A wind turbine cabin intelligent power distribution pyrolysis particle sensing and early warning system

Abstract

The present invention discloses a wind turbine generator cabin intelligent power distribution pyrolytic particle sensing and early warning system, which relates to the technical field of wind power generation equipment safety monitoring technology. The system includes: an integrated particle sampling module, embedded and installed at the inlet of the forced air cooling duct of the power distribution cabinet; a multi-spectral laser analysis module, connected to the integrated particle sampling module, using a dual-wavelength laser to identify particle types and calculate concentration; a power distribution parameter coupling module, which synchronously calculates the particle concentration growth rate, cable temperature rise rate, and insulation resistance drop rate; a risk prediction module, which integrates multi-dimensional input features and outputs fire probability through an LSTM model; and a graded early warning module, which executes graded response measures in a linked manner according to probability values. The present invention has the advantages of real-time capture of submicron pyrolytic particles, realization of multi-parameter dynamic coupling early warning, reduction of false alarm rate, and improvement of fault response timeliness.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation equipment safety monitoring, and in particular to an intelligent power distribution pyrolysis particle sensing and early warning system for a wind turbine generator cabin. Background Art

[0002] As the capacity of wind turbines exceeds 10MW, the integration of electrical equipment within the nacelle has significantly increased, posing a significant challenge to accurately warning of early thermal failures in the power distribution system. Traditional monitoring solutions face technical limitations in three dimensions: at the perception level, conventional smoke detectors are limited by a particle size detection limit of 1μm or larger, making them unable to capture the 0.1-0.5μm characteristic particles released during the initial pyrolysis of insulating materials. At the analysis level, threshold alarm mechanisms based on a single temperature or current parameter make it difficult to distinguish between normal equipment operating fluctuations and true insulation degradation signals. At the decision-making level, fixed threshold response modes cannot adapt to the dynamic changes in risks under varying wind speeds and load conditions. Of particular note, the high humidity and high salt fog characteristics of the offshore wind power environment result in a false alarm rate of over 30% for traditional photoelectric sensors. While the multi-sensor fusion solution proposed in existing patent document CN112987711A incorporates temperature rise rate analysis, it still fails to address the challenge of identifying the submicron pyrolysis particle type. While the laser scattering technology disclosed in CN113551654B improves particle size detection sensitivity, it lacks a collaborative verification mechanism with electrical parameters and fails to establish a mapping relationship between particle spatial distribution and insulation degradation. More critically, traditional systems generally employ a linear "monitoring-alarm" architecture, which is unable to predict the evolution of thermal faults or dynamically adjust response strategies based on risk probabilities. This leaves operations and maintenance personnel facing a dilemma between excessive downtime and delayed response. Furthermore, the complex airflow environment within the cabin makes locating the pyrolysis source difficult. Existing technologies typically require manual point-by-point investigation, severely impacting the timeliness of fault resolution. Summary of the Invention

[0003] In view of this, the present invention provides a wind turbine cabin intelligent power distribution pyrolysis particle perception and warning system, which has the advantages of real-time capture of submicron pyrolysis particles, realization of multi-parameter dynamic coupling warning, reduction of false alarm rate and improvement of fault response timeliness.

[0004] The present invention provides a wind turbine cabin intelligent power distribution pyrolysis particle sensing and early warning system, comprising:

[0005] The integrated particle sampling module is embedded in the forced air cooling duct inlet of the power distribution cabinet to directly capture submicron pyrolysis particles in the airflow;

[0006] A multispectral laser analysis module, connected to the integrated particle sampling module, is used to irradiate captured submicron pyrolytic particles with laser beams at wavelengths of 532nm and 1064nm, identify the type of pyrolytic particles based on the ratio of the scattered light intensity at 532nm to 1064nm, and calculate the concentration of each type of pyrolytic particles;

[0007] The power distribution parameter coupling module is connected to the multi-spectral laser analysis module to synchronously collect the three-phase current of the current transformer, the cable joint temperature of the temperature sensor, and the insulation resistance of the insulation tester. It also calculates the particle concentration growth rate, cable temperature rise rate, and insulation resistance decrease rate in real time. It also determines whether to trigger a secondary warning based on the particle concentration growth rate, cable temperature rise rate, and insulation resistance decrease rate.

[0008] A risk prediction module, connected to the multispectral laser analysis module and the power distribution parameter coupling module, is configured to obtain multidimensional input features, input the multidimensional input features into a pre-trained LSTM model, and output a fire probability value within a preset future time period; the multidimensional input features are obtained based on the three-phase current of the current transformer, the concentration of each type of pyrolytic particles, and a historical fault library;

[0009] The hierarchical warning module is connected to the risk prediction module and is used to link the fire extinguishing, ventilation and power grid tripping devices to execute hierarchical response measures according to the fire probability value.

[0010] In an optional embodiment, the integrated particle sampling module includes an electrostatic enrichment ring, a vortex deceleration chamber, and a ceramic filter membrane assembly that are sequentially sealed and connected along the airflow direction;

[0011] The outer wall of the electrostatic enrichment ring is provided with a bias electrode lead, which is electrically connected to an external high-voltage power supply through a sealed cable and is used to transmit a preset bias voltage applied by the high-voltage power supply to the electrodes of the electrostatic enrichment ring, so as to form a non-uniform electric field between the electrodes of the electrostatic enrichment ring; the electrostatic enrichment ring is used to adsorb charged particles in the airflow under the non-uniform electric field;

[0012] The vortex deceleration chamber is provided with a vortex guide plate inside, which is used to convert the straight airflow into a rotating flow field to reduce the speed of the airflow;

[0013] The ceramic filter membrane assembly comprises multiple layers of stacked ceramic filter membranes, which are used for multi-level gradient filtration of interfering particles in the airflow and outputting purified submicron pyrolysis particles.

