Temperature early warning method and device for fan grid-connected contactor cabinet and medium
By using a BP neural network model to predict the temperature and perform residual analysis on the wind turbine grid-connected contactor cabinet, the problem of real-time and accuracy of temperature monitoring of the wind turbine grid-connected contactor cabinet was solved, enabling early and accurate warning of overheating faults and ensuring the safety of wind farms.
Patent Information
- Application Number
- CN202511710471.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-20
- Publication Date
- 2026-03-03
AI Technical Summary
Existing technologies cannot achieve real-time and accurate temperature monitoring of the contactor cabinet for wind turbine grid connection, leading to false or missed overheating faults, which seriously threatens the safe and stable operation of wind farms.
An independent temperature prediction model is established using a BP neural network. By combining the residual between the real-time temperature and the predicted temperature, an early warning threshold is dynamically set to achieve early and accurate detection of overheating faults.
By using a BP neural network model, the impact of fluctuations in normal operating conditions is reduced, enabling early and accurate warnings of overheating faults, avoiding false alarms and missed alarms, and ensuring the safe and stable operation of wind farms.
Smart Images
Figure CN121595055A_ABST
Abstract
Description
Technical Field
[0001] This document relates to the field of temperature early warning technology for wind turbine grid-connected contactor cabinets, and in particular to a method, equipment and medium for temperature early warning of wind turbine grid-connected contactor cabinets. Background Technology
[0002] The contactor cabinet for wind turbine grid connection is a key piece of equipment for power output in a wind farm. Its internal circuit breaker contacts, busbar connection points, and other critical components are prone to abnormal overheating due to the continuous flow of high currents, caused by factors such as increased contact resistance, loose bolts, and aging insulation. Overheating is a continuous process; if not addressed promptly, it can accelerate the deterioration of insulation materials and even lead to single-phase or phase-to-phase short circuits, equipment burnout, or even fires and explosions, seriously threatening the safe and stable operation of the wind farm.
[0003] Currently, temperature monitoring of contactor cabinets mainly relies on manual inspections, using methods such as temperature-testing wax strips and infrared thermometers. However, these methods have significant limitations: temperature-testing wax strips require close-range manual inspection, resulting in severe response delays and posing a safety hazard of melting and dripping; infrared thermometry is easily affected by cabinet structure obstructions, ambient light, and electromagnetic interference, making it difficult to guarantee measurement accuracy, and it cannot achieve real-time monitoring of exposed high-voltage points within enclosed cabinets. Furthermore, some existing online temperature measurement technologies, such as battery-powered wireless temperature measurement, suffer from problems such as short battery life at high temperatures and high insulation risks; fiber optic temperature measurement has drawbacks such as complex installation, susceptibility to contamination, and potential impact on cabinet insulation performance.
[0004] More importantly, the operating temperature of the contactor cabinet is strongly correlated with the current load, ambient temperature, and other operating conditions. Traditional fixed threshold early warning mechanisms cannot distinguish between normal temperature rise under high load and overheating caused by equipment abnormalities, resulting in low early warning accuracy and a high risk of false alarms or missed alarms. Therefore, there is an urgent need in this field for a temperature monitoring method that can adapt to the real-time operating conditions of the equipment and achieve accurate early warning. Summary of the Invention
[0005] This invention provides a method, device, and medium for early warning of temperature in a wind turbine grid-connected contactor cabinet, aiming to solve the above-mentioned problems.
[0006] According to an embodiment of the present invention, a method for early warning of temperature in a wind turbine grid-connected contactor cabinet is provided, comprising: S1. Real-time acquisition of temperature data from key temperature measurement points inside the fan grid-connected contactor cabinet, as well as corresponding operating load data and environmental parameter data, to construct a multi-source time-series database; S2. Based on the multi-source time-series data under historical normal conditions, establish an independent BP neural network temperature prediction model for each key temperature measurement point; wherein, the input of each model is preprocessed operating load data and environmental parameter data, and the output is the predicted temperature of the corresponding temperature measurement point; S3. Based on the residual between the real-time temperature of each temperature measurement point and the predicted temperature obtained by the BP neural network temperature prediction model, early overheating warning is given.
[0007] According to an embodiment of the present invention, an electronic device is provided, comprising: Processor; and, A memory is configured to store computer-executable instructions, which, when executed, cause the processor to perform steps such as the above-described method for early warning of temperature in a fan grid-connected contactor cabinet.
[0008] According to an embodiment of the present invention, a storage medium is provided for storing computer-executable instructions, which, when executed, implement the above-described method for early warning of temperature in a fan grid-connected contactor cabinet.
