A method for operation and maintenance of a forced air highland type doubly-fed asynchronous wind turbine generator converter

Abstract

The present application relates to the technical field of converter, particularly to a kind of air-cooled highland type doubly-fed asynchronous wind turbine generator set converter operation and maintenance method, comprising the following steps: based on converter cabinet airflow path, monitoring point is arranged on converter cabinet, real-time acquisition inlet airflow parameter, power module heat sink local airflow parameter and exhaust outlet airflow parameter, in combination with the altitude of converter cabinet installation site, altitude gradient correction is carried out to airflow parameter determination threshold, the flow velocity gradient deviation and temperature gradient deviation of three airflow sections of inlet section, heat sink section and exhaust section are calculated, the flow coefficient around power module heat sink is synchronously monitored, by monitoring point along the inherent airflow path of converter cabinet, the real-time quantitative acquisition of three airflow key parameters of inlet, heat sink and exhaust is realized, airflow dead zone problem is converted into controllable and analysable feature data, and the blank of highland converter heat dissipation state online monitoring is filled.

Description

Technical Field

[0001] This invention relates to the field of converter technology, and in particular to an operation and maintenance method for a wind-cooled, high-altitude doubly-fed asynchronous wind turbine converter. Background Technology

[0002] As the wind power industry gradually expands to high-altitude plateau regions, the high-altitude doubly-fed asynchronous wind turbine converter, as a core component for wind power grid connection, directly determines the reliable operating time and power generation efficiency of the wind farm through its heat dissipation stability. In plateau regions, the air is thin and the air pressure is much lower than in plains areas, naturally reducing the airflow heat exchange efficiency of conventional air-cooled converter cabinets. Furthermore, the fixed structure of the cabinet—bottom air intake, middle power module placement, and top exhaust—easily creates fixed airflow dead zones in the areas where the power module heat sinks and busbars overlap. This is the most common and frequent heat dissipation problem in the operation of high-altitude air-cooled converters, directly leading to potential hazards such as localized overheating and accelerated component aging.

[0003] Existing technologies mainly focus on active heat dissipation regulation during converter operation, achieving immediate overheat protection by adjusting fan speed and air intake guide structure. These technologies can only solve the heat dissipation problem under short-term operating conditions and do not provide a dedicated operation and maintenance solution for the core issue of airflow dead zone.

[0004] The existing heat dissipation maintenance methods for high-altitude air-cooled converter cabinets have significant shortcomings. For the core pain point of airflow dead zone, conventional maintenance relies solely on periodic manual inspections and reactive emergency repairs after overheating. There are no quantitative monitoring methods based on the inherent airflow characteristics of the cabinet, making it impossible to predict the aggravation of airflow dead zone and local heat exchange failure in advance. Often, the problem is only addressed when the equipment alarms for overheating, which can easily lead to damage to power modules and unplanned shutdowns of the wind farm.

[0005] Therefore, a method for the operation and maintenance of converters for air-cooled, high-altitude doubly-fed asynchronous wind turbines is proposed. Summary of the Invention

[0006] In this section, as well as in the abstract and title of this application, some simplifications or omissions may be made to avoid obscuring the purpose of this section, the abstract, and the title of this application, and such simplifications or omissions shall not be used to limit the scope of the invention.

[0007] To address the shortcomings of existing technologies, one objective of this invention is to provide an operation and maintenance method for a wind-cooled, high-altitude doubly-fed asynchronous wind turbine converter.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a method for the operation and maintenance of a wind-cooled high-altitude doubly-fed asynchronous wind turbine converter, comprising the following steps: based on the airflow path of the converter cabinet, monitoring points are set up on the converter cabinet to collect in real time the airflow parameters at the inlet, the local airflow parameters of the power module heat dissipation surface, and the airflow parameters at the exhaust outlet. Combined with the altitude of the converter cabinet installation site, the airflow parameter judgment threshold is corrected by altitude gradient. The velocity gradient deviation and temperature gradient deviation of the three airflow sections (inlet section, heat dissipation surface section, and exhaust section) are calculated. The flow around the power module heat dissipation surface is monitored simultaneously. Based on this, it is determined whether an airflow dead zone has been formed inside the cabinet and the overall operating status of the heat dissipation system. According to the deviation level and the corresponding fault type, graded operation and maintenance actions are performed.

[0009] As a preferred embodiment of the operation and maintenance method for the converter of the air-cooled high-altitude doubly fed asynchronous wind turbine described in this invention, the converter cabinet adopts a layout with bottom air intake, power modules arranged in the middle, and exhaust fans set at the top, forming an air intake airflow path from bottom to top. There are no additional guide baffles inside the converter cabinet, forming a vertical air intake airflow path from bottom to top. After entering from the bottom, the cold air flows through the heat sink of the power module in the middle, carrying away the heat generated by the operation of the power module, and is finally forcibly discharged by the top fan.