[0014] In an optional embodiment, the multispectral laser analysis module includes:

[0015] Dual-wavelength laser emission unit, used for synchronously emitting laser beams with wavelengths of 532nm and 1064nm respectively;

[0016] an optical gas chamber, used to guide an air flow containing submicron-sized pyrolytic particles through an intersection area of ​​two laser beams so that the submicron-sized particles are irradiated by the two laser beams;

[0017] A scattered light collection unit is used to capture the side scattered light generated by submicron pyrolytic particles after being irradiated by two laser beams; and a filter wheel is used to separate the scattered light into a pure 532nm scattered light signal and a pure 1064nm scattered light signal;

[0018] a signal processing unit connected to the scattered light collection unit, configured to convert the 532nm scattered light signal and the 1064nm scattered light signal into corresponding electrical pulse signals and perform three-stage filtering;

[0019] The particle identification unit is connected to the signal processing unit and is used to calculate the scattering intensity ratio based on the filtered 532nm electric pulse signal and the 1064nm electric pulse signal; and match the scattering intensity ratio with a preset pyrolysis particle type database to output the type of pyrolysis particles and the concentration of each type of pyrolysis particles.

[0020] In an optional implementation, the power distribution parameter coupling module includes:

[0021] Synchronous acquisition unit, used to synchronously acquire the three-phase current of the current transformer, the cable joint temperature of the temperature sensor, and the insulation resistance of the insulation tester;

[0022] a dynamic calculation unit, connected to the synchronous acquisition unit and the multi-spectral laser analysis module, respectively, for calculating the particle concentration growth rate and the cable joint temperature rise rate based on the cable joint temperature and insulation resistance; and for calculating the particle concentration growth rate based on the concentration of each type of pyrolytic particles;

[0023] The early warning trigger unit is used to activate the secondary early warning when the particle concentration growth rate, cable joint temperature rise rate, and insulation resistance drop rate simultaneously meet the early warning trigger conditions.

[0024] In an optional implementation, the warning triggering condition of the power distribution parameter coupling module is:

[0025] Particle concentration growth rate>10 4 particles / (m 3 min), and

[0026] The cable connector temperature rise rate is >5℃ / min, and

[0027] Insulation resistance drop rate>10% / min.

[0028] In an optional embodiment, the hierarchical response measures adopt a progressive response, including:

[0029] When the fire probability is ≥30%, start the negative pressure exhaust in the cabin;

[0030] When the fire probability is ≥60%, cut off non-critical loads;

[0031] When the fire probability is ≥90%, the perfluorohexanone fire extinguishing agent is triggered and the network is disconnected.

[0032] In an optional embodiment, the system also includes: a fault tracing module, connected to the graded warning module, for constructing a three-dimensional airflow velocity model through 6 ultrasonic anemometers, and calculating the backtracking path of the pyrolysis particles based on the three-dimensional airflow velocity model to locate the pyrolysis source coordinates.

[0033] In an optional embodiment, the signal processing unit performs three-stage filtering on the electrical pulse signal, including:

[0034] First-stage high-pass filtering with a cutoff frequency of 10kHz to eliminate ambient noise;

[0035] Secondary band-pass filtering with a center frequency of 1 MHz to extract particle characteristic signals;

[0036] Three-stage adaptive filtering to suppress electromagnetic interference pulses.

[0037] In an optional embodiment, the multi-dimensional input features include:

[0038] The spatiotemporal distribution matrix of particle concentration, the dynamic change rate of three-phase current imbalance and the matching insulation degradation pattern encoding in the historical fault library.

[0039] The present invention has the following beneficial effects:

[0040] The present invention captures submicron pyrolytic particles through an integrated particle sampling module, combines multi-spectral laser analysis to realize particle type identification, and couples electrical parameters to dynamically calculate risk probability. It has the advantages of real-time capture of submicron pyrolytic particles, realization of multi-parameter dynamic coupling early warning, reduction of false alarm rate and improvement of fault response timeliness. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 12 is a schematic structural diagram of a wind turbine cabin intelligent power distribution pyrolysis particle sensing and warning system according to an embodiment of the present invention;

[0043] Figure 2 is a schematic structural diagram of an integrated particle sampling module according to an embodiment of the present invention;

[0044] Reference numerals:

[0045] 1. Electrostatic enrichment ring; 11. Bias electrode lead; 2. Vortex deceleration chamber; 21. Vortex guide plate; 3. Ceramic filter membrane assembly. DETAILED DESCRIPTION

[0046] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0047] Fire monitoring in wind turbine generator compartment power distribution systems has long faced technical bottlenecks. Traditional methods rely on single-parameter monitoring, failing to effectively capture early-stage pyrolysis particles and suffering from high false alarm rates and delayed responses. When the insulation material inside the distribution cabinet undergoes pyrolysis, the submicron particles initially generated diffuse through the forced air cooling system. Existing sensors struggle to effectively capture these particles due to improper installation locations and insufficient detection accuracy. Furthermore, isolated monitoring of parameters like temperature and current prevents the establishment of multi-dimensional correlation models, resulting in weak risk quantification capabilities.

[0048] To address these issues, researchers discovered that traditional particle detection devices suffer from airflow disturbances that affect sampling efficiency. This led them to propose a design that embeds the sampling module at the air duct entrance. By analyzing the spectral characteristics of pyrolysis particles from different materials, they discovered that the dual-wavelength scattering intensity ratio has potential for type identification. To address the issue of isolated parameter analysis, a dynamic coupling model of particle concentration growth rate and electrical parameters was developed. To achieve accurate predictions, a spatiotemporal distribution matrix was introduced as input features for an LSTM model, establishing a probabilistic early warning mechanism.