[0009] This invention introduces a BP neural network to establish an independent temperature prediction model for each temperature measurement point, enabling accurate prediction of the equipment's expected temperature under current load and environmental parameters. By analyzing the residual between real-time and predicted temperatures, the impact of fluctuations in normal operating conditions is reduced, allowing the early warning mechanism to focus on temperature changes caused by equipment anomalies. This achieves early and accurate detection of overheating faults, overcoming the drawbacks of fixed threshold methods that are prone to false alarms and missed alarms. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a flowchart of the wind turbine grid-connected contactor cabinet temperature early warning method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a BP neural network according to an embodiment of the present invention. Detailed Implementation
[0012] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.
[0013] Method Implementation Examples According to an embodiment of the present invention, a method for early warning of temperature in a fan grid-connected contactor cabinet is provided. Figure 1 This is a flowchart of a wind turbine grid-connected contactor cabinet temperature early warning method according to an embodiment of the present invention. Figure 1 As shown, the wind turbine grid-connected contactor cabinet temperature early warning method of this embodiment of the invention specifically includes: S1. Real-time acquisition of temperature data from key temperature measurement points inside the fan grid-connected contactor cabinet, as well as corresponding operating load data and environmental parameter data, to construct a multi-source time-series database; In this embodiment of the invention, three types of key temperature measurement points are set: main circuit conductive connection points, key environmental monitoring points, and special component monitoring points.
[0014] The main circuit conductive connection points are prone to overheating faults because the contact resistance is easily changed due to the long-term passage of large currents. Specifically, these include: the upper and lower contacts or plugs of the circuit breaker, the crimping points of the busbar connection bolts, the main contacts and inlet and outlet terminals of the contactor, and the transition parts of the flexible connection.
[0015] Key environmental monitoring points inside the cabinet are used to capture the overall thermal environment inside the cabinet, providing background temperature references for temperature prediction models. These points specifically include: the upper space of the cabinet, and the heat dissipation channels or vents at the back of the cabinet. Ambient temperature sensors are installed in locations inside the cabinet that are not directly exposed to heat sources and where airflow is relatively stable.
[0016] Special component monitoring points are set up for components known to be prone to overheating or with special structures, such as the surface of heat-generating elements like reactors and resistors; and the area near insulating supports or partitions, to monitor the risk of insulation degradation caused by local overheating.
[0017] The aforementioned data acquisition is achieved through a sensor network deployed within the cabinet. Temperature sensors are preferably high-temperature resistant, electromagnetic interference-resistant digital sensors or platinum resistance thermometers. Operating load data is acquired through existing current transformers within the cabinet. Environmental parameters are acquired by temperature and humidity sensors deployed at designated locations within the cabinet. All sensor data is transmitted to the data acquisition unit via fieldbus or industrial Ethernet.
[0018] After data acquisition, the raw data undergoes preliminary preprocessing, including invalid value removal, digital filtering, and data alignment. All data is then appended with a uniform timestamp and stored in a multi-source time-series database built upon a time-series database, forming a dataset for model training and real-time analysis.
[0019] S2. Based on the multi-source time-series data under historical normal conditions, establish an independent BP neural network temperature prediction model for each key temperature measurement point; wherein, the input of each model is preprocessed operating load data and environmental parameter data, and the output is the predicted temperature of the corresponding temperature measurement point; The establishment of an independent BP neural network temperature prediction model for each key temperature measurement point specifically includes: Based on the thermoelectric coupling time constant of the physical object being monitored by the temperature measurement point, a differentiated BP neural network structure is configured for it; For the temperature measurement points of the main circuit conductive components, the corresponding neural network model adopts the first network structure; For the temperature measurement points inside the monitoring cabinet, the corresponding neural network model adopts the second network structure; The hidden layer complexity of the first network structure is higher than that of the second network structure, and the training convergence speed of the first network structure is faster than that of the second network structure.
[0020] Furthermore, the total number of hidden layer neurons in the first network structure is greater than the total number of hidden layer neurons in the second network structure, and the learning rate of the first network structure is greater than the learning rate of the second network structure.
[0021] More specifically, in this embodiment of the invention, the first network structure is applied to the temperature measurement points of the conductive components in the main circuit. Because the temperature at these points changes rapidly and is closely related to the current, the model requires stronger nonlinear fitting capabilities. Therefore, the first network structure employs relatively complex hidden layers, for example, containing two hidden layers with 16 and 8 neurons respectively, and uses the ReLU activation function to accelerate convergence.
[0022] The second network structure is used to monitor the temperature at points within the cabinet. Since the temperature changes at these points are relatively gradual, the model complexity can be appropriately reduced. Therefore, the second network structure uses one hidden layer containing eight neurons, and the activation function is also ReLU.