[0010] As a preferred embodiment of the operation and maintenance method for the converter of a wind-cooled high-altitude doubly-fed asynchronous wind turbine described in this invention, the monitoring points are all arranged based on the converter cabinet structure. Specifically, the monitoring points are arranged as follows: the bottom air inlet monitoring point is located at the bottom air inlet channel of the cabinet or at the louver on the side of the bottom of the converter cabinet, used to collect airflow velocity and ambient temperature as reference parameters for airflow attenuation at high altitudes; the power module heat dissipation surface monitoring point is located on the windward side of the power module heat sink in the middle of the converter cabinet, used to collect local airflow velocity, temperature and bypass coefficient to determine the formation state of airflow dead zones; and the top exhaust outlet monitoring point is located at the exhaust fan outlet and / or the top louver at the top of the cabinet, used to collect exhaust air velocity and exhaust air temperature to determine fan extraction efficiency and overall airflow circulation effect.

[0011] As a preferred embodiment of the operation and maintenance method for the air-cooled high-altitude doubly-fed asynchronous wind turbine converter described in this invention, the altitude gradient correction rules include: velocity gradient threshold correction: based on the installation altitude, an altitude compensation correction is performed on a preset velocity gradient reference threshold, the velocity gradient reference threshold including a first ratio reference of the heat dissipation surface velocity to the inlet air velocity, and a second ratio reference of the exhaust air velocity to the heat dissipation surface velocity; temperature gradient threshold correction: based on the installation altitude, an altitude compensation correction is performed on a preset temperature rise gradient reference threshold, the temperature rise gradient reference threshold including a first temperature rise reference of the heat dissipation surface temperature to the inlet air temperature, and a second temperature rise reference of the exhaust air temperature to the heat dissipation surface temperature; and the flow characteristic parameter threshold adopts a uniform preset value across the entire altitude range of the plateau.

[0012] As a preferred embodiment of the operation and maintenance method for the converter of the air-cooled high-altitude doubly fed asynchronous wind turbine described in this invention, the fault determination logic is directly based on the pattern matching of airflow feature vectors. Specifically, it includes: pre-storing feature vector benchmarks corresponding to at least three fault modes, wherein the feature vector benchmarks include a combination of features of velocity gradient correlation deviation, temperature rise gradient correlation deviation and flow around feature parameters; collecting and constructing feature vectors of the current airflow state in real time; calculating the similarity between the real-time feature vectors and the feature vector benchmarks of each fault mode; and matching the corresponding fault type when the similarity exceeds a preset matching threshold.

[0013] As a preferred embodiment of the operation and maintenance method for the converter of the air-cooled high-altitude doubly fed asynchronous wind turbine described in this invention, the fault types corresponding to the matching include: the feature vector showing a significant deviation in the velocity gradient of the exhaust section, while the parameters of the intake section and the flow characteristics around the heat dissipation surface are relatively normal, and the matching judgment is that the fan bearing is worn or the speed is reduced; the feature vector showing a significant exceedance of the flow characteristic parameters around the heat dissipation surface, while the velocity gradient of the intake section and the exhaust section is basically normal, and the matching judgment is that the heat dissipation surface is dusty or the heat dissipation fins are deformed; the feature vector showing a comprehensive deviation in the velocity gradient and temperature rise gradient of the intake section, the heat dissipation surface section, and the exhaust section, and the matching judgment is that the connection between the tower and the cabinet air intake channel is not smooth.

[0014] As a preferred embodiment of the operation and maintenance method for the wind-cooled high-altitude doubly fed asynchronous wind turbine converter described in this invention, the method further includes a dynamic update step for the feature vector benchmark: after each fault handling is completed and the fault is confirmed to be eliminated, the feature vector data before the occurrence of the fault is used as new samples, and the feature vector benchmark corresponding to the fault mode is updated with weight. When the cumulative number of samples of the same fault mode reaches a preset number, the confidence interval of the feature vector benchmark of the fault mode is recalculated, and the preset matching threshold is optimized accordingly.

[0015] As a preferred embodiment of the operation and maintenance method for the wind-cooled high-altitude doubly fed asynchronous wind turbine converter described in this invention, the calculation of the similarity between the real-time feature vector and the feature vector benchmarks of each fault mode further includes: calculating the Euclidean distance, cosine similarity, and Mahalanobis distance between the real-time feature vector and the feature vector benchmarks of each fault mode respectively, generating a multi-dimensional similarity evaluation matrix, and performing weighted fusion on the multi-dimensional similarity evaluation matrix to obtain a comprehensive similarity. When the comprehensive similarity exceeds a preset matching threshold and the similarity difference between the fault mode with the highest similarity is greater than a set discrimination threshold, the fault mode is finally matched.

[0016] As a preferred embodiment of the operation and maintenance method for the converter of the wind-cooled high-altitude doubly-fed asynchronous wind turbine described in this invention, the graded operation and maintenance actions are in line with the conventional operation and maintenance process of the wind farm, including: when there is no deviation in airflow parameters, calibrating the monitoring sensors every quarter and cleaning the dust on the air intake channel and heat dissipation surface; when the gradient deviation of airflow parameters reaches the preset warning range, adding high-altitude special grease to the fan bearing and correcting the spacing of the power module heat sink in combination with the planned shutdown period of the wind farm; when the deviation exceeds the threshold, graded handling is carried out; when the airflow dead zone exceeds the standard, the heat dissipation surface is cleaned on-site after remote load reduction; when a fan failure or overall airflow failure occurs, the unit is immediately shut down for handling.