[0049] Therefore, the present invention proposes an intelligent power distribution pyrolysis particle perception and warning system for wind turbine cabins, including: an integrated particle sampling module, embedded in the forced air cooling duct inlet of the power distribution cabinet; a multi-spectral laser analysis module, connected to the integrated particle sampling module, using dual-wavelength laser to identify particle types and calculate concentrations; a power distribution parameter coupling module, synchronously calculating the particle concentration growth rate, cable temperature rise rate and insulation resistance decrease rate; a risk prediction module, fusing multi-dimensional input features and outputting the fire probability through the LSTM model; and a graded warning module, linking and executing graded response measures according to the probability values.

[0050] The integrated particle sampling module is a composite structure installed directly on the air inlet path of the forced air cooling system. It optimizes the airflow path to efficiently capture submicron particles. The multispectral laser analysis module uses a dual-wavelength laser beam, 532nm and 1064nm, to cross-irradiate particles, leveraging the differences in scattering intensity of particles of different materials at specific wavelengths to identify their type. The power distribution parameter coupling module synchronizes the three-phase current of the current transformer, the cable connector temperature of the temperature sensor, and the insulation resistance of the insulation tester to establish a dynamic correlation model between particle concentration changes and electrical parameter degradation. The risk prediction module uses the three-phase current of the current transformer, the concentration of each type of pyrolytic particles, and a historical fault database to obtain multidimensional input features for the LSTM network. These features are then input into the trained LSTM network to predict fire probability. The graded warning module triggers different levels of emergency measures based on the predicted probability values, achieving precise risk response.

[0051] Specifically, the forced air cooling system's airflow, carrying pyrolytic particles, enters the integrated sampling module, where it undergoes multi-stage filtration to obtain a purified sample. A dual-wavelength laser beam cross-irradiates the particle swarm within the optical chamber. The scattered light signals undergo spectroscopic processing and the intensity ratio is calculated. This is then matched against a preset database to determine the particle type. Simultaneously collected electrical parameters are dynamically calculated to generate a growth rate indicator. A secondary warning is triggered when multiple indicators simultaneously exceed thresholds. Multidimensional input features derived from the current transformer's three-phase current, the concentration of each type of pyrolytic particle, and a historical fault database are fed into an LSTM model, which outputs the probability of fire occurring in the future. Based on this probability, the system gradually activates ventilation, power outages, and fire extinguishing devices, forming a progressive prevention and control system.

[0052] Compared with existing technologies, traditional solutions use external single-wavelength detectors, which are prone to misjudgment due to environmental interference. However, this invention significantly improves particle capture accuracy through duct-embedded sampling and dual-wavelength identification technology. Existing technologies rely on fixed threshold alarms and fail to quantify risk levels. This invention utilizes an LSTM model to achieve dynamic probability prediction and establish a scientific, graded response mechanism. Traditional system parameter analysis in isolation leads to a high false negative rate. This invention utilizes a multi-parameter coupled model to achieve comprehensive judgment, effectively reducing the false positive rate.

[0053] Through the above-mentioned technical solution, the present invention achieves precise capture and type identification of submicron pyrolytic particles, solving the problem of insufficient sensitivity of traditional detection devices. Multi-parameter dynamic coupling analysis effectively overcomes the limitations of single-threshold alarms and improves the reliability of early warning. The risk prediction model based on spatiotemporal characteristics can quantify the probability of fire occurrence and provide data support for graded response. The progressive prevention and control strategy maximizes the operational continuity of the power generation system while ensuring equipment safety.

[0054] In an optional embodiment, as Figure 2 As shown, the integrated particle sampling module includes an electrostatic enrichment ring 1, a vortex deceleration chamber 2, and a ceramic filter membrane assembly 3, which are sealed and connected in sequence along the airflow direction. The outer wall of the electrostatic enrichment ring 1 is provided with a bias electrode lead 11, which is electrically connected to an external high-voltage power supply via a sealed cable. The interior of the vortex deceleration chamber 2 is provided with a vortex guide plate 21. The ceramic filter membrane assembly 3 comprises multiple layers of stacked ceramic filter membranes.

[0055] Among them, the electrostatic enrichment ring refers to a ring structure that forms a non-uniform electric field by applying a preset bias voltage. Specifically, it can be realized by combining a titanium alloy electrode with a polytetrafluoroethylene insulating matrix, and is used to establish a directional electric field gradient in the air flow channel to adsorb charged particles. The vortex guide plate refers to a metal component with a spiral guide surface. Specifically, it can be formed by 316L stainless steel through a CNC bending process, and is used to convert a straight airflow into a rotating flow field to reduce the speed of particle movement. The multi-layered ceramic filter membrane refers to a filter medium with different pore size distributions. Specifically, it can be prepared by a gradient sintering process using an alumina ceramic substrate, and is used to remove large-size interfering particles through a step-by-step interception method.

[0056] Specifically, charged particles are captured by the electrostatic enrichment ring under the influence of a non-uniform electric field. The airflow then forms a spiral motion path through the vortex guide plate to prolong particle retention time, ultimately achieving particle size screening through a multi-layer ceramic membrane. The electrostatic adsorption process prioritizes the capture of pyrolytic particles with significant charge characteristics. The vortex deceleration structure reduces airflow velocity to ensure adsorption efficiency, and the gradient filtration mechanism effectively removes non-target particles such as dust. The three-stage structure is sealed together to form a directional processing flow, maintaining stable particle screening performance under complex airflow conditions.

[0057] Compared to existing technologies, traditional solutions using only mechanical filtration or electrostatic adsorption suffer from high submicron particle penetration and the misattraction of interfering particles. This invention leverages the synergistic effect of electric field adsorption and mechanical filtration to selectively capture submicron target particles while maintaining airflow. Compared to conventional linear duct designs, the vortex deceleration structure significantly extends particle processing time by altering the flow field morphology.

[0058] Through the above technical solution, the present invention effectively improves the capture efficiency of submicron pyrolysis particles, reduces the interference misjudgment rate of non-target particles such as dust, and can still maintain stable particle screening performance in a strong airflow disturbance environment, providing a high-purity sample basis for subsequent particle type identification.

