Power transmission corridor environment monitoring method based on microclimate fusion data, server and storage medium

Abstract

This invention provides a method, server, and storage medium for monitoring the environment of power transmission corridors based on micro-meteorological fusion data. The method acquires time-series data of wind speed, temperature, and humidity collected by micro-meteorological sensors on power transmission towers. This time-series data is input into a conductor / ground wire state estimation neural network. The forget gate in the recursive layer is dynamically modulated by wind speed disturbance pattern features to adaptively adjust the historical state retention ratio. The training loss incorporates loss terms related to icing thermal balance and wind deflection moment balance. Icing thickness and wind deflection angle are obtained from the network output. Then, a physical state recursive model recursively estimates the mechanical state parameters of the conductor / ground wire under thermal and moment balance constraints. Finally, a dynamic control command sequence for the line is generated by combining tower topology and safety constraints to trigger state adjustment. This method achieves deep integration of wind condition adaptive gating and physical recursive verification, improving the accuracy of environmental and mechanical state perception and the timeliness of control response in power transmission corridors.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a method, server and storage medium for monitoring the environment of power transmission corridors based on micro-meteorological fusion data. Background Technology

[0002] Environmental monitoring of transmission corridors refers to the continuous sensing of meteorological conditions and the mechanical condition of conductors and ground wires within overhead transmission line corridors to support the timely detection of risks such as line icing and wind deflection. Currently, the common method for environmental monitoring of transmission corridors is to deploy micro-meteorological sensors on towers to collect meteorological parameters such as wind speed, temperature, and humidity. These meteorological parameters are then converted into estimates of conductor icing thickness or wind deflection angle using preset empirical mapping rules or simplified physical formulas. However, this conversion method based on static rules or simplified formulas cannot accurately characterize the dynamic evolution of conductor and ground wire condition responses under drastic changes in micro-meteorological conditions. This leads to a significant deviation between the obtained estimates of conductor and ground wire mechanical condition and the actual condition, thus affecting the accuracy and timeliness of subsequent line operation and control decisions. Summary of the Invention

[0003] This invention provides a method, server, and storage medium for monitoring the environment of power transmission corridors based on micro-meteorological fusion data.

[0004] In a first aspect, embodiments of the present invention provide a method for monitoring the environment of a power transmission corridor based on micro-meteorological fusion data, comprising: Acquire micro-meteorological time-series data collected by micro-meteorological sensors deployed on power transmission towers within a continuous time window; Micro-meteorological time-series data are input into a trained conductor-ground wire state estimation neural network. The forget gate dynamic value of the recursive processing layer inside the conductor-ground wire state estimation neural network is dynamically modulated by the wind speed disturbance pattern features derived in real time from the wind speed time-series components. The forget gate dynamic value controls the proportion of historical hidden state information retained and discarded by the recursive processing layer. Physical mechanism constraints are introduced into the conductor-ground wire state estimation neural network during the training phase. The training loss function is embedded with conductor-ground wire icing thermal balance loss term and wind deflection moment balance loss term to guide the network to converge within the physical feasible region. The neural network for estimating conductor and ground wire states directly outputs the mechanical state parameters of the conductor and ground wire based on the input micro-meteorological time series data. The mechanical state parameters of the conductor and ground wire include the conductor ice thickness characterization quantity and the conductor wind deflection angle characterization quantity. The conductor ice thickness characterization quantity reflects the continuous change of the degree of ice accumulation on the conductor surface, and the conductor wind deflection angle characterization quantity reflects the change of the conductor's deviation angle under wind load. The mechanical state parameters of the conductor and ground wire are input into the pre-constructed recursive model of the physical state of the transmission corridor. The recursive model of the physical state of the transmission corridor performs a physical mechanism-driven recursive estimation of the mechanical state of the conductor and ground wire at the current moment based on the mechanical state parameters of the conductor and ground wire at the previous moment and the micro-meteorological time series data at the current moment, and generates the recursively estimated mechanical state parameters of the conductor and ground wire. When the recursive model of the physical state of the transmission corridor is running, it is simultaneously subject to the constraints of the ice-heat balance and the wind deflection moment balance. Based on the recursively estimated mechanical state parameters of the conductor and ground wire, combined with the topological connection relationship of each tower segment in the transmission corridor and the line safety operation constraints, a control signal sequence containing line dynamic operation control instructions is generated. The line dynamic operation control instructions are used to initiate the line operation status adjustment operation of the transmission corridor.

[0005] Secondly, embodiments of the present invention provide a server, a computer device including a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to realize the above-mentioned method for monitoring the environment of power transmission corridors based on micro-meteorological fusion data.

[0006] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which is loaded and executed by a processor to implement the above-described method for monitoring the environment of a power transmission corridor based on micro-meteorological fusion data.

[0007] The embodiments of the present invention have the following beneficial effects: This invention acquires time-series data of wind speed, temperature, and humidity from micro-meteorological sensors and inputs this data into a conductor-to-ground wire state estimation neural network that incorporates loss terms related to icing thermal balance and wind deflection moment balance during training. The recursive processing layer within the network dynamically modulates the forget gate value based on disturbance pattern characteristics derived from the wind speed time series, adaptively adjusting the retention and discard ratio of historical state information. This allows the network to output physically accurate conductor icing thickness and wind deflection angle representations when facing complex wind conditions such as isolated sudden wind disturbances or intermittent impacts. These representations are then fed into a recursive model of the transmission corridor's physical state, constrained by icing thermal balance and wind deflection moment balance conditions, for recursive estimation. The system uses the previous state and current meteorological data for physical mechanism-driven state correction, further suppressing potential physical inconsistencies caused by purely data-driven approaches. This ensures that the final conductor mechanical state parameters possess both the responsiveness of data-driven approaches and the reliability of physical constraints. Finally, by combining the tower segment topology and safe operation limitations, dynamic line operation control commands are generated, achieving highly timely and reliable perception and control of icing and wind deflection risks in the transmission corridor. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the application environment provided in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the method for monitoring the environment of a power transmission corridor based on micro-meteorological fusion data provided in an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the application logic of the power transmission corridor environmental monitoring method based on micro-meteorological fusion data provided in this embodiment of the invention. Figure 4 This is a structural block diagram of the server provided in an embodiment of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0010] In some embodiments, the power transmission corridor environmental monitoring method based on micro-meteorological fusion data provided in this invention is applied to, for example... Figure 1 The application environment shown. For example, as... Figure 1 As shown, the application environment includes a micro-weather sensor 10 and a server 20.

[0011] The micro-weather sensor 10 can integrate, for example, a wind speed sensor, a temperature sensor, and a humidity sensor. The server 20 can be a single server, a server cluster consisting of multiple servers, or a cloud computing service center.

[0012] The aforementioned micro-weather sensor 10 and the aforementioned server 20 transmit data via a network. In some embodiments, the micro-weather sensor 10 collects micro-weather time-series data and sends the micro-weather time-series data to the server 20, which then processes the data to generate control signals.

[0013] One point that needs to be clarified is that the above Figure 2 The descriptions provided are merely exemplary and illustrative. In exemplary embodiments, the functions of the micro-weather sensor 10 and the server 20 can be flexibly configured and adjusted, and the embodiments of the present invention do not limit this.

[0014] Please refer to Figure 2 and Figure 3 The power transmission corridor environmental monitoring method based on micro-meteorological fusion data provided by this invention can be derived from the above. Figure 1 The method is executed on server 20. It may include the following steps:

[0015] Step S100: Obtain micro-meteorological time-series data collected by micro-meteorological sensors deployed on transmission towers within a continuous time window. The micro-meteorological time-series data includes wind speed time-series components, temperature time-series components, and humidity time-series components. The time window length of the micro-meteorological time-series data matches the input dimension of the subsequent conductor state estimation neural network.

[0016] Micro-meteorological sensors are integrated meteorological parameter measuring devices installed at specific locations on the tower body or crossarm of transmission lines. These devices can integrate wind speed, temperature, and humidity sensing units. These three units operate synchronously at a unified sampling frequency, enabling continuous sensing of airflow, environmental thermodynamics, and water vapor content in the local area of ​​the transmission corridor. For example, the wind speed sensing unit uses ultrasonic anemometers, calculating the wind speed vector by measuring the time difference of sound wave propagation in flowing air in orthogonal directions using multiple pairs of ultrasonic transducers. The temperature sensing unit uses platinum resistance temperature probes, utilizing the linear change in platinum resistance with temperature to obtain ambient temperature. The humidity sensing unit uses polymer thin-film capacitive humidity sensors, obtaining relative humidity based on the change in dielectric constant after water molecules are adsorbed by the moisture-sensing material.

[0017] Micro-meteorological time-series data is a multi-dimensional structured data set formed by sequentially collecting data from micro-meteorological sensors at fixed time sampling intervals within a set continuous time window and sorting it by timestamp. Each row of records in this data set corresponds to a sampling time point, and each row simultaneously carries the wind speed measurement, temperature measurement, and humidity measurement values ​​at that moment. The wind speed time-series component is a single-variable time series specifically describing the change of air flow speed over time, separated from the micro-meteorological time-series data. Each element in this series corresponds to a scalar value of wind speed at a sampling time. The temperature time-series component is a single-variable time series specifically describing the change of ambient temperature over time, separated from the micro-meteorological time-series data. Each element in this series corresponds to a Celsius temperature value at a sampling time. The humidity time-series component is a single-variable time series specifically describing the change of relative humidity over time, separated from the micro-meteorological time-series data. Each element in this series corresponds to a relative humidity percentage at a sampling time. The time window length is the number of sampling points corresponding to the total duration of continuous sampling performed by the micro-meteorological sensor. This number determines the number of time steps contained in the data sample of a single input to the conductor state inference neural network.

[0018] The conductor-to-ground wire state estimation neural network is an artificial neural network model trained offline on historical data. During the design phase, the length of the sequence that its input layer can receive is fixed; the numerical value of the input dimension is the specific value of this sequence length. The specific implementation method for matching the time window length with the input dimension is as follows: During the deployment phase, the input tensor shape parameter recorded in the conductor-to-ground wire state estimation neural network model configuration file is read, and the value of the time dimension in this shape parameter is extracted. This value is used as the basis for setting the buffer length parameter in the micro-meteorological sensor data acquisition module, so that the sensor cyclically fills the data buffer according to this set length to form a complete time window data block. If the input layer of the conductor state estimation neural network requires receiving sequence data of 64 time steps, the data acquisition firmware of the micro-meteorological sensor obtains the value 64 from the neural network configuration parameters during initialization, configures the capacity of the internal circular buffer to 64 data slots, and writes the data of each time step to the buffer and moves the write pointer. When the write pointer returns to the starting point, a complete time window data block contains exactly 64 sets of synchronous measurements of wind speed, temperature and humidity. This data block is then encapsulated as micro-meteorological time series data and sent to the neural network for forward inference.

[0019] Step S200: Input the micro-meteorological time series data into the trained conductor-ground wire state estimation neural network. The forget gate dynamic value of the recursive processing layer inside the conductor-ground wire state estimation neural network is dynamically modulated by the wind speed disturbance pattern features derived in real time from the wind speed time series components. The forget gate dynamic value controls the retention and discard ratio of historical hidden state information in the recursive processing layer. Physical mechanism constraints are introduced into the conductor-ground wire state estimation neural network during the training phase. The training loss function is embedded with conductor-ground wire icing thermal balance loss term and wind deflection moment balance loss term to guide the network to converge within the physical feasible region.

