault features of the old table belong to the classification attribute features, then method 1 or method 2 is used to calculate the local similarity S. Method 1 is a binary method; specifically, the binary method includes: when δ = θ, then S = 1; when δ ≠ θ, then S = 0; Method 2 is the maximum membership method; the expression for S in the maximum membership method is: S=(δ∩θ) / (δ∪θ) (1) In equation (1), the symbol ∩ represents the intersection and ∪ represents the union; When the fault characteristics of the new table to be calculated and the corresponding fault characteristics of the old table are numerical attribute characteristics, the local similarity S is calculated using method 3 or method 4. Method 3 is the proportional method; the expression for the proportional method S is: S=1-|δ-θ| / (θ max -θ min ) (2) In equation (2), θ max and θ min These represent the maximum and minimum values ​​of this fault characteristic across all old tables in the case library, respectively; the symbol || represents the absolute value. Method 4 is the distance method; the expression for the distance method S is: S=|δ-θ| / max(δ,θ) (3) In equation (3), max(δ,θ) represents the larger of δ and θ; The local similarity between the calculated numerical attribute features of the new table and the corresponding numerical attribute features of the old table is directly summed to obtain the local similarity of the numerical attribute features; when a numerical attribute feature is missing in the new table or the old table, that numerical attribute feature will no longer participate in the local similarity calculation. Step S6.2: Based on the local similarity S calculated in step S6.1, calculate the global similarity S. t ; The global similarity S t The calculation expression is: In equation (4), w i The weight values ​​represent the weights of various attribute features; i represents the attribute feature number. For categorical attribute features, i = 1, and for numerical attribute features, i = 2. When there is a situation where the candidate case is not unique, the local similarity calculation methods will be replaced until the situation where the candidate case is not unique is resolved. The replacement method is implemented in the following order: {method 1, method 3}, {method 1, method 4}, {method 2, method 3}, {method 2, method 4}.

6. The wind turbine generator fault diagnosis method based on case-based reasoning according to claim 5, characterized in that, The weight values ​​w of each attribute feature are determined using the triangular fuzzy number method. i ; as well as The weight values ​​w of each attribute feature are determined using the triangular fuzzy number method. i Specifically, it includes the following steps: Step S6.21: r experts evaluate w i The value of w was evaluated independently, and all experts gave their respective evaluations. i The minimum, tendency, and maximum values, where the minimum value represents the expert's opinion on w. i The value will not be lower than the indicated value; the propensity score represents the expert's opinion on w. i The most likely value, the maximum value represents what experts believe w to be. i The value that will not be exceeded is represented by the variable l. i,k ,m i,k ,u i,k These represent the opinions of the kth expert on w. i The minimum, tendency, and maximum values ​​of the evaluation results, where i = 1, 2, k = 1, 2, ..., r; Step S6.22: Let variable l i ,m i ,u i These represent the minimum, propensity, and maximum values ​​of all expert evaluation results, respectively, and are based on the formula... Calculate l i According to the formula Calculate m i According to the formula Calculate u i ; Step S6.23: According to formula w i =1-(u i -l i ) / 2m i Calculate w i ; Step S6.24: Set variable Ψ, which represents the number of case corrections, and the initial value of Ψ is 0; set the total number of cases in the case library to Φ; when the new result described in step S9 is inconsistent with the PSM detection result, the value of Ψ is automatically incremented by 1; when the condition Ψ / Φ>0.6 is met, return to step S1 for re-voting.

7. The wind turbine generator fault diagnosis method based on case-based reasoning according to claim 2, characterized in that, The method for forming the new case specifically includes the following steps: Step S10.1: If the new result described in step S9 is consistent with the PSM detection result, and the new result includes the faulty component, fault mode, fault troubleshooting method and rectification opinion after semantic normalization processing as described in claim 2, then the updated alternative case described in step S9 is directly treated as a new case, and the new case is added to the case library. Step S10.2: If the new result described in step S9 is inconsistent with the PSM detection result, the semantic normalization method described in claim 2 is used to process the faulty components, fault modes, fault troubleshooting methods and rectification opinions in the PSM detection result, and the processed result replaces the diagnosis result of the updated alternative case described in step S9. The alternative case after replacement is used as a new case, and the new case is added to the case library. The name of the new case added to the case library in steps S10.1 and S10.2 consists of two parts: the first part is the specific time when SCADA, CMS or PSM detects a new fault as described in step S5, accurate to the second and without units; the second part is the three characters "new case". The two parts are connected by the symbol +.

Images