[0017] As a preferred embodiment of the operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine generator described in this invention, it further includes a step of self-optimization based on fault recording data. After a fault is detected, the original waveform data recorded inside the converter for a preset time before and after the fault is retrieved. The original waveform data includes the instantaneous value sequence of airflow parameters at the inlet, local airflow parameters on the heat dissipation surface of the power module, and airflow parameters at the outlet. The original waveform data is compared with the real-time feature vector in the time-frequency domain to extract the gradual change features of the airflow parameters before the fault and the abrupt change features when the fault occurs. When the cumulative number of samples of the same fault type reaches a preset number, the early warning threshold of the fault type is optimized based on the extracted gradual change features, and the matching weight of the feature vector benchmark is optimized based on the abrupt change features to achieve adaptive optimization of the fault diagnosis model.

[0018] The beneficial effects of the operation and maintenance method for air-cooled high-altitude doubly-fed asynchronous wind turbine converters of this invention are as follows: By deploying monitoring points along the inherent airflow path of the converter cabinet, real-time quantitative acquisition of key parameters of the airflow in the three sections—intake, heat dissipation surface, and exhaust—is achieved. This transforms the problem of airflow dead zones into monitorable and analyzable characteristic data, filling the gap in online monitoring of the heat dissipation status of converters in high-altitude areas. By introducing altitude gradient correction rules and dynamically calibrating the judgment thresholds for flow velocity and temperature rise, the interference of thin air in high-altitude areas on heat exchange efficiency assessment is effectively eliminated, ensuring the universality and accuracy of the diagnostic logic in different altitude regions. By jointly analyzing the velocity gradient deviation, temperature gradient deviation, and flow coefficient around the heat dissipation surface of the three airflow sections, it is possible not only to accurately determine whether an airflow dead zone has formed inside the cabinet, but also to comprehensively assess the overall operating status of the cooling system, including fans and ducts, realizing intelligent fault diagnosis from single-parameter alarms to multi-source feature fusion. Based on the deviation level and fault type, a graded operation and maintenance approach is implemented, transforming reactive emergency repairs into proactive prevention and graded control during the event. Under the premise of ensuring equipment safety, the number of unplanned shutdowns is minimized, significantly improving the operational reliability and power generation efficiency of plateau wind turbine units. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the overall invention.

[0021] Figure 2 This is a schematic diagram of the airflow path of the present invention. Detailed Implementation

[0022] To make the objectives, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] 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 those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figure 1 This embodiment provides a method for the operation and maintenance of a converter for a wind-cooled, high-altitude doubly-fed asynchronous wind turbine, including the following steps: First, based on the airflow path of the converter cabinet 100, monitoring point A is set up on the converter cabinet 100 to collect airflow parameters of the air inlet 101, local airflow parameters of the heat dissipation surface of the power module 102, and airflow parameters of the exhaust outlet 103 in real time. Specifically, based on the directional airflow path inside the converter cabinet 100, and relying on the cabinet structure, monitoring point A, which is resistant to high-altitude electromagnetic interference and has a wide temperature range, is set up at key airflow nodes. The points cover the three core areas of the air inlet 101, the heat dissipation surface of the power module 102, and the exhaust outlet 103. The airflow velocity and ambient temperature parameters of the air inlet 101, the local airflow velocity, temperature, and flow coefficient parameters of the heat dissipation surface of the power module 102, and the airflow velocity and exhaust temperature parameters of the exhaust outlet 103 are collected in real time through a high-frequency acquisition module. The acquisition frequency is not less than 10Hz to avoid data distortion caused by high-altitude electromagnetic interference.

[0026] Secondly, based on the altitude of the converter cabinet 100 installation site, the airflow parameter judgment threshold is corrected by altitude gradient. According to the three plateau gradients of 2000~3000m, 3000~4000m, and 4000~5500m, the airflow parameter judgment threshold is dynamically compensated and corrected to eliminate the influence of air density differences caused by altitude.

[0027] Then, through the edge computing unit, the velocity gradient deviation and temperature gradient deviation of the airflow in the three sections of the air intake section, heat dissipation surface section and exhaust section are calculated in real time. The flow coefficient around the heat dissipation surface of the power module 102 is monitored synchronously and continuously. Through multi-dimensional parameter coupling analysis, it is accurately determined whether a local airflow dead zone is formed inside the cabinet. At the same time, the overall operating status of the heat dissipation system composed of the cooling fan, air duct and heat sink is evaluated.

[0028] Finally, based on the gradient deviation level classification, the corresponding fault type is matched, and differentiated hierarchical operation and maintenance actions are performed according to the three-level standards of early warning, load reduction, and shutdown.

[0029] Furthermore, the converter cabinet 100 adopts a bottom air intake, a middle arrangement of power modules 102, and a top exhaust fan layout, forming an upward airflow path. There are no additional airflow guides or baffles inside the cabinet, creating a vertically continuous upward airflow path. After entering from the bottom, the cold air flows through the heat sink of the power module 102 in the middle, carrying away the heat generated by the power module 102 during operation, and is finally forced out by the top fan. The airflow path is single and directional, which facilitates the precise placement of monitoring point A, ensuring that the monitoring data can truly reflect the overall heat dissipation effect. It also facilitates the precise placement of three-segment monitoring points A, ensuring strong data acquisition continuity and avoiding turbulence that could cause monitoring data deviation. It is compatible with the structure of most air-cooled high-altitude converter cabinets 100, has strong versatility, and solves the core problems of conventional threshold failure in high-altitude areas, inability to predict airflow dead zones, and indiscriminate fault handling, enabling accurate monitoring of heat dissipation status in high-altitude environments.