[0020] In one implementation, step S200 specifically includes the following steps S210 to S260: Step S210: Perform multi-time-granularity disturbance detection on the wind speed time series components to generate wind speed disturbance pattern features including instantaneous disturbance amplitude sequence, continuous disturbance intensity sequence and disturbance intermittent frequency sequence. The instantaneous disturbance amplitude sequence captures short-term abrupt changes in wind speed, the continuous disturbance intensity sequence tracks the maintenance state of wind speed fluctuations, and the disturbance intermittent frequency sequence records the switching frequency between periods of stable wind conditions.

[0021] In one implementation, step S210 specifically includes the following steps S211 to S216: Step S211: Input the wind speed time series component into the disturbance separation structure. The disturbance separation structure separates the trend component and fluctuation component of the wind speed time series component to generate a smooth trend sequence and a residual fluctuation sequence. The smooth trend sequence reflects the overall evolution direction of the wind speed, and the residual fluctuation sequence reflects the short-term fluctuation of the wind speed in the overall evolution direction.

[0022] The perturbation separation structure is a signal processing module based on an adaptive time-scale decomposition algorithm. It decomposes the original wind speed time-series component into two components with different physical meanings, allowing for independent analysis of the macroscopic evolution trend and microscopic instantaneous fluctuations of wind speed. This perturbation separation structure is constructed by combining a local weighted regression smoothing algorithm with the extraction of differential residuals from the original signal. After the wind speed time-series component enters the perturbation separation structure, it first moves point-by-point along the time axis through a sliding window. At each window position, wind speed data points within the window are selected as local fitting samples. Weighted least squares is used to perform low-order polynomial fitting on these local fitting samples. The weight of each data point during fitting is determined by a cubic weighting function based on its distance from the window center point; data points closer to the center point have a higher weight, and those farther away have a lower weight. This yields the smoothed value corresponding to the center position of the window. After the sliding window traverses the entire wind speed time-series component, all smoothed values ​​are arranged in chronological order to form a smoothed trend sequence. The smoothed trend sequence characterizes the overall evolution direction of wind speed over a relatively long time scale, filtering out high-frequency random fluctuation components. Subsequently, the original wind speed time series components are subtracted from the smoothed trend sequence point by point along the time axis. The original wind speed value at each time point is subtracted from the corresponding smoothed trend value, and the resulting difference sequence is the residual fluctuation sequence. The residual fluctuation sequence retains the short-term fluctuation information of wind speed in the overall evolution direction, reflecting the turbulent pulsations and instantaneous gust disturbances in the airflow. This separation of trend components from fluctuation components allows subsequent abrupt change detection, fluctuation envelope analysis, and stationary period identification to be performed on the pure fluctuation signal after removing trend interference, improving the accuracy and anti-interference capability of disturbance feature extraction.

[0023] Step S212: Perform mutation detection on the residual fluctuation sequence, mark the points in the residual fluctuation sequence where the change amplitude of the fluctuation value at adjacent time points exceeds the mutation response boundary as mutation candidate points, and merge consecutive mutation candidate points into mutation event segments along the time axis, record the peak amplitude and duration span of each mutation event segment, and generate a mutation event record set.

[0024] Abrupt change detection is performed, for example, using an outlier identification algorithm based on a moving quantile dynamic threshold. First, for each time point in the residual fluctuation sequence, the residual fluctuation values ​​within a neighborhood interval before and after it are taken as local background samples. The upper and lower quantiles of these local background samples are calculated as local fluctuation reference ranges. The abrupt change response boundary is the numerical interval formed by extending an adaptive offset upwards and downwards from this local fluctuation reference range. The adaptive offset is obtained by multiplying the standard deviation of the local background samples by a sensitivity factor, the value of which is pre-calibrated through statistical analysis of historical wind speed disturbance events. For each time point in the residual fluctuation sequence, the absolute value of the difference between the fluctuation value at that time point and the fluctuation value at the previous adjacent time point is calculated. This absolute value is compared with the width of the abrupt change response boundary at that time point. If the absolute value exceeds the width of the abrupt change response boundary, the time point is marked as a candidate abrupt change point. After marking all time points, the candidate abrupt change point marking sequence is scanned forward along the time axis, and temporally consecutive, uninterrupted candidate abrupt change points are merged into the same abrupt change event segment. For each merged mutation event segment, the absolute values ​​of all residual fluctuations are iterated over the time range covered by the event segment. The largest absolute value is identified as the peak amplitude of the mutation event segment. Simultaneously, the time span between the end and start times of the event segment is calculated as the duration span. The peak amplitude and duration span of each mutation event segment are paired and recorded, and the results are aggregated to generate a mutation event record set. This set provides the foundation for constructing subsequent instantaneous disturbance amplitude sequences.

[0025] Step S213: For each mutation event segment in the mutation event record set, extract its peak amplitude as the instantaneous perturbation amplitude of the event segment, and associate the peak amplitude with each time point covered by the mutation event segment to obtain the instantaneous perturbation amplitude sequence.

[0026] The instantaneous disturbance amplitude sequence is constructed based on a set of abrupt event records. Each abrupt event segment record in the set is traversed, and the stored peak amplitude value is read and directly used as the instantaneous disturbance amplitude value for all time points within that abrupt event segment. For each time point covered by a abrupt event segment, its instantaneous disturbance amplitude is assigned the peak amplitude value; for time points not covered by any abrupt event segment, its instantaneous disturbance amplitude is assigned zero. The instantaneous disturbance amplitude values ​​of all time points are arranged sequentially along the time axis to obtain the instantaneous disturbance amplitude sequence. This sequence characterizes the maximum instantaneous abrupt change in wind speed at each time point, reflecting the distribution of the intensity of gust impact along the time axis.

[0027] Step S214: Perform wave envelope analysis on the residual wave sequence to generate upper and lower envelopes. Determine the wave energy curve based on the change of the envelope width between the upper and lower envelopes over time. Perform sliding accumulation on the wave energy curve to generate the continuous disturbance intensity value at each time point. Arrange the values ​​in chronological order to obtain the continuous disturbance intensity sequence.

[0028] Fluctuation envelope analysis can employ a method based on extremum point interpolation reconstruction to generate the envelope. First, local extrema are detected in the residual fluctuation sequence, identifying the maxima corresponding to peaks and the minima corresponding to troughs. A maxima is determined by its value being greater than its left and right neighboring values, while a minima is determined by its value being less than its left and right neighboring values. For all identified maxima, a piecewise cubic polynomial interpolation algorithm is used to interpolate between them, generating a smooth, continuous upper envelope that passes through all maxima. Similarly, the same piecewise cubic polynomial interpolation algorithm is used for all minima, generating a smooth, continuous lower envelope that passes through all minima. The upper and lower envelopes delineate the fluctuation boundaries of the residual fluctuation sequence along the positive and negative axes, respectively. At each time point, the difference between the values ​​of the upper and lower envelopes is calculated. This difference represents the envelope width at that time point, reflecting the amplitude range of wind speed fluctuations at that moment. A larger envelope width indicates more severe wind speed fluctuations. The envelope widths of all time points are arranged chronologically to form a fluctuation energy curve. Subsequently, a sliding accumulation process is performed on the fluctuation energy curve. A sliding accumulation window of a specified length is set. Starting from the beginning of the fluctuation energy curve, the envelope width values ​​of all time points within the window are summed. The sum is divided by the window length to obtain the persistent disturbance intensity value corresponding to the center time point of the window. Then, the window is slid forward one time step along the time axis, and the above summation and averaging operations are repeated until the entire fluctuation energy curve is traversed. The persistent disturbance intensity values ​​of all time points are arranged chronologically to obtain a persistent disturbance intensity sequence. This sequence reflects the continuous accumulation state of wind speed fluctuation energy over time and can effectively distinguish between short-term disturbances and persistent fluctuations.

[0029] Step S215: Identify the stationary period of the residual fluctuation sequence. Mark the time period in the residual fluctuation sequence where the fluctuation value is continuously lower than the stationary determination boundary as a stationary segment. Record the time distance between the end time of each stationary segment and the start time of the next stationary segment as the interval duration. Take the frequency of the interval duration within the preset backtracking interval as the disturbance interval frequency at the corresponding time point. Arrange them in time to obtain the disturbance interval frequency sequence.

[0030] The stationarity determination boundary is a pre-defined threshold for fluctuation values. This threshold is determined through statistical distribution analysis of the residual fluctuation sequence. Specifically, the average absolute value of all fluctuation values ​​in the residual fluctuation sequence is calculated, and this average is multiplied by a pre-defined stationarity tolerance coefficient, which is pre-calibrated based on statistical analysis of historical stable wind periods. The residual fluctuation sequence is scanned point-by-point along the time axis, and the absolute value of the fluctuation value at each time point is compared with the stationarity determination boundary. If the absolute value of the fluctuation value is less than the stationarity determination boundary, that time point is marked as a stationary point. Temporally adjacent stationary points are grouped into the same stationary segment, and the start and end timestamps of each stationary segment are recorded. All identified stationary segments are traversed. For two adjacent stationary segments, the time distance obtained by subtracting the end timestamp of the previous stationary segment from the start timestamp of the latter segment is the interval duration. For a target time point, a pre-defined backtracking interval is traced backward from that time point. The total number of all intermittent durations occurring within this backtracking interval is counted, and this total number is taken as the disturbance intermittent frequency at the target time point. The disturbance intermittent frequencies of all time points are arranged in chronological order to obtain a disturbance intermittent frequency sequence. This sequence quantifies the frequency with which wind conditions switch back and forth between stable and disturbed states. A higher intermittent frequency indicates more unstable wind conditions and more frequent interruptions to stable periods.

[0031] Step S216: Align and integrate the instantaneous disturbance amplitude sequence, the continuous disturbance intensity sequence, and the disturbance intermittent frequency sequence on the time axis, and eliminate the time offset introduced by the detection delay based on the original timestamp of the wind speed time series component to generate wind speed disturbance pattern features.

[0032] Because the instantaneous disturbance amplitude sequence, persistent disturbance intensity sequence, and disturbance intermittent frequency sequence may introduce different time offsets during their respective calculation processes due to operations such as sliding windows, interpolation, and smoothing, there may be slight misalignments among the three on the time axis. Therefore, alignment and integration processing is required. The alignment and integration process uses the original timestamp sequence of the wind speed time series components as the reference axis. For each of the three sequences, the values ​​at each time point are mapped to the nearest original timestamp position on the reference time axis according to their own timestamps. If multiple values ​​correspond to a reference timestamp position, the average value is taken; if a reference timestamp position is missing a value, linear interpolation is used to fill it in using two adjacent valid values. After mapping and filling, each of the three sequences has a one-to-one corresponding value at each original timestamp position. The instantaneous disturbance amplitude value, persistent disturbance intensity value, and disturbance intermittent frequency value at the same timestamp are combined into a three-dimensional feature vector. The three-dimensional feature vectors of all timestamps, arranged in chronological order, constitute the wind speed disturbance model features. This feature fully preserves the temporal evolution information of wind speed disturbances across three dimensions: transient impact intensity, sustained fluctuation energy, and intermittent switching frequency.

[0033] Step S220: Input the instantaneous disturbance amplitude sequence, the continuous disturbance intensity sequence, and the disturbance intermittent frequency sequence into the disturbance mode encoding to generate a disturbance type distribution vector. The disturbance type distribution vector indicates the degree to which the current wind speed disturbance tends to be an isolated sudden disturbance, a continuous fluctuation, or an intermittent impact.