[0059] In an optional embodiment, the multi-spectral laser analysis module includes a dual-wavelength laser emission unit, an optical gas chamber, a scattered light collection unit, a signal processing unit and a particle identification unit. The dual-wavelength laser emission unit synchronously emits laser beams with wavelengths of 532nm and 1064nm respectively. The optical gas chamber guides the airflow containing submicron pyrolysis particles through the intersection area of ​​the two laser beams. The scattered light collection unit captures the side scattered light generated after the particles are irradiated, and uses a filter wheel to separate the 532nm and 1064nm scattered light signals. The signal processing unit converts the optical signal into an electrical pulse signal and performs three-level filtering. The particle identification unit calculates the scattering intensity ratio based on the filtered signal, and outputs the particle type and concentration after matching the pyrolysis particle type database.

[0060] Among them, the dual-wavelength laser emission unit refers to a light source system that can generate two lasers of different wavelengths. Specifically, it can use Nd:YAG lasers in combination with frequency-doubling crystals to achieve 532nm wavelength output, while retaining the fundamental frequency 1064nm wavelength, and achieve dual-beam coaxial output through a beam splitter prism. This unit provides an optical feature basis for particle identification through the differences in the scattering characteristics of materials at different wavelengths. The optical gas chamber refers to a closed cavity with a specific flow channel design. Specifically, it can use quartz glass material with a conical air inlet to achieve laminar flow guidance, ensuring that particles are fully irradiated when flowing through the intersection area of ​​the two laser beams. The filter wheel in the scattered light collection unit refers to a rotary filter device with a precision positioning mechanism. Specifically, a stepper motor can be used to drive a turntable on which 532nm and 1064nm bandpass filters are installed to achieve temporal separation of the two wavelengths of scattered light and eliminate interference from ambient stray light. The three-stage filtering processing of the signal processing unit includes high-pass filtering, band-pass filtering and adaptive filtering. Specifically, an operational amplifier can be used to build an active filter to eliminate high-frequency noise, and the characteristic frequency band can be extracted through a digital signal processor, combined with a reference signal generation algorithm to suppress electromagnetic interference pulses.

[0061] Specifically, the dual-wavelength laser emission unit generates 532nm and 1064nm laser beams, which are collimated to form a cross-irradiation area. The airflow containing particles passes through this area along a laminar flow path in the optical air chamber, and the particles are irradiated by two laser beams at the same time to produce scattered light. The scattered light collection unit introduces the side scattered light into the filter wheel through the reflector group. The filter wheel alternately switches the two filters at a set speed to separate the scattered light signals of the corresponding wavelengths. The photodetector converts the separated light signal into an electrical pulse. The signal processing unit sequentially performs 10kHz high-pass filtering to eliminate environmental noise, 1MHz band-pass filtering to extract particle characteristic signals, and finally suppresses electromagnetic interference through adaptive filtering. The particle identification unit calculates the ratio of the intensities of the scattered light of the two wavelengths in real time, compares the ratio with the characteristic value of the pyrolysis particles in the pre-stored database, and determines the particle type and then counts the concentration of each type of particle based on the pulse count.

[0062] Compared with existing technologies, traditional single-wavelength laser detection technology cannot distinguish particle types and relies solely on scattered light intensity threshold judgment, which is easily affected by dust interference. The present invention uses dual-wavelength scattering intensity ratio feature recognition combined with a filter wheel to achieve spectral separation, effectively distinguishing between pyrolytic particles of insulating materials and interfering particles. Existing technologies use fixed-frequency filtering to make it difficult to eliminate interference from complex electromagnetic environments. The three-level dynamic filtering mechanism of the present invention can adaptively adjust the filtering parameters to improve the sensitivity of weak signal detection. Conventional particle detection equipment relies on single concentration statistics. The present invention achieves material-level fault tracing by establishing a pyrolytic particle type database.

[0063] Through the above-mentioned technical solution, the present invention can accurately identify the material type of submicron pyrolytic particles, distinguishing characteristic particles produced by the pyrolysis of insulating materials from non-pyrolytic particles in the environment, and avoiding false alarms caused by dust and water mist. Furthermore, based on the correspondence between the intensity of the dual-wavelength scattering signal and the number of particles, the concentration of different types of pyrolytic particles can be accurately quantified, providing reliable data support for early fire warning. The three-stage filtering process effectively suppresses electromagnetic interference generated by the operation of electrical equipment, ensuring detection stability under complex operating conditions.

[0064] In an optional embodiment, the power distribution parameter coupling module includes a synchronous acquisition unit, a dynamic calculation unit, and an early warning trigger unit. The synchronous acquisition unit synchronously acquires the three-phase current of the current transformer, the cable joint temperature of the temperature sensor, and the insulation resistance of the insulation tester. The dynamic calculation unit calculates the particle concentration growth rate, the cable joint temperature rise rate, and the insulation resistance decrease rate based on the cable joint temperature, insulation resistance, and the concentration of each type of pyrolytic particles. The early warning trigger unit activates a secondary early warning when the particle concentration growth rate, cable joint temperature rise rate, and insulation resistance decrease rate simultaneously meet the early warning trigger conditions.

[0065] The synchronous acquisition unit is a hardware interface that enables the simultaneous measurement of multiple electrical parameters. This is achieved using a multi-channel synchronous sampling chip coupled with the RS485 communication protocol. Timestamp alignment technology ensures that the acquisition time deviation of three-phase current, temperature, and insulation resistance data is less than 1ms. This unit eliminates timing errors and provides a consistent data foundation for dynamic calculations.