[0030] By deploying monitoring points along the inherent airflow path of the converter cabinet 100, real-time quantitative acquisition of key parameters of the airflow in the three sections—intake, heat dissipation surface, and exhaust—was achieved. This transformed the problem of airflow dead zones into monitorable and analyzable characteristic data, filling the gap in online monitoring of the heat dissipation status of converters at high altitudes. By introducing altitude gradient correction rules and dynamically calibrating the thresholds for velocity and temperature rise, the interference of thin air at high altitudes on heat exchange efficiency assessment was effectively eliminated, ensuring the universality and accuracy of the diagnostic logic across different altitudes. Through joint analysis of velocity gradient deviation, temperature gradient deviation, and the flow coefficient around the heat dissipation surface in the three airflow sections, it is possible not only to accurately determine whether airflow dead zones have formed within the cabinet but also to comprehensively assess the overall operating status of the cooling system, including fans and ducts. This achieves intelligent fault diagnosis, moving from single-parameter alarms to multi-source feature fusion. Based on deviation levels and fault types, graded operation and maintenance are implemented, transforming reactive post-incident repairs into proactive pre-incident prevention and in-process graded control. While ensuring equipment safety, this minimizes unplanned downtime and significantly improves the operational reliability and power generation efficiency of high-altitude wind turbine units.

[0031] Furthermore, all monitoring points A are deployed based on the 100-meter structure of the converter cabinet, requiring no modification to the cabinet structure and without compromising the cabinet's protection level. The specific deployment locations are as follows: The bottom air inlet 101 monitoring point is located at the bottom air inlet channel of the cabinet or inside the louver on the bottom side of the converter cabinet 100, avoiding the dust accumulation blind area. An integrated flow rate and temperature sensor is used to collect the airflow velocity and ambient reference temperature in real time. This data serves as the benchmark parameter for calculating the attenuation of airflow at high altitudes and heat dissipation efficiency, eliminating the interference of ambient temperature.

[0032] The monitoring point for the heat dissipation surface of the power module 102 is located at the center of the windward side of the heat sink of the power module 102 in the middle of the converter cabinet 100. One monitoring point is set up for each row of power modules 102. A miniature three-dimensional airflow sensor is used to synchronously collect local airflow velocity, temperature and flow coefficient. The fluctuation value of the flow coefficient is used to directly determine whether the local airflow is turbulent or whether a dead zone of airflow has been formed.

[0033] The top exhaust outlet monitoring point 103 is located at the center of the exhaust fan outlet at the top of the cabinet and on the inside of the top exhaust louver. Two monitoring points are symmetrically arranged to collect the exhaust air velocity and exhaust air temperature. The exhaust air velocity is used to determine the fan extraction efficiency, and the temperature difference between the inlet and outlet air is used to determine the overall airflow circulation and heat exchange effect.

[0034] The non-destructive deployment does not damage the cabinet structure, is compatible with the retrofitting of existing operating units, has a clear division of labor among the three monitoring points, and the baseline parameters, local parameters, and overall parameters corroborate each other, resulting in high data reliability. It can specifically monitor the core area of ​​the heat dissipation surface, directly pinpoint the location of airflow dead zones, and accurately locate faults, solving the problems of arbitrary deployment of monitoring points A, unrepresentative data, and inability to accurately correspond to fault locations.

[0035] Furthermore, the altitude gradient correction adopts a piecewise linear compensation rule to address the characteristics of air density changes at high altitudes. The correction parameters conform to the GB / T 37921-2019 standard for high-altitude wind turbine generators. Specific correction rules include: Flow velocity gradient threshold correction: Based on the installation altitude, the preset flow velocity gradient reference threshold is corrected by altitude compensation. The flow velocity gradient reference threshold includes the first ratio reference K1 of the flow velocity of the heat dissipation surface to the flow velocity of the air inlet 101, and the second ratio reference K2 of the flow velocity of the exhaust air to the flow velocity of the heat dissipation surface.

[0036] Temperature gradient threshold correction: Based on the installation altitude, the preset temperature rise gradient reference threshold is corrected by altitude compensation. The temperature rise gradient reference threshold includes the first temperature rise reference ΔT1 for the heat dissipation surface temperature and the inlet air temperature, and the second temperature rise reference ΔT2 for the exhaust air temperature and the heat dissipation surface temperature.

[0037] The threshold values ​​of the flow around the air are not affected by the air density at altitude. A uniform preset value is used across the entire altitude range of the plateau. If the flow around the air coefficient exceeds the standard, it is directly determined that there is local airflow turbulence, and no correction is required.

[0038] The specific amendment rules are as follows: Flow rate ratio baseline correction: K1=K1×f(p), K2=K2×f(p).

[0039] Temperature rise baseline correction: ΔT1=ΔT1×g(p), ΔT2=ΔT2×g(p).