[0034] The perturbation pattern encoding is an encoder module with classification mapping capabilities. This module consists of a fully connected feedforward neural network. Its input layer receives a three-dimensional vector composed of the values ​​of three feature sequences at the current time step. The hidden layer contains two layers of neurons; the first and second hidden layers use a linear rectified function as their activation function. The output layer contains three neurons, corresponding to three perturbation types: isolated sudden disturbance, continuous fluctuation, and intermittent impact. The output layer uses a normalized exponential function as its activation function, compressing the output values ​​of the three neurons into three probability values ​​between 0 and 1, summing to 1. These three probability values ​​represent the degree to which the current wind speed perturbation belongs to the isolated sudden disturbance type, the continuous fluctuation type, and the intermittent impact type, respectively. These three probability values ​​together constitute the perturbation type distribution vector. This fully connected feedforward neural network uses historical data samples with manually labeled wind speed disturbance types for supervised learning during the training phase. The loss function is the classification cross-entropy loss function, and the optimizer is the adaptive moment estimation optimizer. Through repeated iterative learning of historical samples, the encoder can accurately map the corresponding disturbance type tendency distribution based on the three input disturbance feature values.

[0035] Step S230: Based on the perturbation type with the highest tendency in the perturbation type distribution vector, determine the adjustment strategy tendency of the forgetting gate dynamic value. When the perturbation type is isolated sudden perturbation, the tendency is to suppress the forgetting amplitude to retain the historical state. When the perturbation type is intermittent impact, the tendency is to amplify the forgetting amplitude to accelerate the state update.

[0036] In one implementation, step S230 specifically includes the following steps S231 to S236: Step S231: Perform a primary type determination on the disturbance type distribution vector, compare the magnitudes of the isolated sudden disturbance tendency value, the continuous fluctuation tendency value, and the intermittent impact tendency value in the disturbance type distribution vector, mark the type with the largest tendency value as the primary disturbance type, and mark the other two types as auxiliary disturbance types.

[0037] In step S231, the primary type determination is implemented using a three-input comparator logic. The three tendency values ​​from the disturbance type distribution vector are fed into the three inputs of the comparator. The comparator performs pairwise comparisons sequentially, first comparing the isolated sudden disturbance tendency value with the continuous fluctuation tendency value, then comparing the larger one with the intermittent impact tendency value, and finally outputting the disturbance type identifier corresponding to the maximum value among the three tendency values. This maximum value's disturbance type identifier is written into the primary disturbance type register, and the remaining two disturbance type identifiers are written into the first auxiliary disturbance type register and the second auxiliary disturbance type register, respectively, completing the primary type determination.

[0038] Step S232: When the main control disturbance type is marked as isolated sudden disturbance, extract the instantaneous disturbance amplitude corresponding to the current time step from the instantaneous disturbance amplitude sequence, and determine whether the instantaneous disturbance amplitude is within the preset isolated sudden disturbance response sensitive range. If it is within the sensitive range, set the adjustment strategy tendency direction to the forgetting suppression direction. If it is not within the sensitive range, revert the adjustment strategy tendency direction to the neutral direction.

[0039] The sensitive interval for isolated sudden disturbance response is a pre-defined range of amplitude values. The lower bound of this interval is determined by the minimum statistical value of the instantaneous disturbance amplitude that has an observable mechanical effect on the conductor during historical wind speed abrupt changes, while the upper bound is determined by the historical highest observed value in the instantaneous disturbance amplitude sequence. When the dominant disturbance type is marked as an isolated sudden disturbance, the instantaneous disturbance amplitude value corresponding to the current time step is read from the instantaneous disturbance amplitude sequence according to the time index, and this value is compared with the lower and upper bounds of the sensitive interval for isolated sudden disturbance response. If the instantaneous disturbance amplitude falls within the sensitive range, the adjustment strategy will be biased towards the forgetting suppression direction. The forgetting suppression direction means that the dynamic value of the forgetting gate will be reduced in the future, so that the network retains more historical hidden state information to resist the pollution of the state estimation by a single instantaneous disturbance. If the instantaneous disturbance amplitude is lower than the lower bound of the sensitive range, it means that the disturbance amplitude is not enough to have a substantial impact on the wire state, and the adjustment strategy will be biased back to the neutral direction. If the instantaneous disturbance amplitude is higher than the upper bound of the sensitive range, it means that the disturbance amplitude is abnormally severe, and this case will be responded to according to the intermittent impact type processing logic.

[0040] Step S233: When the master disturbance type is marked as intermittent impact type, extract the disturbance interval frequency of the current time step from the disturbance interval frequency sequence, compare the disturbance interval frequency with the intermittent impact frequency reference value, if the disturbance interval frequency is greater than the intermittent impact frequency reference value, then set the adjustment strategy tendency direction to forget acceleration direction, otherwise maintain the adjustment strategy tendency direction of the previous time step.

[0041] The intermittent impact frequency reference value is a frequency reference value calibrated based on the statistical characteristics of historical intermittent strong wind periods. When the main control disturbance type is marked as intermittent impact, the disturbance intermittent frequency value corresponding to the current time step is read from the disturbance intermittent frequency sequence according to the time index. The disturbance intermittent frequency value is compared with the intermittent impact frequency reference value. If the disturbance intermittent frequency value is greater than the intermittent impact frequency reference value, it indicates that the current wind condition is in a high-frequency intermittent switching state, and the historical hidden state information is no longer able to effectively represent the current wind condition pattern. The adjustment strategy tendency direction is set to the forgetting acceleration direction. The forgetting acceleration direction means that the forgetting gate dynamic value will be increased in the future, so that the network can discard old state information more quickly to adapt to the rapidly switching wind conditions. If the disturbance intermittent frequency value is not greater than the intermittent impact frequency reference value, it indicates that the intermittent switching frequency has not yet reached the level that requires accelerated forgetting. The adjustment strategy tendency direction determined in the previous time step remains unchanged to ensure the continuity of the control strategy.

[0042] Step S234: When the master disturbance type is marked as continuous fluctuation type, extract the continuous disturbance intensity of the current time step from the continuous disturbance intensity sequence, compare the continuous disturbance intensity with the continuous fluctuation intensity reference value, and if the continuous disturbance intensity is greater than the continuous fluctuation intensity reference value, set the adjustment strategy tendency direction to forget fine-tuning direction, and limit the adjustment step size of the forget gate dynamic value to a narrow step size range.

[0043] The sustained fluctuation intensity reference value is an intensity reference value calibrated based on the median of the energy distribution of sustained wind periods in history. When the dominant disturbance type is marked as sustained fluctuation, the sustained disturbance intensity value corresponding to the current time step is read from the sustained disturbance intensity sequence according to the time index. The sustained disturbance intensity value is compared with the sustained fluctuation intensity reference value. If the sustained disturbance intensity value is greater than the sustained fluctuation intensity reference value, it indicates that the current wind speed fluctuation is in a strong and sustained state. In this state, it is neither necessary to significantly suppress forgetting nor to significantly accelerate forgetting, but a more moderate adjustment method should be adopted. Therefore, the adjustment strategy tendency direction is set to the forgetting fine-tuning direction. When the forgetting fine-tuning direction is executed, the adjustment step size of the forgetting gate dynamic value is limited to a preset narrow step size range. The absolute values ​​of the upper and lower limits of this narrow step size range are much smaller than the normal adjustment step sizes corresponding to the forgetting suppression direction and the forgetting acceleration direction, so as to ensure the smoothness of the adjustment process and avoid network state oscillations caused by over-adjustment. If the sustained disturbance intensity value is not greater than the sustained fluctuation intensity reference value, the adjustment strategy tendency direction of the previous time step is also maintained unchanged.

[0044] Step S235: Obtain the tendency value of the auxiliary disturbance type and its direction of change. When the tendency value of the auxiliary disturbance type shows a unidirectional increasing trend in a continuous time step and the auxiliary disturbance type and the main disturbance type are conflict types, generate a conflict warning signal of the adjustment strategy tendency, and weaken the execution strength of the current adjustment strategy tendency according to the conflict warning signal.

[0045] The tendency value and direction of change of auxiliary disturbance types are determined by calculating the first-order difference sign sequence of the tendency value of the auxiliary disturbance type over several consecutive time steps. The identifiers of the two auxiliary disturbance types are read from the first and second auxiliary disturbance type registers. The tendency values ​​of these two auxiliary disturbance types over the current time step and several previous historical time steps are extracted from the disturbance type distribution vector. For each auxiliary disturbance type, the difference between adjacent time steps is calculated sequentially along the time axis. If the signs of all consecutive differences are positive, the tendency value of that auxiliary disturbance type is determined to show a unidirectional increasing trend. Conflict types are disturbance types that are opposed to the main disturbance type in the forgetting gate adjustment direction. Isolated sudden disturbances and intermittent impacts are conflict types because the former requires suppressing forgetting while the latter requires accelerating forgetting. Continuous fluctuations are not conflicting with isolated sudden disturbances or intermittent impacts. When a unidirectional increasing trend is detected in the tendency value of an auxiliary disturbance type and it is a conflict type with the current main disturbance type, a conflict warning signal is generated. The conflict warning signal carries a conflict weighting factor, the value of which is positively correlated with the increment of the auxiliary disturbance type tendency value. The execution intensity of the current adjustment strategy tendency is weakened based on the conflict weighting factor. Specifically, the adjustment amount corresponding to the adjustment strategy tendency determined in step S232, S233, or S234 is multiplied by a discount coefficient less than 1. The larger the conflict weighting factor, the smaller the discount coefficient, thus weakening the execution intensity and preventing network state jumps due to abrupt changes in the adjustment strategy tendency during wind condition transitions.

[0046] Step S236: Perform a continuity fusion of the adjustment strategy tendency after the execution intensity is weakened with the adjustment strategy tendency of the previous time step to generate the adjustment strategy tendency of the current time step, and output it to the inertial connection processing step.

[0047] Tendency continuity fusion can be achieved through first-order exponential smoothing recursion. The original adjustment value corresponding to the adjustment strategy tendency after weakening execution intensity, and the historical adjustment value corresponding to the adjustment strategy tendency finally executed in the previous time step, are recorded. The original adjustment value is multiplied by a smoothing coefficient between 0 and 1, and the historical adjustment value is multiplied by (1 - the smoothing coefficient). The sum of these two values ​​yields the final adjustment value of the adjustment strategy tendency for the current time step. This smoothing coefficient determines the relative proportion of influence of the current adjustment strategy tendency on the newly calculated value versus the historical value. A larger smoothing coefficient results in a more sensitive response to the newly calculated value, while a smaller smoothing coefficient indicates stronger historical continuity. The calculated final adjustment value and its corresponding adjustment direction are encapsulated as the adjustment strategy tendency for the current time step and output to the inertial connection processing step.

[0048] Step S240: Obtain the adjustment trajectory of the forget gate dynamic value output in the previous time step of the recursive processing layer. The adjustment trajectory records the trend of forget gate dynamic value changes in the past multiple time steps. Inertially connect the adjustment strategy tendency of the current time step with the adjustment trajectory to generate the forget gate dynamic value of the current time step.

[0049] The adjustment trajectory is a fixed-length first-in-first-out (FIFO) queue data structure that stores the historical records of forget gate dynamic values ​​from the most recent time steps. At each time step, after the recursive processing layer completes the forward computation, it pushes the forget gate dynamic value used for that time step to the tail of the queue, while simultaneously popping the earliest time step record from the head of the queue, keeping the queue length constant. The specific execution method of inertial connection is as follows: First, all historical forget gate dynamic values ​​in the adjustment trajectory queue are read, and the average of these historical values ​​is calculated as the forget gate inertial reference value. Simultaneously, the trend of change from old to new in the queue is calculated, i.e., whether the forget gate dynamic value shows an upward trend, a downward trend, or a stable trend in the most recent time steps, which is used as the inertial direction. The adjustment amount carried in the adjustment strategy tendency of the current time step output in step S236 is coordinated with the inertial direction. If the direction of the adjustment amount is the same as the inertial direction, the full amplitude of the adjustment amount is retained; if the direction of the adjustment amount is opposite to the inertial direction, the amplitude of the adjustment amount is attenuated according to the degree of reversal. The attenuation coefficient is jointly determined by the smoothing coefficient in the current adjustment strategy tendency and the fluctuation amplitude of the forget gate dynamic values ​​in the most recent time steps in the adjustment trajectory. The coordinated adjustment is superimposed on the forget gate inertial reference value, and then passed through a range pruner to restrict the result to within the legal range of the forget gate dynamic value, ultimately generating the forget gate dynamic value for the current time step. This value is the forget gate parameter actually used by the recursive processing layer when performing gating filtering at the current time step.