[0066] The dynamic calculation unit is a processor module that performs parameter rate-of-change calculations. Specifically, it uses an FPGA chip to perform real-time differential operations. It calculates particle concentration growth rate, cable joint temperature rise rate, and insulation resistance drop rate using a sliding time window. This unit quantifies the dynamic process of equipment degradation and establishes a multi-dimensional degradation assessment index.

[0067] The early warning trigger unit is a control module that performs multi-conditional logical judgments. Specifically, it can use a programmable logic controller to implement an AND gate judgment circuit. It triggers an early warning only when three rate-of-change indicators simultaneously exceed preset thresholds. By establishing a multi-parameter coordinated deterioration judgment, this unit eliminates false triggers caused by single parameter anomalies.

[0068] Specifically, the three-phase current data is converted into a 0-5V analog signal by a current transformer and then input into the synchronous acquisition unit. The cable joint temperature is collected by a K-type thermocouple, and the insulation resistance is transmitted as a digital signal by an insulation tester. The dynamic calculation unit uses a 10-second time window period and a central difference method to calculate the particle concentration growth rate. The slope of the cable joint temperature curve is fitted as the temperature rise rate using the least squares method. The insulation resistance data is logarithmically transformed and its rate of decrease is calculated. The early warning trigger unit compares the three rate of change indicators with the preset thresholds in real time. If and only if all three exceed the limit at the same time, a second-level early warning instruction is sent to the graded early warning module.

[0069] Compared to existing technologies, traditional monitoring systems only monitor a single parameter or use sequential judgment logic. For example, they trigger an alarm simply by detecting a temperature exceeding a limit, making them susceptible to interference from sudden changes in ambient temperature or transient current surges. This invention eliminates timing deviations by synchronously collecting multiple parameters. It uses a logic and judgment mechanism to require that three key degradation indicators exceed their limits in a coordinated manner, effectively filtering out transient interference signals. For example, it can distinguish between the continuous concentration increase caused by a true pyrolysis process and the particle concentration fluctuations caused by transient dust interference.

[0070] Through the above technical solution, the present invention solves the technical problems of traditional monitoring systems, which suffer from high false alarm rates and high rates of missed early fault detection due to single-parameter judgment. Through the simultaneous acquisition of multi-dimensional parameters and dynamic coupled analysis, it can accurately identify the synergistic deterioration trends of equipment insulation degradation, increased contact resistance, and pyrolytic particle release. It captures early fault characteristics while eliminating transient interference, reducing the false alarm rate of the second-level warning system to less than 18% of traditional systems and accelerating early fault detection by more than 30 minutes.

[0071] In an optional embodiment, the warning triggering condition of the power distribution parameter coupling module is: the particle concentration growth rate is greater than 10 4 particles / (m 3 ·min), the cable joint temperature rise rate is greater than 5℃ / min, and the insulation resistance drop rate is greater than 10% / min. When the three conditions are met at the same time, the second level warning will be activated.

[0072] The particle concentration growth rate refers to the change in the concentration of pyrolysis particles per unit time. Specifically, a differential algorithm can be used to calculate the concentration difference between adjacent sampling periods to characterize the severity of the material's pyrolysis reaction.

[0073] Among them, the temperature rise rate of the cable joint refers to the instantaneous rate of temperature change at the conductor connection part. Specifically, the temperature sensor can be used to collect data and then the first-order derivative can be calculated through a sliding window to reflect the Joule heat accumulation rate caused by contact resistance degradation.

[0074] The insulation resistance drop rate refers to the attenuation ratio of the insulation medium resistance value. Specifically, the resistance value can be measured with an insulation tester and then the logarithmic rate of change can be calculated to quantify the conductive channel formation process caused by the carbonization of the insulation material.

[0075] Specifically, by establishing a multi-parameter joint triggering mechanism, the system continuously monitors the dynamic changes of three parameters: particle concentration, temperature, and resistance. When the particle concentration growth rate exceeds the set threshold, it indicates that there is abnormal pyrolysis particle release; when the temperature rise rate of the cable joint exceeds the set threshold, it indicates that the conductor contact surface is abnormally heated; when the insulation resistance drop rate exceeds the set threshold, it indicates that the insulation layer is deteriorating at an accelerated rate. Only when the three parameters exceed their respective thresholds at the same time, the system determines that there is a risk of compound failure, and triggers the second-level warning. This mechanism effectively filters out abnormal fluctuations in a single parameter caused by environmental interference through multi-dimensional parameter cross-validation, such as the instantaneous temperature rise of the temperature sensor caused by airflow disturbances in the cabin, and random changes in insulation resistance due to humidity fluctuations.

[0076] Compared with existing technologies, traditional solutions only trigger alarms based on a single parameter exceeding a threshold. For example, if the temperature exceeds 80°C, the warning will be activated. However, it cannot distinguish between normal operating fluctuations and real faults. The present invention adopts a composite criterion design, requiring that the three key parameters characterizing material pyrolysis, conductor overheating, and insulation degradation must simultaneously reach a dangerous evolution rate, eliminating false triggers caused by environmental factors such as vibration, temperature difference, and humidity. For example, in a sudden wind condition, although vibration may cause a transient increase in contact resistance and cause a temperature rise, the system will not trigger an alarm if it is not accompanied by an increase in particle concentration and a decrease in insulation resistance.

[0077] Through the above-mentioned technical solution, the present invention can accurately distinguish between true faults and environmental interference signals, reducing the false alarm rate from 32% in traditional solutions to below 5%. Specifically, when encountering cabin vibration, the system automatically blocks temperature rise anomalies caused by mechanical shock by detecting that there is no sustained decrease in insulation resistance. When encountering high humidity environments, it automatically ignores short-term fluctuations in insulation resistance by verifying that particle concentration does not increase exponentially. This multi-parameter collaborative verification mechanism enhances the credibility of early warning signals and avoids unplanned downtime caused by false triggering.