[0040] Where f(p) is the air density compensation coefficient, g(p) is the heat dissipation efficiency compensation coefficient, f(p) = p / p0, g(p) = p / p0, p is the air density at the current altitude, p0 is the air density at sea level (1.225 kg / m³), and the air density p is calculated according to the standard atmospheric formula: p = p0 × (1 - 0.0065H / 288)^4.255, where H is the altitude.

[0041] For example, when the altitude H=2500m, p / p0≈0.78, then f(p)=0.78, g(p)=1.28. If the plain baseline K1=0.85, K2=1.10, ΔT1=25℃, ΔT2=8℃, then the corrected values ​​are: K1=0.85×0.78=0.663, K2=1.10×0.78=0.858, ΔT1=25×1.28=32℃, ΔT2=8×1.28=10.2℃.

[0042] Furthermore, the flow coefficient λ is defined as a dimensionless parameter reflecting the uniformity of airflow distribution on the heat dissipation surface, and its calculation formula is: λ = (v max -v min ) / v avg vmax v represents the maximum value of the airflow velocity at multiple points on the front side of the heat dissipation surface. min For the minimum value, v avg The average value is obtained in real time by a miniature three-dimensional airflow sensor array (each column of power module 102 corresponds to 3 to 5 measuring points) deployed at the heat dissipation surface monitoring points. When λ exceeds the preset threshold λ_th=0.3, it is determined that a local airflow dead zone has formed on the heat dissipation surface.

[0043] Furthermore, the fault determination logic abandons the traditional single-point threshold exceedance determination mode and directly relies on pattern matching of airflow feature vectors, resulting in higher accuracy. Specifically, it includes: The system pre-stores the feature vector benchmarks corresponding to the three most common fault modes at high altitudes: fan bearing wear, dust / deformation of heat dissipation area, and poor connection of air intake channel. The feature vector benchmarks include three-dimensional combination features of flow velocity gradient correlation deviation, temperature rise gradient correlation deviation, and flow around characteristic parameters. Each fault corresponds to a unique feature vector interval.

[0044] The system collects airflow parameters from three segments in real time, constructs a three-dimensional feature vector of the current airflow state in real time through the edge computing unit, and calculates the Euclidean distance and cosine similarity between the real-time feature vector and the pre-stored feature vector benchmarks of each fault mode.

[0045] When the similarity exceeds the preset 85% matching threshold, the corresponding unique fault type is directly matched to eliminate interference factors.

[0046] Multi-dimensional feature vector matching achieves a fault identification accuracy of over 95%, directly pinpointing the fault type without requiring on-site investigation, significantly reducing the difficulty and time consumption of high-altitude operation and maintenance, and solving the problems of traditional fault judgment being singular, vague in location, and unable to distinguish the root cause of the fault.

[0047] The fault determination logic is implemented using pattern matching of airflow feature vectors. The airflow state feature vector X consists of 5 components: X=[δK1,δK2,δΔT1,δΔT2,λ]. The meaning and calculation method of each component are as follows: δK1=|K 1实测 -K 1修正 | / K 1修正 ×100%

[0048] δK2=|K 2实测 -K 2修正 | / K 2修正 ×100%

[0049] δΔT1=|ΔT 1实测 -ΔT 1修正 | / ΔT 1修正 ×100%

[0050] δΔT2=|ΔT 2实测 -ΔT2修正 | / ΔT 2修正 ×100%

[0051] λ = measured flow coefficient

[0052] Pre-store the feature vector reference X0 (sea level calibration value) for three typical fault modes:

[0053] Fan failure benchmark: On the test platform, fan bearing wear (running with sand for 100 hours) and speed decay (reduced power supply voltage) were artificially simulated. The mean of the feature vector 30 minutes before the failure was statistically analyzed in 12 tests. The 95% confidence interval of δK2 was [31.4%, 39.8%], so δK2≥30% was set as the threshold.

[0054] Heat dissipation surface blockage benchmark: Simulated dust layers of different thicknesses were sprayed onto the surface of the heat sink. When the hot spot temperature was 10°C higher than the average temperature, it was determined that an airflow dead zone was formed. At this time, λ≈0.35. In 10 tests, hot spots exceeding the standard were observed when λ≥0.3. Therefore, λ≥0.3 was set as the threshold.

[0055] The standard for poor air intake channel is: when the air intake louvers are partially blocked, the air velocity in all three sections decreases by more than 20% when the blocking rate is ≥50%. The threshold is set as δK1, δK2, δΔT1, and δΔT2 are all ≥20%.

[0056] Considering the influence of altitude, deviation calculations need to be performed in conjunction with the altitude-corrected baseline values ​​during actual use. For example, at an altitude of 2500m, the corrected physical quantity thresholds are as follows: K2≤0.858×(1~30%)=0.60 when the fan fails; λ≥0.3 when the heat dissipation surface is blocked; K1≤0.663×(1~20%)=0.53 when the air intake channel is obstructed; K2≤0.858×(1~20%)=0.69; ΔT1≥32×(1+20%)=38.4℃; ΔT2≥10.2×(1+20%)=12.2℃.

[0057] Real-time data acquisition constructs the current feature vector X, and the similarity between feature vector X and the baseline X0 for each fault mode is calculated. The similarity calculation employs a multi-dimensional fusion method: Calculate the Euclidean distance d, cosine similarity s, and Mahalanobis distance d of the feature vectors respectively.