[0023] During the model training phase, historical normal operating data of the equipment was selected from a multi-source time-series database as the training set. After standardizing the training data, the backpropagation algorithm and the Adam optimizer were used for training, with mean squared error as the loss function. Early stopping was employed during training to prevent overfitting.
[0024] S3. Based on the residual between the real-time temperature of each temperature measurement point and the predicted temperature obtained by the BP neural network temperature prediction model, early overheating warning is given.
[0025] S3 specifically includes: The real-time collected operating load data and environmental parameter data are input into the BP neural network temperature prediction model to obtain the predicted temperature of each temperature measurement point. The residual between the real-time temperature and the predicted temperature of each temperature measurement point is calculated. Based on the residual and combined with the rated parameters of the equipment, an early warning threshold is dynamically set. When the residual of any temperature measurement point continues to exceed its early warning threshold, an early overheat warning signal for that temperature measurement point is triggered. Figure 2This is a schematic diagram of a BP neural network according to an embodiment of the present invention.
[0026] Furthermore, the operating load data includes three-phase current. Before inputting the three-phase current data into the BP neural network model, its current imbalance is calculated. The current imbalance is input into the model along with the effective value of the three-phase current, so that the model can identify the local overheating effect caused by the inter-phase load imbalance.
[0027] The current imbalance is calculated as the ratio of the negative sequence component to the positive sequence component of the current, or characterized as the percentage of the maximum deviation of the effective value of the three-phase current to the average value.
[0028] The dynamically set early warning threshold specifically includes: The residual sequence is segmented using a sliding time window; For the residual data within each time window, the peak threshold method is used to filter out the sequence of extreme points that exceed the threshold value; The extreme point sequence is fitted using a generalized Pareto distribution, and the conditional quantile at a given confidence level is calculated. This conditional quantile is then used as the dynamic warning threshold for the current time window.
[0029] The threshold value is adaptively adjusted based on the statistical characteristics of the residuals within the sliding window, specifically as follows: ; Where μ is the mean of the residuals within the window, σ is the standard deviation, and k is the sensitivity coefficient set according to the safety operation requirements of the contactor cabinet.
[0030] The specific conditions for triggering the early overheat warning signal for this temperature measurement point are as follows: Set an early warning time window T and a counting threshold N; If, within a continuous time period T, the number of times the residual at a certain temperature measurement point exceeds its dynamic warning threshold reaches or exceeds N times, it is determined that there is a risk of continuous overheating at that point, triggering a warning.
[0031] More specifically, a concrete implementation of step S3 is as follows: Generating the residual sequence specifically includes: for each temperature measurement point i, at each sampling time t, calculating the measured temperature. Temperature prediction using a BP neural network model The residual: ; Perform this operation in parallel at all temperature measurement points to generate residual time series for each point. ; The calculation of dynamic early warning thresholds specifically includes: A sliding time window of fixed length L is used to segment the continuous residual sequence. The window slides as new data arrives, each time sliding by one sampling interval. For the residual data within the current window, its mean and standard deviation are calculated, and the threshold value is determined according to the following formula: ; Using the peak thresholding method, all residual values greater than the threshold value within the current window are selected to form a sequence of extreme points.
[0032] Assume these extreme point sequences follow a generalized Pareto distribution. Use maximum likelihood estimation to fit the shape parameters of the GPD distribution. and scale parameters The formula is: ; Where m is the number of extreme points. This represents the total number of data points in the current window.
[0033] This will be used as the dynamic warning threshold for this temperature measurement point within the current time window.
[0034] The early warning triggering logic includes: setting an early warning time window T and a counting threshold N; and maintaining a counter Count(i) for each temperature measurement point. Within any consecutive time period T, whenever the real-time residual of temperature measurement point i exceeds its current dynamic early warning threshold, the counter Count(i) increments by 1. If, within a consecutive time period T, the cumulative counter Count(i) of a certain temperature measurement point i reaches or exceeds the set counting threshold N, it is determined that the temperature measurement point has a persistent overheating risk, and the system immediately triggers an early overheating early warning signal for that point.
[0035] By combining the local statistical characteristics of residual sequences with extreme value theory, this algorithm can adaptively set a reasonable and sensitive early warning threshold. Through continuous judgment logic, it can effectively distinguish between short-term random fluctuations and real, continuously developing overheating fault precursors, thereby achieving early and accurate overheating early warning while ensuring a low false alarm rate.