[0078] In an optional embodiment, the graded response measures adopt a progressive response, including starting negative pressure exhaust in the cabin when the fire probability reaches 30%, cutting off non-critical loads when it reaches 60%, and triggering perfluorohexanone fire extinguishing agent and disconnecting from the grid when it reaches 90%.

[0079] Fire probability refers to the probability of a fire occurring within a target future time period, as output by the fire risk prediction model. Specifically, an LSTM neural network can be used to perform time-series analysis on the spatiotemporal distribution matrix of particle concentration, three-phase current imbalance, and historical fault characteristics to generate a probability value. This probability value is then used to dynamically adjust the response strategy. A progressive response involves implementing measures of varying intensity in stages based on risk levels. Specifically, this can be achieved by triggering corresponding control commands through pre-set multi-level thresholds, thereby optimizing the allocation of emergency resources. Negative pressure ventilation involves using a fan system to create a negative pressure environment within a confined space to quickly expel flammable gases. Specifically, variable-frequency fans combined with airflow guides can be used to achieve directional exhaust, reducing the risk of combustion. Non-critical loads refer to auxiliary equipment power circuits that do not affect the core functions of the generator set. These loads can be automatically disconnected based on a pre-set load priority list to maintain continuous power to the main circuit. Perfluorohexanone fire extinguishing agent is a clean gas extinguishing medium with chemical inhibition. Specifically, a high-pressure storage tank combined with an atomizing nozzle can be used to rapidly spray and interrupt the combustion chain reaction. Off-grid means disconnecting the electrical connection between the wind turbine and the power grid. Specifically, the grid connection point contactor can be cut off by the circuit breaker actuator to prevent the accident from expanding.

[0080] Specifically, when the probability value output by the fire risk prediction model reaches 30%, the control system activates the negative pressure exhaust device to reduce the possibility of combustion by actively discharging the combustible gas accumulated in the cabin. At this time, the generator set remains in normal operation. When the probability value rises to 60%, the system automatically identifies and cuts off the power supply circuit of non-critical loads, reducing the accumulation of heat loads on electrical equipment while maintaining the continuous operation of the core power generation function. When the probability value exceeds 90%, the system synchronously triggers the perfluorohexanone fire extinguishing agent release device and the power grid tripping mechanism to block the development of the fire through the dual mechanisms of chemical suppression and physical isolation. The three response levels dynamically adjust the intervention intensity based on real-time risk assessment results, avoiding unnecessary equipment shutdowns in low-risk stages and ensuring rapid and effective emergency response in high-risk stages.

[0081] Compared with existing technologies, traditional solutions typically use a single threshold to trigger the injection of fire extinguishing agents, which can easily lead to equipment damage or delayed response due to misjudgment. This invention establishes a multi-level response mechanism, using non-destructive measures to slow the development of accidents in the early stages of risk accumulation, and gradually increasing the intensity of control as the risk escalates. This reduces the probability of misoperation and improves the accuracy of capturing critical disposal opportunities. For example, while existing technologies directly activate the fire extinguishing system when smoke is detected, this solution prioritizes exhaust risk reduction measures at the same risk stage, significantly reducing equipment maintenance costs caused by the misinjection of fire extinguishing agents.

[0082] Through this technical solution, the present invention addresses the issues of over-intervention and delayed response caused by the crude emergency response of traditional systems, achieving precise matching and dynamic optimization of fire prevention and control measures within wind turbine cabins. By implementing control strategies of varying intensities in phases, the system maximizes generator set uptime while ensuring equipment safety, while also avoiding the risk of fire spread caused by delayed response in high-risk situations.

[0083] In an optional embodiment, the system of the present invention further includes a fault tracing module, which constructs a three-dimensional airflow velocity model using six ultrasonic anemometers, calculates the backtracking path of the pyrolysis particles based on the three-dimensional airflow velocity model, and locates the coordinates of the pyrolysis source.

[0084] Among them, the six ultrasonic anemometers refer to ultrasonic velocity measuring devices arranged in six orthogonal spatial directions inside the distribution cabinet. Specifically, they can be implemented by an ultrasonic sensor array with three-dimensional vector measurement capabilities, which is used to synchronously collect airflow velocity components at different positions. The three-dimensional airflow velocity model refers to the reconstruction of discrete wind speed measurement values ​​into a continuous three-dimensional flow field distribution through a spatial interpolation algorithm. Specifically, the finite volume method in computational fluid dynamics can be used to model the flow field, which is used to characterize the airflow motion law inside the distribution cabinet. The backtracking path refers to the inverse solution of the particle motion trajectory based on the Lagrangian particle tracking method. Specifically, the fourth-order Runge-Kutta method can be used to solve the inverse motion equation, which is used to invert the migration path of pyrolysis particles in the flow field. The pyrolysis source coordinates refer to the convergence of the intersection points of multiple backtracking paths into spatial coordinate values ​​through the least squares method. Specifically, the particle trajectory clustering algorithm can be used for coordinate calculation to determine the release position of the pyrolysis particles.

[0085] Specifically, six ultrasonic anemometers form a spatial measurement network inside the distribution cabinet, acquiring six orthogonal airflow velocity components in real time. The three-dimensional airflow velocity model reconstructs discrete wind velocity data into a continuous three-dimensional flow field using a spatial interpolation algorithm, accurately describing the motion state of the airflow inside the distribution cabinet. When pyrolytic particles are detected, an inverse particle tracking algorithm is used to calculate the particle's motion trajectory in the flow field based on the current three-dimensional flow field distribution, and the initial position of the particle at the time of release is determined through time inversion. The pyrolytic source coordinates are determined through spatial clustering analysis of the trajectory intersection points, and the release source location information of the pyrolytic particles is ultimately output.

[0086] Compared to existing technologies, traditional solutions can only detect particle concentration using a single position sensor, failing to establish a correlation between airflow and particle migration, making it impossible to trace the source of particles. This invention combines three-dimensional flow field modeling with inverse particle tracking to overcome the spatial limitations of single-point detection and achieve precise localization of particle release sources.