[0050] Step S250: Perform gated filtering on the cell state vector passed from the previous time step to the current time step using the forget gate dynamic value of the current time step, retain the historical information in the cell state vector that is compatible with the current wind condition pattern and discard the historical information that is not compatible, and obtain the filtered cell state vector.

[0051] The cell state vector is a continuous vector within the gated recurrent unit structure of the recursive processing layer, used to transfer long-term memory information between time steps. Its dimension is the same as the hidden state vector. Each element of the cell state vector corresponds to a memory channel, storing the long-range dependency information accumulated in the previous time step. After the recursive processing layer of the previous time step completes its forward computation, it passes its cell state vector to the current time step. The gated filtering operation uses the forget gate dynamic value of the current time step as the coefficient for element-wise multiplication, performing a Hadamard product operation with the cell state vector passed from the previous time step. Each element of the forget gate dynamic value is multiplied by the element of the corresponding channel in the cell state vector, and the product constitutes the filtered cell state vector. Since each element in the forget gate dynamic value takes a value between 0 and 1, when the forget gate dynamic value of a channel is close to 1, the historical memory of the corresponding channel is almost completely preserved; when the forget gate dynamic value of a channel is close to 0, the historical memory of the corresponding channel is almost entirely discarded. The forget gate dynamic value is dynamically modulated and generated by the wind speed disturbance pattern characteristics of the current wind condition. Therefore, it can adaptively retain historical state information channels that match the current wind condition pattern, while discarding those historical state information channels that have become invalid under the current wind condition.

[0052] Step S260: Generate the hidden state vector for the current time step based on the filtered cell state vector and the micro-meteorological time series data input to the recursive processing layer at the current time step. Output the hidden state vector to the subsequent processing layer of the conductor state inference neural network and copy it to the recursive processing layer at the next time step.

[0053] The hidden state vector is generated by the remaining gated computation path of the gated recurrent unit structure in the recursive processing layer. At the current time step, the wind speed, temperature, and humidity time-series components from the current time-step meteorological time-series data are concatenated into an input vector. This input vector is then concatenated with the hidden state vector passed from the previous time step to the current time step, resulting in a joint input vector. This joint input vector is then fed into the update gate computation path and the candidate state computation path of the gated recurrent unit structure, respectively. The update gate computation path consists of a fully connected layer followed by a logistic activation function, outputting an update gate vector. Each element of this vector takes a value between 0 and 1, controlling the interpolation ratio between the new and old states. The candidate state computation path consists of a fully connected layer followed by a hyperbolic tangent activation function, outputting a candidate state vector. Each element of this vector takes a value between -1 and 1, representing possible new state information at the current time step. The filtered cell state vector and candidate state vector are linearly interpolated element-wise based on the updated gate vector. The interpolated result is the cell state vector at the current time step. This interpolated result is then passed through an output gate to generate the hidden state vector at the current time step. The output gate consists of a fully connected layer with a joint input vector followed by a logistic activation function. The output gate vector and the cell state vector at the current time step, after hyperbolic tangent activation, are subjected to a Hadamard product to obtain the hidden state vector. This hidden state vector is output to the subsequent processing layer of the conductor state inference neural network to map and generate conductor mechanical state parameters. It is also copied and passed to the recursive processing layer at the next time step as input for historical hidden state information, thus maintaining the continuity of the temporal recursive computation.

[0054] Step S300: Obtain the conductor and ground wire state estimation neural network directly outputting the conductor and ground wire mechanical state parameters based on the input micro-meteorological time series data. The conductor and ground wire mechanical state parameters include conductor ice thickness characterization and conductor wind deflection angle characterization. The conductor ice thickness characterization reflects the continuous change in the degree of ice accumulation on the conductor surface, and the conductor wind deflection angle characterization reflects the change in the deviation angle of the conductor under wind load.

[0055] In one implementation, step S300 specifically includes the following steps S310 to S360: Step S310: Extract the hidden state vector sequence generated by the recursive processing layer inside the conductor state estimation neural network at all time steps covered by the micro-meteorological time series data. Each hidden state vector in the hidden state vector sequence is a fusion representation of the micro-meteorological information and historical state information at the corresponding time step.

[0056] The extraction of the hidden state vector sequence is accomplished by acquiring signals at the output port of the recursive processing layer at each time step. During the forward inference of the micro-meteorological time-series data by the conductor-to-ground wire state inference neural network, the recursive processing layer performs recursive calculations step by step, starting from the first time step. After processing each time step, a hidden state vector is generated at the output port of that time step and cached in an internal buffer. After all time steps covered by the micro-meteorological time-series data have been processed, the internal buffer of the recursive processing layer stores a number of hidden state vectors equal to the number of time steps. These hidden state vectors are retrieved and arranged sequentially according to the time step order to obtain the hidden state vector sequence. Each hidden state vector in this sequence is a fusion representation of the current micro-meteorological input information at the corresponding time step and all historical state information before that time step. The fusion is achieved through cell state vector recursion and gating mechanisms within the recursive processing layer. The hidden state vectors encode the dynamic pattern characteristics of the micro-meteorological environment of the power transmission corridor over time.

[0057] Step S320: Input the hidden state vector sequence into the conductor state mapping structure connected to the output of the recursive processing layer. The conductor state mapping structure performs dimensional projection transformation on the hidden state vector at each time step to generate conductor icing-related projection components and conductor wind deflection-related projection components. The conductor icing-related projection components and conductor wind deflection-related projection components are generated by mutually independent branch mapping channels.

[0058] The conductor state mapping structure is a dual-branch fully connected mapping module connected after the output of the recursive processing layer. This module contains two structurally identical but parameter-independent branch mapping channels that work in parallel, respectively mapping the hidden state vector into conductor icing-related projection components and conductor wind-related projection components. Each branch mapping channel consists of a fully connected layer whose input dimension is equal to the dimension of the hidden state vector, and whose output dimension is a pre-defined projection dimension value, which is smaller than the input dimension. This projection dimension value is used to compress and project the high-dimensional hidden state vector onto a low-dimensional feature space related to the corresponding physical quantity. The conductor icing-related projection component is a low-dimensional vector output by the fully connected layer of the first branch mapping channel after performing matrix multiplication and bias addition operations on the hidden state vector. Each element in this vector corresponds to the activation intensity of different latent physical factors related to the conductor icing process. The conductor wind deflection-related projection component is a low-dimensional vector output by the fully connected layer of the second branch mapping channel after performing matrix multiplication and bias addition operations on the hidden state vector. Each element in this vector corresponds to the activation intensity of different latent physical factors related to the conductor wind deflection response. The parameters of the two branch mapping channels are independently optimized through backpropagation during the training phase, without sharing them, allowing each mapping channel to focus on extracting feature representations related to icing and wind deflection respectively.

[0059] Step S330: Perform temporal memory accumulation on the icing-related projection components of the conductor, maintain an icing accumulation memory chain, and perform directional recursive superposition of the icing-related projection components of the conductor at each time step along the icing accumulation memory chain to generate a sequence of conductor icing thickness characterization quantities.

[0060] In one implementation, step S330 specifically includes the following steps S331 to S336: Step S331: Initialize the icing accumulation memory chain. The accumulated state quantity of the icing accumulation memory chain at the start time step is the preset icing baseline quantity.

[0061] The icing accumulation memory chain is a recursive state register specifically designed to track the continuous accumulation of conductor icing thickness. This register stores a scalar value representing the total accumulated conductor icing thickness up to the current time step. Before processing begins at the first time step covered by the micro-meteorological time-series data, this register is initialized by assigning the stored accumulated state value to a preset icing reference value. The icing reference value is an initial reference value calibrated based on the baseline state of the transmission line under icing-free conditions. This initial reference value typically corresponds to the state characterization when the conductor surface is icy or has only minimal icing. The initialization operation ensures that the icing accumulation memory chain has a defined and reasonable initial state at the start of the time-series recursion.

[0062] Step S332: Read the icing-related projection component of the conductor at the current time step, and extract the temperature time series component corresponding to the current time step from the micro-meteorological time series data. Generate a temperature modulation coefficient based on the temperature time series component, and scale the amplitude of the icing-related projection component of the conductor using the temperature modulation coefficient to obtain the temperature effect corrected projection component.

[0063] As one implementation method, step S332 specifically includes the following steps S3321 to S3326: Step S3321: Read the temperature time series component of the current time step from the micro-meteorological time series data, extract the historical temperature sequence of the temperature time series component within the preset backtracking interval, and arrange the historical temperature sequence in chronological order as a temperature evolution trajectory.

[0064] The temperature evolution trajectory is a one-dimensional sequence formed by tracing back a specified time interval from the current time step, and arranging the temperature time series components of all time steps within that interval in chronological order. The length of this specified time interval is pre-set based on the time lag characteristics of the temperature's impact on the icing process, typically selecting the length of several consecutive time steps to ensure the temperature evolution trajectory covers the complete response cycle of temperature changes to icing. The extraction operation is implemented using a sliding window mechanism. A fixed-length temperature history buffer is maintained at each time step. Upon reaching a new time step, the new temperature time series component is pushed to the end of the buffer, and the oldest record at the head of the buffer is popped. The data retained in the buffer is the historical temperature sequence within the preset backtracking interval corresponding to the current time step.

[0065] Step S3322: Divide the temperature evolution trajectory into temperature change stages, identify the heating segment, cooling segment and constant temperature segment in the temperature evolution trajectory, extract the stage type label of the current time step in the temperature evolution trajectory, and record the duration of the current stage.

[0066] The temperature change stages are divided by analyzing the first-order difference numerical sequence of the temperature evolution trajectory. The temperature differences between adjacent time steps in the trajectory are calculated sequentially to obtain the first-order difference sequence. A positive threshold for determining temperature rise and a negative threshold for determining temperature fall are set. When the first-order difference value is greater than the positive threshold for temperature rise for several consecutive time steps, these time steps are marked as a temperature rise stage; when the first-order difference value is less than the negative threshold for temperature fall for several consecutive time steps, these time steps are marked as a temperature fall stage; when the absolute value of the first-order difference value does not exceed a small threshold for determining isothermal temperature for several consecutive time steps, these time steps are marked as an isothermal stage. After traversing the entire temperature evolution trajectory and assigning labels to each time step, the position of the current time step in the trajectory is read, and the stage type label assigned to that position is extracted. This stage type label can be one of the following: temperature rise stage, temperature fall stage, or isothermal stage. At the same time, by tracing back along the temperature evolution trajectory from the current time step, the number of consecutive time steps with the same stage type label as the current time step is counted, and this number is the duration of the current stage.

[0067] Step S3323: Analyze the degree of temperature fluctuation on the temperature evolution trajectory and generate a temperature fluctuation activity characterization quantity. The temperature fluctuation activity characterization quantity reflects the frequency and range of temperature value fluctuations in the historical temperature sequence.