[0058] The distance index is converted into similarity (e.g., s=1 / (1+d)) to generate a three-dimensional similarity evaluation matrix [s1,s2,s3].

[0059] The three-dimensional similarity was weighted and fused according to the weight coefficients [0.4, 0.3, 0.3] to obtain the comprehensive similarity S = 0.4×s1 + 0.3×s2 + 0.3×s3. The weight coefficients were determined based on the analysis of historical fault data using principal component analysis to maximize the distinguishability of the three fault modes.

[0060] When the overall similarity S exceeds the preset matching threshold (e.g., 85%), and the similarity difference between it and the fault mode with the highest similarity is greater than the set discrimination threshold (e.g., 10%), the fault type is finally matched.

[0061] Furthermore, based on big data from the operation and maintenance of high-altitude converters, standardized judgment rules were developed, as follows: The feature vector shows that the velocity gradient deviation in the exhaust section is significantly excessive, while the velocity and temperature parameters in the intake section and the flow characteristics of the heat dissipation surface are all within the normal range. The matching judgment is that the fan bearing is worn or the fan speed is reduced, which belongs to the mechanical failure of the fan body.

[0062] The eigenvectors show that the flow characteristics parameters around the heat dissipation surface are significantly exceeded, with a sudden drop in local flow velocity and a sudden rise in temperature. The flow velocity gradient between the air inlet and exhaust sections is basically normal. The matching judgment is that the heat dissipation area is blocked by dust or the heat dissipation fins are deformed and bent, forming a local airflow dead zone.

[0063] The eigenvectors show that the velocity gradients and temperature rise gradients of the air inlet section, heat dissipation section, and exhaust section deviate from the baseline, resulting in low overall heat exchange efficiency. The matching judgment indicates that the air inlet channel between the tower and the cabinet is not well connected and the air duct is leaking, which is a general air duct failure.

[0064] It can quickly identify fault types without the need for professional personnel, match the root cause of the fault in a targeted manner, avoid ineffective operation and maintenance, and shorten downtime.

[0065] Furthermore, to adapt to different wind field conditions and improve long-term diagnostic accuracy, a dynamic update step for the feature vector benchmark is also included: After each fault handling is completed, the unit restarts and operation stabilizes, and the fault is confirmed to be completely eliminated, the feature vector data within the 30 minutes prior to the fault occurrence is automatically retrieved as new samples. These samples are then updated using a weighted average algorithm based on the feature vector benchmark corresponding to the fault mode, with more recent fault samples receiving higher weights. The update formula is: X0 new =(1~β)×X0 old +β×X avg X0 new As the new eigenvector benchmark, X0 old The old feature vector is the baseline, β is the weight coefficient of recent samples, which is set to 0.2 (meaning that new samples contribute 20%), X avg This represents the mean of the feature vectors prior to the current fault.

[0066] When the cumulative number of samples for the same fault mode reaches a preset number N (e.g., N=50), the confidence interval (95% confidence level) of the feature vector baseline for that fault mode is recalculated using the formula: μ±1.96×σ / Where μ is the sample mean, σ is the standard deviation, and n is the number of samples, the matching threshold is adjusted according to the confidence interval width (for example, new threshold = original threshold × 1 + 0.1 × confidence interval width coefficient) to achieve adaptive optimization of the diagnostic model.

[0067] It enables adaptive iterative optimization of diagnostic models, becoming more accurate with use, and adapts to wind fields with different altitudes and dust environments. It eliminates the need for on-site re-adjustment, significantly improving versatility and durability, and solving problems such as rigid operation and maintenance models, decreased accuracy over long-term operation, and inability to adapt to differentiated wind fields.

[0068] Furthermore, the similarity between the real-time feature vector and the baseline feature vectors of each fault mode is calculated. To further improve diagnostic accuracy and avoid misjudgment of similar faults, this is further optimized into a multi-dimensional similarity fusion calculation. The Euclidean distance, cosine similarity, and Mahalanobis distance between the real-time feature vector and the baseline feature vector of each fault mode are calculated respectively to generate a three-dimensional similarity evaluation matrix.

[0069] The three-dimensional matrix is ​​weighted and fused according to a weight ratio of 4:3:3 to obtain the comprehensive similarity.

[0070] When the overall similarity exceeds the preset matching threshold of 85%, and the similarity difference between it and the fault mode with the highest similarity is greater than the set discrimination threshold of 10%, the fault type is finally matched to completely eliminate interference and avoid misjudgment.

[0071] Multi-dimensional weighted fusion similarity calculation further improves diagnostic accuracy, reducing the false positive rate to below 1%. The discriminant threshold setting completely avoids confusion between similar faults, making fault determination more reliable and solving the problems of feature confusion and high false positive rates associated with similar faults.

[0072] The weighting coefficients of Euclidean distance, cosine similarity, and Mahalanobis distance, in a ratio of 4:3:3, were determined using principal component analysis based on a large amount of field fault data, which can distinguish the three fault modes to the greatest extent possible.