[0087] Through this technical solution, the present invention can quickly locate the release source of pyrolytic particles, eliminating the need for extensive inspections of equipment within the power distribution cabinet by maintenance personnel and significantly reducing fault location time. The precise output of pyrolytic source coordinates can directly guide maintenance operations, effectively improving fault handling efficiency.

[0088] In an optional embodiment, the signal processing unit performs three-stage filtering on the original electrical pulse signal, including a first-stage high-pass filter with a cutoff frequency of 10kHz to eliminate environmental noise; a second-stage band-pass filter with a center frequency of 1MHz to extract particle characteristic signals; and a third-stage adaptive filter to suppress electromagnetic interference pulses.

[0089] Among them, the first-level high-pass filtering refers to a filtering method that allows signals above the cutoff frequency to pass through. Specifically, it can be implemented using a Butterworth filter. It is used to eliminate low-frequency environmental noise below 10kHz and avoid signal interference caused by mechanical vibration of the equipment. The second-level band-pass filtering refers to a filtering method that allows signals in a specific frequency band to pass through. Specifically, it can be implemented using a Chebyshev filter. By setting the center frequency to 1MHz to match the inherent frequency band of the submicron particle scattering signal, the particle characteristic signal can be effectively extracted. The third-level adaptive filtering refers to a filtering method that dynamically adjusts the filter parameters according to the real-time signal characteristics. Specifically, it can be implemented using the least mean square algorithm. It is used to suppress the high-frequency electromagnetic interference pulses generated by the operation of power equipment.

[0090] Specifically, during signal processing, the raw electrical pulse signal first undergoes a first-stage high-pass filter to remove low-frequency noise generated by cabin vibration and fan operation. It then enters a second-stage bandpass filter to retain the 1MHz frequency band signal that matches the scattering characteristics of submicron particles. Finally, a third-stage adaptive filter dynamically eliminates randomly occurring electromagnetic interference pulses. Through this three-stage filtering process, non-target signals such as environmental noise and equipment electromagnetic interference are gradually removed, retaining the particle signature signal with a high signal-to-noise ratio.

[0091] Compared to existing technologies, traditional particle detection systems typically use single-stage fixed-frequency filtering or dual-stage filtering designs, which are unable to simultaneously address the combined interference of low-frequency mechanical vibration noise, high-frequency electromagnetic interference, and the separation of target signal frequency bands. This invention, through a three-stage progressive filtering architecture, achieves the first coordinated processing of low-frequency environmental noise, high-frequency electromagnetic interference, and target signal frequency bands, solving the challenge of effectively extracting submicron particle signals in complex electromagnetic environments.

[0092] Through the above technical solution, the present invention can accurately distinguish between pyrolytic particle scattering signals and environmental interference signals, avoid false alarms caused by electromagnetic interference or mechanical vibration, and significantly improve the reliability of early pyrolytic particle detection.

[0093] In an optional embodiment, the multi-dimensional input features include: a particle concentration spatiotemporal distribution matrix, a dynamic change rate of three-phase current imbalance, and an insulation degradation pattern code matched in a historical fault library.

[0094] Among them, the particle concentration spatiotemporal distribution matrix refers to the matrix data that records the changes in the concentration of pyrolysis particles over time through a spatial coordinate grid. Specifically, it can be achieved by dividing the cabin space with a three-dimensional coordinate system and recording the concentration value change gradient of each grid point. It is used to characterize the diffusion path and aggregation trend of the pyrolysis particles. The dynamic change rate of the three-phase current imbalance refers to the rate of change of the difference in the effective value of the three-phase current over time. Specifically, it can be achieved by calculating the first-order derivative of the standard deviation of the three-phase current in real time, and is used to capture the correlation between transient electrical anomalies and the pyrolysis process. Insulation degradation pattern encoding refers to the digital identification of degradation features predefined in the historical fault library. Specifically, it can be achieved by using a hash algorithm to extract features and encode and store typical fault modes to provide prior knowledge constraints for the model.

[0095] Specifically, the present invention constructs a multi-dimensional input system by integrating spatiotemporal dynamic features, electrical transient features and historical experience features. After the spatiotemporal feature vectors of the spatiotemporal distribution matrix of the particle concentration are extracted, the evolution law of the spatial concentration gradient is analyzed through a sliding time window to identify the aggregation trend of pyrolytic particles in specific areas. The dynamic change rate of the three-phase current imbalance is calculated in real time by a differential circuit to calculate the instantaneous change of the current imbalance state, and capture the local temperature rise anomaly caused by poor contact or overload. The insulation degradation pattern code in the historical fault library uses a pattern matching algorithm to compare the current monitoring data with the typical fault characteristics for similarity, and outputs the closest degradation pattern code. These three types of features are input into the LSTM neural network after normalization, and the dynamic coupling relationship between the features is captured through time series modeling, and finally a fire risk probability prediction value based on multi-dimensional parameter fusion is output.

[0096] Compared with existing technologies, traditional methods only use particle concentration or current parameters at a single time point as input features, failing to capture the spatial distribution characteristics of particle diffusion and the transient changes in electrical parameters. This invention introduces a spatiotemporal distribution matrix and dynamic rate of change parameters, combined with the constraints of historical fault mode encoding, to construct a feature input system covering four dimensions: spatial, temporal, electrical, and empirical. This effectively addresses the problem of misjudgment caused by single-parameter analysis.

[0097] Through the above technical solution, the present invention realizes the multi-dimensional dynamic prediction of the fire risk of the wind turbine cabin distribution system, significantly reduces the probability of false alarms caused by environmental interference, improves the recognition accuracy of early hidden pyrolysis behavior, and at the same time enhances the model's adaptability to complex working conditions by introducing historical fault mode coding.