[0068] Temperature fluctuation analysis is achieved, for example, by calculating the local fluctuation statistics of the temperature evolution trajectory. The temperature evolution trajectory is divided into several overlapping short-time analysis windows, each containing a fixed number of continuous time steps. The sample standard deviation is calculated for the temperature values ​​within each short-time analysis window; this standard deviation reflects the dispersion of temperature values ​​within that short period. The average standard deviation of all short-time analysis windows is then taken as the mean temperature fluctuation standard deviation. Simultaneously, a first-order differencing operation is performed on the temperature evolution trajectory, calculating the proportion of times the signs of two adjacent difference values ​​change relative to the total number of time steps. This proportion reflects the frequency of reversals in the direction of temperature change. The mean temperature fluctuation standard deviation and the reversal proportion are then weighted and summed. The weighting coefficients are pre-determined based on the sensitivity analysis of historical temperature data to the impact of icing. The result of the weighted sum is the temperature fluctuation activity characterization quantity. A higher value for the temperature fluctuation activity characterization quantity indicates more frequent and wider fluctuations in temperature values ​​within the historical temperature series.

[0069] Step S3324: Input the stage type label, the duration of the current stage, and the temperature fluctuation activity characterization quantity into the temperature modulation decision. The temperature modulation decision process outputs the modulation tendency type, which includes one of the following: enhancement tendency, suppression tendency, and neutral maintenance tendency.

[0070] Temperature modulation decision-making is implemented by a rule-based reasoning logic module. This module internally maintains a modulation tendency reasoning rule table. The rule table uses the stage type label, the discretization level of the current stage's duration, and the discretization level of the temperature fluctuation activity representation as conditional attributes, and the modulation tendency type as the decision attribute. During reasoning, the stage type label obtained in step S3322, the duration level obtained in step S3322 after discretization mapping of the current stage's duration, and the activity level obtained in step S3323 after discretization mapping of the temperature fluctuation activity representation are combined into a conditional tuple. This conditional tuple is then matched against the conditional parts of each rule in the modulation tendency reasoning rule table. Upon successful matching, the decision attribute value corresponding to the matching rule is read. This decision attribute value is the modulation tendency type, which can be one of three: enhancement tendency, suppression tendency, or neutral maintenance tendency. An enhancing tendency indicates that temperature conditions are favorable for ice growth or maintenance, requiring the generation of amplification-type temperature modulation coefficients in the subsequent process; an inhibiting tendency indicates that temperature conditions are unfavorable for ice maintenance, and ice may melt, requiring the generation of attenuation-type temperature modulation coefficients; a neutral maintenance tendency indicates that temperature conditions have no significant effect on ice changes, requiring the generation of neutral temperature modulation coefficients.

[0071] Step S3325: Determine the modulation coefficient generation path according to the modulation tendency type. When the modulation tendency type is enhancement tendency, generate amplification-type temperature modulation coefficients along the enhancement mapping path. When the modulation tendency type is suppression tendency, generate attenuation-type temperature modulation coefficients along the suppression mapping path. When the modulation tendency type is neutral maintenance tendency, generate neutral temperature modulation coefficients.

[0072] The enhanced mapping path can be implemented using an amplification factor generation function. This function takes the duration of the current stage and the temperature fluctuation activity level as input variables and outputs an amplification-type temperature modulation factor greater than 1. This generation function uses a two-dimensional lookup table structure. The row index of the lookup table represents the discretization level of the duration of the current stage, and the column index represents the discretization level of the temperature fluctuation activity level. Each cell in the table pre-stores the corresponding amplification-type temperature modulation factor value. These values ​​are pre-calibrated through statistical analysis of temperature conditions and icing rate in historical icing events. During a query, the corresponding cell in the lookup table is located based on the duration level and activity level of the current time step, and the value is read as the amplification-type temperature modulation factor. The suppressed mapping path is implemented using a decay factor generation function. This function also uses a two-dimensional lookup table structure, with the same row and column index definitions as the enhanced mapping path. However, the values ​​stored in the table are all decay-type temperature modulation factor values ​​less than 1. These values ​​are pre-calibrated through statistical analysis of temperature conditions and melting rate in historical icing melting events. When the neutrality is maintained, the generated neutral temperature modulation coefficient is always 1, indicating that no amplitude scaling is applied to the icing-related projection components of the conductor.

[0073] Step S3326: Combine the generated temperature modulation coefficient with the projection component related to conductor icing at the current time step by amplitude scaling to obtain the projection component after temperature effect correction, and output the projection component after temperature effect correction to the step of damp heat co-correction.

[0074] Amplitude scaling is a process of multiplying the temperature modulation coefficient by each element of the conductor icing-related projected component. The temperature modulation coefficient is a single scalar value. Multiplying this scalar value by the entire vector of the conductor icing-related projected component yields a product vector where each element is the corresponding element in the original projected component scaled by the temperature modulation coefficient. This resulting vector constitutes the temperature-corrected projected component. This element-by-element scaling process preserves the relative proportions between elements within the projected component, adjusting only its overall amplitude, thus injecting the influence of temperature conditions on the icing process into the projected component through amplitude modulation.

[0075] Step S333: Read the humidity time series component corresponding to the current time step in the micro-meteorological time series data, generate a humidity modulation coefficient based on the humidity time series component, and perform amplitude scaling again on the temperature effect-corrected projection component using the humidity modulation coefficient to obtain the humidity-heat co-corrected projection component.

[0076] The humidity modulation coefficient can be generated using a piecewise linear mapping method. A humidity reference saturation value is preset, representing the relative humidity value corresponding to the saturation state of water vapor content in the ambient air. The humidity time-series component value at the current time step is compared with this humidity reference saturation value. If the humidity time-series component value is greater than or equal to the humidity reference saturation value, the humidity modulation coefficient is set to a preset maximum amplification factor. If the humidity time-series component value is less than the humidity reference saturation value but greater than a preset low humidity threshold, the humidity modulation coefficient is interpolated according to the linear mapping relationship between the humidity time-series component values ​​from the low humidity threshold to the humidity reference saturation value. The calculation method is to start with the minimum modulation coefficient corresponding to the low humidity threshold and end with the maximum amplification factor, linearly determining the corresponding modulation coefficient value based on the relative position of the humidity time-series component value within the interval. If the humidity time-series component value is not greater than the low humidity threshold, the humidity modulation coefficient is set to the minimum modulation coefficient, which is a positive number less than 1, representing the inhibitory effect of the low humidity environment on icing. The generated humidity modulation coefficient is multiplied term by term by each element in the temperature-corrected projection component to obtain the humidity- and heat-coordinated corrected projection component. This process further integrates the influence of humidity conditions on top of the temperature correction, achieving coordinated regulation of the icing process by temperature and humidity.

[0077] Step S334: Read the accumulated state quantity of the previous time step from the icing accumulation memory chain, and perform directional recursive superposition of the projected component after wet and heat co-correction with the accumulated state quantity of the previous time step. The directional recursive superposition process restricts the direction of decrease of the state quantity according to the unidirectional accumulation characteristics of the icing process.

[0078] The core of directed recursive superposition is to selectively accumulate the projected components after thermo-humidity co-correction. First, the algebraic sum of all elements in the projected components after thermo-humidity co-correction is calculated. This algebraic sum represents the value of the increase or decrease in icing change due to micrometeorological conditions at the current time step. The accumulated state value stored in the previous time step is read from the register of the icing accumulation memory chain, and the accumulated state value of the previous time step is added to the value of the icing change at the current time step. The unidirectional accumulation characteristic of the icing process means that the conductor icing thickness will not decrease negatively beyond the existing icing amount under natural conditions. The reduction in icing thickness can only melt existing icing and cannot make the icing thickness negative. In directional recursive superposition, directional constraints are imposed on the summation result. Specifically, the sign of the icing change value is determined. If the icing change value is positive, it indicates that the current conditions are favorable for icing growth, and normal accumulation is allowed. If the icing change value is negative, it indicates that the current conditions are favorable for icing melting. Then, the difference between the cumulative state quantity of the previous time step and the absolute value of the icing change value is calculated. If the difference is non-negative, normal decay is allowed. If the difference is negative, it indicates that the melting amount exceeds the existing accumulation amount. In this case, the summation result is forcibly truncated to zero so that the state quantity will not cross the zero value due to unidirectional melting.

[0079] Step S335: Perform lower bound truncation on the superimposed state quantity. When the superimposed state quantity is lower than the icing reference quantity, restore the state quantity to the icing reference quantity to obtain the conductor icing thickness characterization quantity at the current time step.

[0080] The lower bound truncation operation can be implemented through a numerical comparison and conditional assignment logic. The state quantity obtained by the directed recursive superposition in step S334 is compared with the icing reference quantity. The value of the icing reference quantity is the same as the value used when initializing the icing accumulation memory chain in step S331. If the superimposed state quantity is greater than or equal to the icing reference quantity, the superimposed state quantity remains unchanged; if the superimposed state quantity is less than the icing reference quantity, the value of the state quantity is forcibly replaced with the icing reference quantity. The state quantity after the lower bound truncation process is the conductor icing thickness representation quantity at the current time step. This representation quantity ensures that the conductor icing thickness is not lower than the reference state at any time step, avoiding physically unreasonable negative icing or icing thickness representations lower than the reference.

[0081] Step S336: Write the conductor icing thickness characterization value of the current time step into the icing accumulation memory chain as the accumulated state value of the next time step, and arrange the conductor icing thickness characterization values ​​of each time step in chronological order to generate a conductor icing thickness characterization value sequence.

[0082] The write operation overwrites the existing value in the icing accumulation memory chain register with the current time step's conductor icing thickness characterization value calculated in step S335. This value will be read and used as the accumulated state value of the previous time step in step S334 of the next time step. After the conductor icing thickness characterization value is calculated at each time step, the characterization value is associated with the corresponding timestamp and stored in an output buffer. After all time steps covered by the micrometeorological time series data have completed the above recursive calculation, all conductor icing thickness characterization values ​​sorted by timestamp in the output buffer constitute the conductor icing thickness characterization value sequence.

[0083] Step S340: Perform time-series response dynamics on the wind deflection-related projection components of the conductor, and simulate the inertial hysteresis characteristics of the conductor wind deflection angle response process by combining the wind speed time-series components of the corresponding time step in the micro-meteorological time-series data, and generate a sequence of conductor wind deflection angle characterization quantities.

[0084] Dynamic time-series response is a mathematical simulation of the mechanical inertial hysteresis characteristics exhibited by a conductor when it undergoes wind-induced angular displacement under wind load. The wind-induced angular response of the conductor is not instantaneous but exhibits a response time lag determined by the conductor's mass and mechanical damping. This simulation is implemented using a first-order inertial response filter. At each time step, the filter receives the conductor's wind-induced angular displacement related projection component as the static wind-induced angular displacement driving force for the current time step. Simultaneously, it dynamically adjusts the filter's response time parameters based on the wind speed time-series component of the corresponding time step. The filter's state recursion calculation process is as follows: the conductor's wind-induced angular displacement characteristic output by the filter at the previous time step is read as the initial state value; the difference between the static wind-induced angular displacement driving force at the current time step and the output value at the previous time step is calculated; this difference is multiplied by a response rate coefficient dynamically determined by the wind speed time-series component; and the product is added to the output value at the previous time step to obtain the output value at the current time step. The response rate coefficient increases with the increase of the wind speed time-series component value, indicating that the conductor's wind deflection response reaches steady state faster at high wind speeds and is more sluggish at low wind speeds. The value output by this first-order inertial response filter at each time step is the conductor wind deflection angle characterization quantity for the corresponding time step. The conductor wind deflection angle characterization quantities for all time steps are arranged in chronological order to form a conductor wind deflection angle characterization quantity sequence.

[0085] Step S350: Register the conductor icing thickness characterization sequence and the conductor wind deflection angle characterization sequence according to their respective associated timestamps, so that the conductor icing thickness characterization and conductor wind deflection angle characterization at the same timestamp constitute a state pair.