[0073] Furthermore, the tiered operation and maintenance procedures fully align with the routine operation and maintenance processes of plateau wind farms, balancing equipment safety and power generation benefits, and are divided into three levels of handling: Level 1 (no deviation in airflow parameters, normal operating conditions): Regularly calibrate the monitoring sensors quarterly, clean the dustproof screen of the air intake channel and the floating dust on the heat dissipation surface, and use high-altitude special dry compressed air for cleaning to avoid secondary dust adhesion.

[0074] Level 2 (Airflow parameter gradient deviation reaches the preset warning range, slight abnormality): Do not shut down immediately. In conjunction with the planned shutdown period of the wind farm, add low-temperature resistant and high-altitude aging-resistant special grease to the fan bearing, correct the spacing of the heat sink of power module 102, and clean the local dust accumulation.

[0075] Level 3 (deviation exceeds threshold, fault condition): Implement tiered handling. If only the airflow dead zone on the heat dissipation surface exceeds the standard, remotely issue a load reduction command. After the unit operates at half load, clean the heat dissipation surface on-site; a complete shutdown is not required. In the event of fan failure or overall airflow failure, immediately trigger a safety shutdown to prevent equipment overheating and damage, ensuring equipment safety.

[0076] The three-tiered approach aligns with the realities of operation and maintenance in high-altitude wind farms, minimizing unplanned outages, maximizing power generation revenue, and balancing equipment safety with operational efficiency. This approach reduces the risks and workload for maintenance personnel working at heights, and addresses the issues of blindly shutting down for minor faults, delayed early warnings for serious faults, and the conflict between operation and maintenance and power generation.

[0077] Furthermore, it also includes a self-optimization step based on fault recording data to further enhance early warning capabilities: The converter has an automatic fault-triggered waveform recording function, which can record the original waveform data for 1 second before and after the fault. The sampling frequency is not less than 2kHz and there are no less than 32 analog channels. After the fault is triggered, the system automatically retrieves the original waveform data for 30 minutes before and after the fault (including the instantaneous values ​​of flow velocity and temperature at each monitoring point of air inlet 101, heat dissipation surface and exhaust outlet 103).

[0078] Wavelet transform was used to perform time-frequency domain analysis on the original waveform data: the db4 wavelet basis was selected, and a 5-level decomposition was performed. The low-frequency approximation coefficients were extracted to reflect the gradual trend, and the high-frequency detail coefficients reflected the abrupt change characteristics. The gradual change characteristics were defined as the slope of the linear fitting of the wavelet approximation coefficients within 15 to 30 minutes before the fault, and the abrupt change characteristics were defined as the peak value of the wavelet detail coefficients at the time of the fault.

[0079] When the cumulative samples of the same fault type reach 50 groups, the early warning threshold for that fault type is optimized based on the extracted gradual features. For example, if a slow upward trend of δK2 is generally observed before a fan failure, the warning threshold is adjusted from δK2≥30% to δK2≥25% or a continuous upward trend is detected, thus providing a warning 15-30 minutes in advance. At the same time, the matching weights of the feature vector benchmark are optimized based on abrupt change features to enhance the diagnostic model's sensitivity to abrupt change signals.

[0080] It enables early warning of faults, avoids the risk of equipment damage in advance, forms a closed-loop operation and maintenance system, maximizes the system's self-optimization capabilities, significantly reduces the operation and maintenance cost throughout the entire life cycle, and solves the problems of failing to capture early signs of faults and delayed early warnings.

[0081] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for operation and maintenance of a converter for a wind-cooled, high-altitude doubly-fed asynchronous wind turbine, characterized in that: Includes the following steps: Based on the airflow path of the converter cabinet (100), monitoring points (A) are set up on the converter cabinet (100) to collect the airflow parameters of the air inlet (101), the local airflow parameters of the heat dissipation surface of the power module (102), and the airflow parameters of the exhaust outlet (103) in real time. Based on the altitude of the converter cabinet (100) installation site, the airflow parameter judgment threshold is corrected by altitude gradient. Calculate the velocity gradient deviation and temperature gradient deviation of the airflow in the three sections of air intake section, heat dissipation section and exhaust section, and simultaneously monitor the flow coefficient around the heat dissipation surface of the power module (102) to determine whether an airflow dead zone is formed inside the cabinet and the overall operating status of the heat dissipation system. Based on the deviation level and the corresponding fault type, perform graded operation and maintenance actions.

2. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in claim 1, characterized in that: The converter cabinet (100) adopts a bottom air intake, a middle power module (102) arrangement and a top exhaust fan layout, forming an air intake airflow path from bottom to top. There are no additional guide baffles inside the converter cabinet (100), forming a vertical air intake airflow path from bottom to top. After the cold air enters from the bottom, it flows through the heat sink of the middle power module (102) and carries away the heat generated by the operation of the power module (102), and is finally forced out by the top fan.

3. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in claim 1 or 2, characterized in that: The monitoring points (A) are all arranged based on the structure of the converter cabinet (100), and the specific locations are as follows: The bottom air inlet (101) monitoring point is set at the bottom air inlet channel of the cabinet or at the louver on the bottom side of the converter cabinet (100) to collect airflow velocity and ambient temperature as the benchmark parameters for airflow attenuation at high altitude. The heat dissipation monitoring point of the power module (102) is located on the windward side of the heat sink of the power module (102) in the middle of the converter cabinet (100) to collect local airflow velocity, temperature and flow coefficient, and to determine the formation state of the airflow dead zone; The top exhaust outlet (103) monitoring point is located at the exhaust fan outlet and / or top louver of the cabinet, and is used to collect exhaust air velocity and exhaust air temperature to determine the fan extraction efficiency and overall airflow circulation effect.

4. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in claim 3, characterized in that: The elevation gradient correction rule includes: Flow velocity gradient threshold correction: Based on the installation altitude, the preset flow velocity gradient reference threshold is corrected by altitude compensation. The flow velocity gradient reference threshold includes a first ratio reference of the flow velocity of the heat dissipation surface to the flow velocity of the air inlet (101), and a second ratio reference of the flow velocity of the exhaust air to the flow velocity of the heat dissipation surface. Temperature gradient threshold correction: Based on the installation altitude, the preset temperature rise gradient reference threshold is corrected by altitude compensation. The temperature rise gradient reference threshold includes a first temperature rise reference of heat dissipation surface temperature and air inlet temperature, and a second temperature rise reference of exhaust air temperature and heat dissipation surface temperature. The threshold values ​​for flow-around characteristic parameters are uniformly preset across the entire altitude range of the plateau.

5. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in claim 4, characterized in that: The fault determination logic is directly based on pattern matching of airflow feature vectors, specifically including: Pre-store feature vector references corresponding to at least three fault modes, wherein the feature vector references include a combination of features of velocity gradient correlation deviation, temperature rise gradient correlation deviation and flow around feature parameters; The feature vector of the current airflow state is collected and constructed in real time, and the similarity between the real-time feature vector and the benchmark feature vector of each fault mode is calculated. When the similarity exceeds the preset matching threshold, the corresponding fault type is matched.

6. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in claim 5, characterized in that: The fault types corresponding to the matching include: The feature vector shows a significant deviation in the velocity gradient of the exhaust section, while the parameters of the intake section and the flow characteristics around the heat dissipation surface are relatively normal. The matching judgment is that the fan bearing is worn or the speed is reduced. The feature vector shows that the flow characteristic parameters around the heat dissipation surface are significantly exceeded, while the flow velocity gradients in the air inlet and exhaust sections are basically normal. The matching judgment is that the heat dissipation area is dusty or the heat sink is deformed. The feature vectors show that the velocity gradients and temperature rise gradients of the air inlet section, heat dissipation surface section, and exhaust section are all deviated, and the matching judgment is that the air inlet channel between the tower and the cabinet is not well connected.

7. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in claim 6, characterized in that: It also includes a dynamic update step for the eigenvector baseline: After each fault handling is completed and the fault is confirmed to be eliminated, the feature vector data before the occurrence of the fault is used as new samples, and the feature vector benchmark corresponding to the fault mode is updated with weights. When the cumulative number of samples for the same fault mode reaches a preset number, the confidence interval of the feature vector benchmark for that fault mode is recalculated, and the preset matching threshold is optimized accordingly.

8. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in claim 7, characterized in that: The calculation of the similarity between the real-time feature vector and the baseline feature vectors of each fault mode further includes: The Euclidean distance, cosine similarity, and Mahalanobis distance between the real-time feature vector and the baseline feature vector of each fault mode are calculated respectively to generate a multi-dimensional similarity evaluation matrix. The multidimensional similarity evaluation matrix is ​​weighted and fused to obtain a comprehensive similarity score; When the overall similarity exceeds the preset matching threshold, and the similarity difference between the fault mode with the highest similarity is greater than the set discrimination threshold, the fault mode is finally matched.

9. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in any one of claims 1 to 8, characterized in that: The tiered operation and maintenance actions are consistent with the routine operation and maintenance procedures of wind farms, including: When the airflow parameters are within tolerance, the monitoring sensors are calibrated quarterly, and the dust in the air intake channel and heat dissipation surface is cleaned. When the gradient deviation of the airflow parameters reaches the preset warning range, in conjunction with the planned shutdown period of the wind farm, add plateau-specific grease to the fan bearing and correct the spacing of the heat sink of the power module (102); When the deviation exceeds the threshold, a graded approach is taken. If only the airflow dead zone exceeds the standard, the heat dissipation surface is cleaned on-site after remote load reduction. If a fan failure or overall airflow failure occurs, the machine is immediately shut down for handling.

10. The operation and maintenance method for the converter of a wind-cooled, high-altitude doubly-fed asynchronous wind turbine as described in any one of claims 5 to 8, characterized in that: It also includes a self-optimization step based on fault recording data: After the fault determination is triggered, the original waveform data of the preset time before and after the fault recorded inside the converter is retrieved. The original waveform data includes the instantaneous value sequence of the airflow parameters of the air inlet (101), the local airflow parameters of the heat dissipation surface of the power module (102), and the airflow parameters of the exhaust outlet (103). The original waveform data is compared with the real-time feature vector in the time-frequency domain to extract the gradual change features of airflow parameters before the fault occurs and the abrupt change features when the fault occurs. When the cumulative number of samples of the same fault type reaches a preset number, the early warning threshold of the fault type is optimized based on the extracted gradual features, and the matching weight of the feature vector benchmark is optimized based on the mutation features, so as to achieve adaptive optimization of the fault diagnosis model.

Images