[0036] This invention introduces a BP neural network to establish an independent temperature prediction model for each temperature measurement point, enabling accurate prediction of the equipment's expected temperature under current load and environmental parameters. By analyzing the residual between real-time and predicted temperatures, the impact of fluctuations in normal operating conditions is reduced, allowing the early warning mechanism to focus on temperature changes caused by equipment anomalies. This achieves early and accurate detection of overheating faults, overcoming the drawbacks of fixed threshold methods that are prone to false alarms and missed alarms.
[0037] Device Example 1 According to an embodiment of the present invention, an electronic device is provided, comprising: Processor; and, A memory is configured to store computer-executable instructions, which, when executed, cause the processor to perform the steps of the above-described method for temperature warning of the wind turbine grid-connected contactor cabinet.
[0038] Device Example 2 According to an embodiment of the present invention, a storage medium is provided for storing computer-executable instructions, which, when executed, implement the steps of the above-described method for temperature early warning of the wind turbine grid-connected contactor cabinet.
[0039] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for early warning of temperature in a fan grid-connected contactor cabinet, characterized in that... include: S1. Real-time acquisition of temperature data from key temperature measurement points inside the fan grid-connected contactor cabinet, as well as corresponding operating load data and environmental parameter data, to construct a multi-source time-series database; S2. Based on the multi-source time-series data under historical normal conditions, establish an independent BP neural network temperature prediction model for each key temperature measurement point; wherein, the input of each model is preprocessed operating load data and environmental parameter data, and the output is the predicted temperature of the corresponding temperature measurement point; S3. Based on the residual between the real-time temperature of each temperature measurement point and the predicted temperature obtained by the BP neural network temperature prediction model, early overheating warning is given.
2. The method according to claim 1, characterized in that, The establishment of an independent BP neural network temperature prediction model for each key temperature measurement point specifically includes: Based on the thermoelectric coupling time constant of the physical object being monitored by the temperature measurement point, a differentiated BP neural network structure is configured for it; For the temperature measurement points of the main circuit conductive components, the corresponding neural network model adopts the first network structure; For the temperature measurement points inside the monitoring cabinet, the corresponding neural network model adopts the second network structure; The hidden layer complexity of the first network structure is higher than that of the second network structure, and the training convergence speed of the first network structure is faster than that of the second network structure.
3. The method according to claim 2, characterized in that, The total number of hidden layer neurons in the first network structure is greater than the total number of hidden layer neurons in the second network structure, and the learning rate of the first network structure is greater than the learning rate of the second network structure.
4. The method according to claim 1, characterized in that, S3 specifically includes: The real-time collected operating load data and environmental parameter data are input into the BP neural network temperature prediction model to obtain the predicted temperature of each temperature measurement point. The residual between the real-time temperature and the predicted temperature of each temperature measurement point is calculated. Based on the residual and combined with the rated parameters of the equipment, an early warning threshold is dynamically set. When the residual of any temperature measurement point continues to exceed its early warning threshold, an early overheat warning signal for that temperature measurement point is triggered.
5. The method according to claim 1, characterized in that, The operating load data includes three-phase current. Before inputting the three-phase current data into the BP neural network model, its current imbalance is calculated. The current imbalance is then input into the model along with the effective value of the three-phase current, so that the model can identify the local overheating effect caused by the inter-phase load imbalance.
6. The method according to claim 4, characterized in that, The dynamically set early warning threshold specifically includes: The residual sequence is segmented using a sliding time window; For the residual data within each time window, the peak threshold method is used to filter out the sequence of extreme points that exceed the threshold value; The extreme point sequence is fitted using a generalized Pareto distribution, and the conditional quantile at a given confidence level is calculated. This conditional quantile is then used as the dynamic warning threshold for the current time window.
7. The method according to claim 6, characterized in that, The threshold value is adaptively adjusted based on the statistical characteristics of the residuals within the sliding window, specifically as follows: ； Where μ is the mean of the residuals within the window, σ is the standard deviation, and k is the sensitivity coefficient set according to the safety operation requirements of the contactor cabinet.
8. The method according to claim 1, characterized in that, The specific conditions for triggering the early overheat warning signal for this temperature measurement point are as follows: Set an early warning time window T and a counting threshold N; If, within a continuous time period T, the number of times the residual at a certain temperature measurement point exceeds its dynamic warning threshold reaches or exceeds N times, it is determined that there is a risk of continuous overheating at that point, triggering a warning.
9. An electronic device, comprising: processor; as well as, A memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the steps of the wind turbine grid-connected contactor cabinet temperature early warning method as described in any one of claims 1-8.
10. A storage medium for storing computer-executable instructions, which, when executed, implement the wind turbine grid-connected contactor cabinet temperature early warning method as described in any one of claims 1-8.