[0086] Temporal registration is achieved through a database join operation using timestamps as the association key. Each record in the conductor icing thickness characterization sequence contains a timestamp field and an icing thickness characterization value field, and each record in the conductor wind deflection characterization sequence contains a timestamp field and a wind deflection value field. During temporal registration, the two sequences are joined equi-valued according to the timestamp field. That is, for each timestamp, if a record corresponding to that timestamp exists in both sequences, the two records are merged into one record. This record contains both the conductor icing thickness characterization value and the conductor wind deflection value for that timestamp. The merged record is a state pair. If, due to computational delays or data gaps, only one sequence contains a record for a certain timestamp, nearest neighbor timestamp matching is used to supplement it, selecting the record value with the closest timestamp in the missing sequence for pairing. After all timestamps are registered, all state pairs are sorted in ascending order of timestamp, forming a complete set of temporal state pairs.

[0087] Step S360: Organize the state pairs into conductor and ground wire mechanical state parameters according to the timestamp order, and output the conductor and ground wire mechanical state parameters to the transmission corridor physical state recursive model.

[0088] The mechanical state parameters of the conductor and ground wire are organized as follows: the conductor icing thickness and conductor wind deflection angle characteristics obtained from registration in step S350, along with the corresponding timestamps of each state pair, are combined into a structured data unit. All data units are arranged in ascending order of timestamps to form a sequence of conductor and ground wire mechanical state parameters. This sequence of conductor and ground wire mechanical state parameters is packaged as a whole and output to the input end of the transmission corridor physical state recursive model through memory data transfer or inter-process communication interface.

[0089] Step S400: Input the conductor and ground wire mechanical state parameters into the pre-constructed transmission corridor physical state recursive model. The transmission corridor physical state recursive model performs a physical mechanism-driven recursive estimation of the conductor and ground wire mechanical state at the current moment based on the conductor and ground wire mechanical state parameters at the previous moment and the micro-meteorological time series data at the current moment, generating the recursively estimated conductor and ground wire mechanical state parameters. When the transmission corridor physical state recursive model is running, it is simultaneously subject to the constraints of ice and heat balance and wind deflection moment balance.

[0090] In one implementation, step S400 specifically includes the following steps S410 to S460: Step S410: Extract the conductor icing thickness and conductor wind deflection from the conductor mechanical state parameters of the previous time step as the recursive initial state of the transmission corridor physical state recursive model.

[0091] The conductor icing thickness characteristic of the previous time step is the conductor icing thickness characteristic value corresponding to the time step immediately preceding the current time step in the conductor-ground wire mechanical state parameter sequence; the conductor wind deflection characteristic of the previous time step is the conductor wind deflection characteristic value corresponding to the time step immediately preceding the current time step in the conductor-ground wire mechanical state parameter sequence. The extraction operation is completed through time index positioning. After the timestamp of the current time step is determined, the record with the smallest difference between its timestamp and the timestamp of the current time step is searched in the conductor-ground wire mechanical state parameter sequence. The conductor icing thickness characteristic field and the conductor wind deflection characteristic field of this record are read. The two read values ​​are written into the state register of the transmission corridor physical state recursive model as the initial state for subsequent recursive branches.

[0092] Step S420: Input the wind speed time series component, temperature time series component, and humidity time series component from the micro-meteorological time series data of the current time step into the recursive model of the physical state of the transmission corridor. The icing recursive branch of the recursive model of the physical state of the transmission corridor calculates the recursive estimate of the conductor icing thickness of the current time step based on the conductor icing thickness characterization quantity of the previous time step in the recursive initial state and the temperature time series component and humidity time series component of the current time step.

[0093] The icing recursive branch of the physical state recursive model of the transmission corridor is calculated recursively based on the conductor icing heat balance equation. This equation describes the energy balance relationship between the rate of change of ice mass on the conductor surface and ambient temperature, humidity, wind speed, and the Joule heat of the conductor itself. During calculation, the icing recursive branch first reads the conductor icing thickness characterization from the previous time step in the initial recursive state, using it as the starting point. Then, it reads the temperature and humidity time-series components of the current time step. Based on the temperature time-series component, it determines the supercooled water droplet freezing efficiency coefficient under the current ambient temperature conditions. This efficiency coefficient is obtained through a lookup table, which is pre-discretized based on the experimentally measured relationship between the supercooled water droplet freezing probability and temperature. Finally, it determines the amount of supercooled water droplets in the air that can be captured based on the humidity time-series component. The freezing efficiency coefficient of supercooled water droplets, the supercooled water droplet content characterization, and the conductor icing thickness characterization from the previous time step are all substituted into the icing heat balance recursive calculation to obtain the increment of icing thickness change at the current time step. This increment is then added to the conductor icing thickness characterization from the previous time step to obtain the recursive estimate of conductor icing thickness at the current time step.

[0094] Step S430: The recursive branch of the physical state recursive model of the transmission corridor calculates the recursive estimate of the conductor wind deflection angle for the current time step based on the conductor wind deflection angle characterization quantity of the previous time step in the recursive initial state and the wind speed time series component of the current time step, combined with the conductor-ground wire mechanical damping characteristics and the response delay of wind load.

[0095] In one implementation, step S430 specifically includes the following steps S431 to S436: Step S431: Perform wind pressure conversion on the wind speed time series component of the current time step to generate the wind pressure load time series quantity acting on the conductor and ground wire. The wind pressure conversion process converts the wind speed into the pressure action borne by the surface of the conductor and ground wire.

[0096] In one implementation, step S431 specifically includes the following steps S4311 to S4316: Step S4311: Read the wind speed time sequence component of the current time step, and determine the current angle relationship between the wind direction and the conductor axis based on the wind speed time sequence component. The current angle relationship affects the effective component of wind speed in the direction perpendicular to the conductor.

[0097] The current angle between the wind direction and the conductor / ground wire axis is determined by querying the geometric relationship between the geographical orientation information of the transmission corridor tower segment and the wind direction measurement at the current time step. The conductor / ground wire axis direction of each tower segment in the transmission corridor is recorded as a static parameter in the configuration database during project deployment, and this axis direction is represented as a geographical azimuth. The wind speed time-series component includes two measurement information: wind speed magnitude and wind direction azimuth. The wind direction azimuth at the current time step is read, and the conductor / ground wire axis azimuth corresponding to the current tower segment is read. The absolute difference between the two azimuths is calculated; this difference is the current angle between the wind direction and the conductor / ground wire axis, and the angle ranges from 0 degrees to 90 degrees.

[0098] Step S4312: Input the current included angle relationship into the wind speed decomposition, and decompose the normal wind speed component perpendicular to the conductor axis and the tangential wind speed component parallel to the conductor axis from the wind speed time series component.

[0099] Wind speed decomposition is based on the principle of vector orthogonal projection. The magnitude of the wind speed in the time-series components is used as the composite wind speed magnitude, and the current angle obtained in step S4311 is used as the deviation angle between the composite wind speed direction and the normal direction of the conductor axis. The normal wind speed component is calculated by multiplying the composite wind speed magnitude by the sine of the current angle; the tangential wind speed component is calculated by multiplying the composite wind speed magnitude by the cosine of the current angle.

[0100] Step S4313: Generate wind pressure for the normal wind speed component. Based on the air density characteristics and the geometric characteristics of the windward surface of the conductor, convert the normal wind speed component into the normal wind pressure load of the conductor. The normal wind pressure load of the conductor is a representation of the pressure distribution acting on the surface of the conductor in the normal direction.

[0101] The wind pressure generation process is based on the relationship between wind pressure and wind speed in aerodynamics. The air density characteristic is taken from the standard air density reference value corresponding to the altitude of the transmission corridor, which is a common constant in the engineering field. The geometric characteristics of the conductor's windward surface are the conductor's outer diameter and surface shape factor. The conductor's outer diameter is read from the engineering parameter table according to the actual conductor type used in the line. The surface shape factor is taken as a drag coefficient constant commonly used in the engineering field for standard circular cross-section conductors. The calculation process for wind pressure generation is as follows: the normal wind speed component is squared, multiplied by the air density reference value, then multiplied by the surface shape factor, and finally multiplied by the conductor's outer diameter and divided by 2. The result is the conductor's normal wind pressure load, which represents the resultant normal wind pressure force borne per unit length of conductor surface.

[0102] Step S4314: Generate tangential friction force for the tangential wind speed component, and convert the tangential wind speed component into the conductor tangential friction load. The conductor tangential friction load is less than the conductor normal wind pressure load.

[0103] The generation of tangential friction is based on the principle of frictional resistance generated by fluid flowing tangentially along a wall. The tangential wind speed component is squared, multiplied by a reference air density value, and then multiplied by the conductor surface friction coefficient. This friction coefficient is an empirical constant much smaller than 1, determined by the surface roughness of the conductor. This result is then multiplied by the outer circumference of the conductor. The final result is the conductor tangential friction load. This load characterizes the frictional force generated on the conductor surface when wind sweeps tangentially along the conductor's axial direction. Because its physical generation mechanism differs from the pressure drag on the normal windward surface, its value is naturally smaller than the conductor normal wind pressure load.

[0104] Step S4315: Vector synthesize the normal wind pressure load and the tangential friction load of the conductor to generate the synthesized wind pressure load force and the wind pressure load direction, which are used as the wind pressure load time series quantities.

[0105] Vector composition is performed according to the parallelogram law. The direction of the normal wind pressure load is the normal direction of the conductor axis, and the direction of the tangential friction load is the tangential direction of the conductor axis; the two directions are orthogonal to each other. The vector sum of the two is calculated. Specifically, the square root of the sum of the squares of the normal wind pressure load and the tangential friction load is taken to obtain the magnitude of the synthesized wind pressure load. The arctangent angle is calculated by dividing the tangential friction load by the normal wind pressure load, yielding the deviation angle of the synthesized wind pressure load direction relative to the normal direction of the conductor. The combined magnitude and direction of the synthesized wind pressure load are recorded as a time series quantity of the wind pressure load.

[0106] Step S4316: Associate the wind pressure load time sequence with the timestamp of the current time step and output it to the joint recursive processing step.

[0107] The associated operation takes the wind pressure load time series quantity generated in step S4315 as a data field, and combines it with the timestamp field of the current time step to form a key-value pair and store it in the intermediate data cache area for subsequent wind deflection recursion branches to call.

[0108] Step S432: Based on the mechanical damping characteristics and inertial parameters of the conductor and ground wire, damping attenuation and inertial continuation are performed on the conductor wind deflection angle characterization quantity and its rate of change in the previous time step to obtain the residual oscillation state quantity in the previous time step. The residual oscillation state quantity includes the residual angular component and the residual angular velocity component.

[0109] The mechanical damping characteristics of the conductor are characterized by a damping attenuation factor, which describes the rate of energy dissipation during the conductor's free oscillation due to material damping and air resistance. Its value is calculated from the damping ratio parameter determined by the conductor's material and structure. The inertial parameters of the conductor are characterized by an inertial continuation factor, which describes the time-duration characteristic of the conductor's oscillation state under no external force. Its value is determined by the conductor's moment of inertia and the time step between the current and previous time steps. The calculation process for the residual oscillation state quantity is as follows: The residual angle and residual angular velocity components stored in the wind deflection recursive branch of the previous time step are read as the initial oscillation state. The residual angular velocity component is multiplied by the damping attenuation factor to obtain the attenuated residual angular velocity component. The residual angle component is then added to the product of the attenuated residual angular velocity component and the time step to obtain the inertial continuation residual angle component. The inertial continuation residual angle component and the attenuated residual angular velocity component are combined to output the residual oscillation state quantity of the previous time step.

[0110] Step S433: Obtain the wind deflection response time constant of the conductor and determine the dynamic response transfer ratio of the wind pressure load to the conductor swing response based on the wind deflection response time constant and the time interval between the current time step and the previous time step.