[0098] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A wind turbine cabin intelligent power distribution pyrolysis particle sensing and warning system, characterized in that: include: The integrated particle sampling module is embedded in the forced air cooling duct inlet of the power distribution cabinet to directly capture submicron pyrolysis particles in the airflow; A multispectral laser analysis module, connected to the integrated particle sampling module, is used to irradiate captured submicron pyrolytic particles with laser beams at wavelengths of 532nm and 1064nm, identify the type of pyrolytic particles based on the ratio of the scattered light intensity at 532nm to 1064nm, and calculate the concentration of each type of pyrolytic particles; The power distribution parameter coupling module is connected to the multi-spectral laser analysis module to synchronously collect the three-phase current of the current transformer, the cable joint temperature of the temperature sensor, and the insulation resistance of the insulation tester. It also calculates the particle concentration growth rate, cable temperature rise rate, and insulation resistance decrease rate in real time. It also determines whether to trigger a secondary warning based on the particle concentration growth rate, cable temperature rise rate, and insulation resistance decrease rate. a risk prediction module, connected to the multispectral laser analysis module and the power distribution parameter coupling module, respectively, for obtaining multidimensional input features, inputting the multidimensional input features into a pre-trained LSTM model, and outputting a fire probability value within a preset time period in the future; The multi-dimensional input feature is obtained based on the three-phase current of the current transformer, the concentration of each type of pyrolytic particles and the historical fault library; The hierarchical warning module is connected to the risk prediction module and is used to link the fire extinguishing, ventilation, and power grid tripping devices to execute hierarchical response measures according to the fire probability value; The system also includes: a fault tracing module, which is connected to the graded warning module and is used to construct a three-dimensional airflow velocity model through six ultrasonic anemometers, and calculate the backtracking path of the pyrolysis particles based on the three-dimensional airflow velocity model to locate the coordinates of the pyrolysis source.

2. The system according to claim 1, wherein: The integrated particle sampling module includes an electrostatic enrichment ring, a vortex deceleration chamber and a ceramic filter membrane assembly that are sealed and connected in sequence along the airflow direction; A bias electrode lead is provided on the outer wall of the electrostatic enrichment ring, and the bias electrode lead is electrically connected to an external high-voltage power supply through a sealed cable, and is used to transmit a preset bias voltage applied by the high-voltage power supply to the electrodes of the electrostatic enrichment ring, so as to form a non-uniform electric field between the electrodes of the electrostatic enrichment ring; The electrostatic enrichment ring is used to adsorb charged particles in the airflow under the non-uniform electric field; The vortex deceleration chamber is provided with a vortex guide plate inside, which is used to convert the straight airflow into a rotating flow field to reduce the speed of the airflow; The ceramic filter membrane assembly comprises multiple layers of stacked ceramic filter membranes, which are used for multi-level gradient filtration of interfering particles in the airflow and outputting purified submicron pyrolysis particles.

3. The system according to claim 1, wherein: The multi-spectral laser analysis module includes: Dual-wavelength laser emission unit, used for synchronously emitting laser beams with wavelengths of 532nm and 1064nm respectively; an optical gas chamber, used to guide an air flow containing submicron-sized pyrolytic particles through an intersection area of ​​two laser beams so that the submicron-sized particles are irradiated by the two laser beams; A scattered light collection unit is used to capture the side scattered light generated by submicron pyrolytic particles after being irradiated by two laser beams; and a filter wheel is used to separate the scattered light into a pure 532nm scattered light signal and a pure 1064nm scattered light signal; a signal processing unit connected to the scattered light collection unit, configured to convert the 532nm scattered light signal and the 1064nm scattered light signal into corresponding electrical pulse signals and perform three-stage filtering; The particle identification unit is connected to the signal processing unit and is used to calculate the scattering intensity ratio based on the filtered 532nm electric pulse signal and the 1064nm electric pulse signal; and match the scattering intensity ratio with a preset pyrolysis particle type database to output the type of pyrolysis particles and the concentration of each type of pyrolysis particles.

4. The system according to claim 3, characterized in that The power distribution parameter coupling module includes: Synchronous acquisition unit, used to synchronously acquire the three-phase current of the current transformer, the cable joint temperature of the temperature sensor, and the insulation resistance of the insulation tester; a dynamic calculation unit, connected to the synchronous acquisition unit and the multi-spectral laser analysis module, respectively, for calculating the particle concentration growth rate and the cable joint temperature rise rate based on the cable joint temperature and insulation resistance; and for calculating the particle concentration growth rate based on the concentration of each type of pyrolytic particles; The early warning trigger unit is used to activate the secondary early warning when the particle concentration growth rate, cable joint temperature rise rate, and insulation resistance drop rate simultaneously meet the early warning trigger conditions.

5. The system according to claim 4, characterized in that The warning triggering conditions of the power distribution parameter coupling module are: ,and The cable connector temperature rise rate is >5℃ / min, and Insulation resistance drop rate>10% / min.

6. The system according to claim 1, wherein: The tiered response measures adopt a progressive response and include: When the fire probability is ≥30%, start the negative pressure exhaust in the cabin; When the fire probability is ≥60%, cut off non-critical loads; When the fire probability is ≥90%, the perfluorohexanone fire extinguishing agent is triggered and the network is disconnected.

7. The system according to claim 3, wherein: The signal processing unit performs three-stage filtering on the electrical pulse signal, which includes: First-stage high-pass filtering with a cutoff frequency of 10kHz to eliminate ambient noise; Secondary band-pass filtering with a center frequency of 1 MHz to extract particle characteristic signals; Three-stage adaptive filtering to suppress electromagnetic interference pulses.

8. The system according to claim 1, wherein: The multi-dimensional input features include: The spatiotemporal distribution matrix of particle concentration, the dynamic change rate of three-phase current imbalance and the matching insulation degradation pattern encoding in the historical fault library.

Images