[0111] The wind deflection response time constant of the conductor is a characteristic parameter describing the time required for the conductor's deflection angle to reach a steady-state value under a step wind load. This parameter is determined by the mass, moment of inertia, tension, and mechanical damping of the conductor system. It is determined through dynamic parameter identification experiments before system deployment and stored as a constant in the model configuration parameters. The dynamic response transfer ratio is determined as follows: calculate the time interval between the current time step and the previous time step, divide this time interval by the wind deflection response time constant to obtain a dimensionless time ratio; map this time ratio to the time domain response function of a first-order inertial system by calculating 1 minus the exponential function value with the Euler number as the base and the negative value of this time ratio as the exponent. The result is the dynamic response transfer ratio, which takes a value between 0 and 1, representing the relative degree to which the conductor's oscillation response can follow the changes in wind pressure load within the current time step.

[0112] Step S434: Convert the wind pressure load time series quantity into the wind deflection driving torque acting on the conductor and ground wire. Combine the residual oscillation state quantity and the dynamic response transfer ratio, and perform recursive calculation based on the conductor and ground wire oscillation motion equation to obtain the initial recursive quantity of the wind deflection angle at the current time step. The recursive calculation integrates the effects of wind pressure torque driving, historical oscillation state continuation and dynamic response delay.

[0113] When converting wind pressure load time series quantities into wind deflection driving torque, the combined wind pressure load force modulus value is read from the wind pressure load time series quantities. This force modulus value is multiplied by the lever arm length of the conductor, which is the equivalent distance from the suspension point of the conductor to the center of the conductor cross section. Then, it is multiplied by the sine value of the angle between the combined wind pressure load action direction and the lever arm direction. The resulting product is the wind deflection driving torque. When performing recursive calculations based on the conductor-ground wire oscillation motion equation, the wind deflection driving torque is multiplied by the dynamic response transfer ratio to obtain the effective wind deflection driving torque. This process reflects the physical delay in which the wind pressure load cannot be fully transferred into an oscillation response within a finite time step. The effective wind deflection driving torque is divided by the moment of inertia of the conductor-ground wire system to obtain the wind deflection angular acceleration contribution at the current time step. The wind deflection angular acceleration contribution is multiplied by the square of the time step, and then the residual angular component in the residual oscillation state quantity is added, along with the product of the residual angular velocity component in the residual oscillation state quantity and the time step. The sum of these three terms constitutes the initial recursive wind deflection angle for the current time step.

[0114] Step S435: Read the torque balance deviation record generated during the wind deflection recursive estimation process of the previous time step, determine the angle compensation amount based on the torque balance deviation record, perform deviation compensation on the initial recursive amount of wind deflection angle, and obtain the compensated wind deflection angle recursive amount.

[0115] The torque balance deviation record is the deviation record data generated during the constraint comparison and projection correction process in steps S440 and S450 of the previous time step. This record stores the deviation direction and magnitude information of the recursive estimate from the previous time step exceeding the physically feasible interval. The angle compensation amount is determined as follows: read the deviation magnitude value in the torque balance deviation record. If the deviation magnitude value is not zero, multiply the deviation magnitude value by a preset compensation gain factor. This compensation gain factor is a positive number less than 1, and its function is to avoid control oscillation caused by overcompensation. The product is the angle compensation amount, and the compensation direction is opposite to the deviation direction. Add this angle compensation amount to the initial recursive wind deflection angle to obtain the compensated recursive wind deflection angle.

[0116] Step S436: Output the compensated wind deflection angle recursive value as the recursive estimate of the traverse wind deflection angle for the current time step, and generate the torque balance deviation record for the current time step for the next time step.

[0117] The output recursive estimate of the conductor wind deflection angle is simultaneously written to the output buffer and the recursive state register. The torque balance deviation record for the current time step is generated after the subsequent steps S440 and S450 are executed. If step S440 determines that the deviation has not exceeded the constraint boundary, the deviation magnitude in the record is set to zero; if it exceeds the constraint boundary, the deviation magnitude in the record is set to the difference before and after the projection correction in step S450. After the record is completed, it is stored in the internal state variables of the model for the next time step, step S435, to read.

[0118] Step S440: Compare the recursive estimates of conductor icing thickness and conductor wind deflection angle with the pre-set icing thermal balance constraints and wind deflection moment balance constraints in the recursive model of the physical state of the transmission corridor, and determine whether the recursive estimates fall within the physically feasible range allowed by the constraints.

[0119] The icing thermal equilibrium constraint specifies the reasonable range of variation for the recursive estimate of conductor icing thickness under given temperature and humidity conditions. The upper bound of this range is determined by the product of the theoretical maximum icing growth rate under the most favorable icing meteorological conditions and the time step, while the lower bound is determined by the product of the theoretical maximum ablation rate under the most favorable ablation meteorological conditions and the time step. The specific values ​​of the upper and lower bounds are obtained by substituting the temperature and humidity time series components of the current time step into pre-calibrated icing growth rate and ablation rate mapping tables, respectively. The wind deflection moment equilibrium constraint specifies the reasonable range of values ​​for the recursive estimate of conductor wind deflection angle under given wind speed conditions. The upper bound of this range is the theoretical maximum wind deflection angle of the conductor-ground wire system at the current wind speed, obtained by solving the static equilibrium equation of wind pressure load moment and conductor-ground wire gravity restoring moment. The lower bound is 0. The comparison operation is an interval inclusion judgment, checking whether the recursive estimate value is greater than or equal to the lower bound and less than or equal to the upper bound. If this condition is met, the recursive estimate falls within the physically feasible interval; otherwise, it is determined to be outside the range.

[0120] Step S450: If the recursive estimate exceeds the physical feasible interval, the excess portion of the recursive estimate is projected back into the physical feasible interval along the constraint boundary to obtain the constraint-corrected recursive estimate of the conductor icing thickness and the constraint-corrected recursive estimate of the conductor wind deflection angle.

[0121] The processing method for projection along the constraint boundary is as follows: if the recursive estimate is greater than the upper bound, it is forcibly assigned the upper bound value; if the recursive estimate is less than the lower bound, it is forcibly assigned the lower bound value. The value after projection correction is the recursive estimate after constraint correction. The difference between the original recursive estimate before projection correction and the constraint correction value after projection correction is recorded. This difference is the deviation record for that time step and is stored in the torque balance deviation record or icing deviation record variable of the corresponding recursive branch.

[0122] Step S460: Output the recursive estimate of conductor icing thickness and the recursive estimate of conductor wind deflection angle after constraint correction as the recursive estimated conductor-ground wire mechanical state parameters, and feed back the recursive estimated conductor-ground wire mechanical state parameters as the recursive initial state for the next time step.

[0123] The output operation packages the two constraint-corrected recursive estimates along with their timestamps into a recursive estimate of the conductor's mechanical state parameters data unit, and sends it to the model output data stream. The feedback operation overwrites the constraint-corrected recursive estimate of the conductor's icing thickness to the storage location of the conductor's icing thickness representation in the recursive initial state described in step S410, and overwrites the constraint-corrected recursive estimate of the conductor's wind deflection angle to the storage location of the conductor's wind deflection angle representation in the recursive initial state, for use as the starting point for the recursive calculation in the next time step.

[0124] Step S500: Based on the recursively estimated mechanical state parameters of the conductor and ground wire, combined with the topological connection relationship of each tower segment in the transmission corridor and the line safety operation constraints, a control signal sequence containing line dynamic operation control instructions is generated. The line dynamic operation control instructions are used to initiate the line operation status adjustment operation of the transmission corridor.

[0125] In one implementation, step S500 specifically includes the following steps S510 to S560: Step S510: Spatial decomposition of the recursively estimated conductor and ground wire mechanical state parameters according to the tower segment identifier, so that each tower segment obtains the recursively estimated conductor icing thickness and the recursively estimated conductor wind deflection angle within that tower segment.

[0126] Tower segment identifiers are unique coded identifiers for each segment within a transmission corridor, consisting of two adjacent towers and the conductor / ground wire between them. These identifiers are predefined and stored in the topology database of the transmission corridor's geographic information system. Spatial decomposition is performed based on the mapping relationship between the deployment locations of micro-meteorological sensors and the tower segments. Each set of micro-meteorological sensors is associated with one or more tower segment identifiers upon deployment, and this association is recorded in the sensor configuration table. The recursively estimated conductor / ground wire mechanical state parameters are generated by the conductor / ground wire state inference neural network corresponding to a specific micro-meteorological sensor and the transmission corridor physical state recursive model, thus naturally carrying the sensor's identifier information. During spatial decomposition, the sensor configuration table is queried using the sensor identifier as an index to obtain one or more tower segment identifiers associated with that sensor. The recursively estimated conductor icing thickness and conductor wind deflection angle from the recursively estimated conductor / ground wire mechanical state parameters are copied and assigned to each associated tower segment identifier, ensuring that each tower segment obtains the recursively estimated conductor icing thickness and conductor wind deflection angle corresponding to the micro-meteorological conditions of that segment.

[0127] Step S520: Read the preset graded safety operation limit conditions for transmission lines. The graded safety operation limit conditions divide the conductor icing thickness into multiple progressive icing response intervals and the conductor wind deflection angle into multiple progressive wind deflection response intervals. Each response interval corresponds to a control action prototype.

[0128] The graded safety operation constraints for transmission lines are a set of safety criterion rules pre-compiled and stored in the configuration database of the transmission corridor operation control system. The progressive icing response interval divides the continuous numerical range of conductor ice thickness from safe to dangerous values ​​into several contiguous intervals. Each interval has a unique icing response interval identifier, and the boundary values ​​for interval division are set based on the conductor's designed ice thickness tolerance and historical ice disaster statistics. The progressive wind deflection response interval divides the continuous numerical range of conductor wind deflection angle from zero degrees to the maximum permissible wind deflection angle into several contiguous intervals. Each interval has a unique wind deflection response interval identifier, and the boundary values ​​for interval division are set based on the tower insulation gap design value and wind deflection flashover accident statistics. The control action prototype is a pre-set standardized operation scheme template, including but not limited to the recommended values ​​for the initiation level and duration of DC de-icing operations, the operation type and tension adjustment direction of conductor and ground wire tension adjustment operations, etc. Each combination of icing response interval identifier and wind deflection response interval identifier is associated with at least one control action prototype in the control action prototype library.

[0129] Step S530: The recursive estimate of the conductor icing thickness corresponding to each tower segment is matched with the landing area in the progressive icing response interval to generate an icing response interval identifier. At the same time, the recursive estimate of the conductor wind deflection angle corresponding to each tower segment is matched with the landing area in the progressive wind deflection response interval to generate a wind deflection response interval identifier.

[0130] The matching operation is implemented through sequential traversal comparison of intervals. For the recursive estimate of conductor icing thickness, starting from the first interval of the progressive icing response interval, the value of the recursive estimate is compared sequentially with the lower and upper bounds of each interval. When the value of the recursive estimate falls between the lower and upper bounds of a certain interval, the traversal stops, and the icing response interval identifier for that interval is output. For the recursive estimate of conductor wind deflection angle, the same sequential traversal comparison method is used to find matching intervals in the progressive wind deflection response interval, and the wind deflection response interval identifier is output.

[0131] Step S540: Based on the combination relationship between the icing response interval identifier and the wind deflection response interval identifier, retrieve the corresponding control action prototype in the preset control action prototype library, and instantiate the retrieved control action prototype into a candidate control operation that includes the target tower segment, operation start time, and operation duration.

[0132] In one implementation, step S540 specifically includes the following steps S541 to S546: Step S541: Input the icing response interval identifier into the icing control mapping table. The icing control mapping table stores the correspondence between the icing response interval and the DC de-icing operation level. Output the DC de-icing operation level and the recommended value of the de-icing current duration based on the icing response interval identifier.

[0133] The icing control mapping table is a key-value pair mapping table that uses the icing response interval identifier as the index key and stores the DC icing melting operation level and the recommended duration of the icing melting current as the storage values. The DC icing melting operation level is divided into multiple levels from low to high according to the icing melting power and coverage area, with each level corresponding to different icing melting device capacity and operational complexity. The recommended duration of the icing melting current is the recommended duration of continuous application of the icing melting current under each operation level, determined based on historical icing melting experience. During a query, the icing response interval identifier is used as the search key to locate the corresponding record in the mapping table, returning the DC icing melting operation level field value and the recommended duration of the icing melting current field value from that record.

[0134] Step S542: Input the wind deflection response range identifier into the wind deflection control mapping table. The wind deflection control mapping table stores the correspondence between the wind deflection response range and the conductor tension adjustment operation type. Output the conductor tension adjustment operation type and tension force adjustment direction according to the wind deflection response range identifier.

[0135] The wind deflection control mapping table is a key-value pair mapping table that uses the wind deflection response range identifier as the index key and the conductor / ground wire tension adjustment operation type and tension force adjustment direction as the storage values. The conductor / ground wire tension adjustment operation type includes two types: increasing tension and releasing tension; the tension force adjustment direction indicates whether the current operation is increasing or decreasing. During a query, the wind deflection response range identifier is used as the search key to locate the corresponding record in the mapping table, returning the values ​​of the conductor / ground wire tension adjustment operation type and tension force adjustment direction fields from that record.

[0136] Step S543: When both DC de-icing operation level and conductor tensioning adjustment operation type exist in the same tower segment, start the operation compatibility review, obtain the control operation history sequence of the tower segment before the current time point, and record the type of operation recently executed, the execution completion time, and the trajectory of conductor status change after the operation in the control operation history sequence.

[0137] The historical sequence of control operations is stored in the operation log database of the transmission corridor operation control system. Each record includes the tower segment identifier, operation type code, operation completion timestamp, and data on the changes in conductor icing thickness and conductor wind deflection angle within a specified period after the operation. When an operation compatibility review is initiated, the latest records in the operation log database are retrieved using the current tower segment identifier as the search criterion. These records are then arranged in reverse chronological order by execution completion time to form the historical sequence of control operations.

[0138] Step S544: Determine whether the time interval between the currently applied DC de-icing operation and the most recent de-icing operation meets the minimum cooling interval requirement for de-icing operation based on the historical sequence of control operations, and generate a de-icing operation permit mark.

[0139] The minimum cooling interval for ice melting operations is a minimum time interval set to prevent overheating or shortened lifespan of the DC ice melting device due to continuous high-frequency operation. The determination process is as follows: Records of operation type DC ice melting are selected from the historical control operation sequence. The record with the most recent execution completion time is taken, and the time interval between the current time and the execution completion time of that record is calculated. This time interval is compared with the minimum cooling interval for ice melting operations. If the time interval is greater than or equal to the minimum cooling interval, a permitted ice melting operation flag is generated; if the time interval is less than the minimum cooling interval, a prohibited ice melting operation flag is generated.

[0140] Step S545: Based on the historical sequence of control operations, determine whether the cumulative adjustment amount of the current applied conductor tensioning adjustment operation and the recent tensioning adjustment operations in the same direction exceeds the cumulative upper limit of mechanical adjustment, and generate a tensioning operation permission mark.

[0141] In one implementation, step S545 specifically includes the following steps S5451 to S5455: Step S5451: Filter out all historical operation records related to conductor tension adjustment from the control operation history sequence, and sort them according to the operation execution time to obtain the tension operation history sequence.

[0142] The filtering operation involves reading the operation type code field of each record in the historical sequence of control operations, retaining records whose operation type codes belong to the conductor tensioning adjustment operation code category, and filtering out the remaining records. The retained records are then sorted in ascending order according to their execution completion timestamps to obtain the historical sequence of tensioning operations.

[0143] Step S5452: Extract the same-direction operation records from the tensioning operation history sequence that have the same adjustment direction as the current applied conductor tensioning adjustment operation, and accumulate the single adjustment amount in each same-direction operation record in chronological order to obtain the total cumulative adjustment amount in the same direction.

[0144] The adjustment direction field can take two values: increase and decrease. The adjustment direction of the currently requested conductor tensioning operation is compared with the adjustment direction field of each record in the tensioning operation history sequence. Records with the same direction are selected as same-direction operation records. The single adjustment value in each same-direction operation record is read and added sequentially according to the execution completion time of the record from earliest to latest. The accumulated result is the total cumulative adjustment amount in the same direction.

[0145] Step S5453: Obtain the preset cumulative upper limit threshold for conductor tension mechanical adjustment. The cumulative upper limit threshold limits the total tension adjustment in the same direction to not exceed the maximum stroke range allowed by the mechanical structure. Compare the total cumulative adjustment in the same direction with the cumulative upper limit threshold.

[0146] The cumulative upper limit threshold for conductor tension mechanical adjustment is the mechanical travel limit of the conductor tension adjustment device. This value is given by the technical parameters provided by the device manufacturer and stored as a configuration parameter in the global parameter table of the control action prototype library. The comparison operation calculates whether the total cumulative adjustment in the same direction is less than the cumulative upper limit threshold. If so, the result is marked as not exceeding the limit; otherwise, the result is marked as exceeding the limit.

[0147] Step S5454: When the total cumulative adjustment amount in the same direction plus the single adjustment amount of the current applied conductor tensioning adjustment operation is still lower than the cumulative upper limit threshold, a tensioning operation permission mark is generated in the permission state.

[0148] The total cumulative adjustment amount obtained in step S5452 is added to the single adjustment amount of the current application output in step S542, and the sum is compared with the cumulative upper limit threshold. If the sum is less than the cumulative upper limit threshold, it is determined that the tension adjustment operation of the current application will not cause the total cumulative amount to exceed the mechanical travel limit, and a tension operation permission mark is generated in the permission state.

[0149] Step S5455: When the total cumulative adjustment in the same direction has reached or will exceed the cumulative upper limit threshold after adding the current single adjustment amount, read the conductor status feedback after the last same-direction operation record, determine whether the conductor is currently in the resettable range, generate a permission mark with a reset operation suggestion if it is in the resettable range, generate a tensioning operation permission mark in the prohibited state if it is in the non-resettable range, and output the tensioning operation permission mark to the joint determination step.

[0150] The conductor / ground wire status feedback is the tension and sag values ​​measured by sensors after the last unidirectional tensioning adjustment operation. This data is recorded in the conductor status change trajectory field of the corresponding record in the historical sequence of control operations. The resettable range is a preset range of tension and sag values, representing the range within which the conductor / ground wire can be safely restored to its mechanical centerline state through a reverse operation to release tension. During the judgment process, the status change trajectory recorded from the last unidirectional operation is read, and the latest tension and sag feedback values ​​are extracted. These two values ​​are then compared with the upper and lower bounds of the resettable range for tension and sag, respectively, for inclusion. If both values ​​fall within the resettable range, the conductor / ground wire is currently in the resettable range, and a tensioning operation permission mark with a reset operation suggestion is generated. The reset operation suggestion recommends performing a reverse tensioning adjustment operation to release the accumulated adjustment amount. If at least one value does not fall within the resettable range, the conductor / ground wire is currently in the non-resettable range, and a tensioning operation permission mark is generated for a prohibited state.

[0151] Step S546: Jointly determine the de-icing operation permission flag and the tensioning operation permission flag. When both are permitted, output the DC de-icing operation level and the conductor tensioning adjustment operation type simultaneously. When either flag is prohibited, downgrade or delay the operation to generate the final candidate control operation.

[0152] The logic for joint determination is as follows: If both the ice-melting operation permission mark and the tensioning operation permission mark are in a permitted state, then the DC ice-melting operation level and the recommended value of the ice-melting current duration output in step S541, as well as the conductor tensioning adjustment operation type and tension force adjustment direction output in step S542, are combined into a candidate control operation; if the ice-melting operation permission mark is in a permitted state but the tensioning operation permission mark is in a prohibited state, then the tensioning adjustment operation is downgraded and replaced with a conservative scheme of reducing the tensioning adjustment amplitude or not performing tensioning adjustment, while the DC ice-melting operation is retained as usual, and a candidate control operation is generated; if the ice-melting operation permission mark is in a prohibited state, then the operation is delayed until the next available time window after the minimum cooling interval is met, and the tensioning adjustment operation is retained or downgraded according to the permission status, and a candidate control operation is generated.

[0153] Step S550: Obtain the topology connection description of each tower segment in the transmission corridor. The topology connection description includes the electrical connection sequence and power flow direction between tower segments. Perform electrical correlation impact analysis on candidate control operations of adjacent tower segments and identify the cascading state changes caused by the implementation of control operations between adjacent tower segments.

[0154] The topology connection description is extracted from the topology database of the power transmission corridor geographic information system. This database uses an adjacency list structure to store the connection relationships between each tower segment. Each tower segment node is associated with the identifiers of its electrical upstream neighbor tower segments and its downstream neighbor tower segments, while also recording the power flow direction from upstream to downstream. The electrical association impact analysis is completed by traversing the topology adjacency list level by level. Taking each tower segment with a candidate control operation as the starting point, the impact analysis is propagated along the power flow direction to its downstream neighbor tower segments. The analysis content is as follows: if the current tower segment performs a DC de-icing operation, causing the conductor temperature to rise, the temperature rise will affect the local micro-meteorological conditions of the downstream tower segment through conductor heat conduction and air convection, thereby affecting the icing state of the downstream tower segment; if the current tower segment performs a conductor and ground wire tension adjustment operation, causing a change in conductor tension, the tension change will affect the conductor sag and tower stress of adjacent tower segments through conductor mechanical transmission. For each affected downstream tower segment, re-estimate the changing trends of conductor icing thickness and conductor wind deflection angle to determine whether new safe operation constraints will be triggered, and record the triggered new conditions as cascading state change events.

[0155] Step S560: Based on the identification results of the interlocking state changes, coordinate and correct the candidate control operations, generate a corrected control operation sequence, and arrange the corrected control operation sequence into a control signal sequence containing the line dynamic operation control instructions according to the order of operation start time.

[0156] The coordination and correction method is as follows: For each downstream tower segment that triggers a new safety constraint due to a cascading state change event, a corresponding control action is added to the candidate control operation queue for that tower segment. The start time of this added action is set to the start time of the upstream tower segment's operation plus the delay duration of the cascading state change propagation. This delay duration is estimated jointly by the conductor's thermal conduction velocity or mechanical transmission velocity and the tower segment spacing. When the control operations of multiple tower segments overlap or conflict in time, the execution sequence is arranged in the order of prioritizing the upstream tower segments and then processing the downstream tower segments. After all coordination and corrections are completed, the final control operations of all tower segments are uniformly sorted according to the order of their start times. Each control operation is formatted as a line dynamic operation control instruction, and the instruction field includes the target tower segment identifier, operation type code, operation start time, operation duration, and operation parameter values. All control instructions are arranged in ascending order of time to form a control signal sequence. This control signal sequence is distributed to the field actuators of each tower segment through the execution interface of the transmission corridor operation control system to initiate the line operation status adjustment operation of the transmission corridor.

[0157] Please refer to Figure 4 This diagram illustrates the structural block diagram of a server 20 provided in one embodiment of the present invention. This server can be used to implement the functions of the aforementioned method for monitoring the environment of power transmission corridors based on micro-meteorological fusion data. Specifically:

[0158] Server 20 includes a Central Processing Unit (CPU) 21, a system memory 24 including Random Access Memory (RAM) 22 and Read Only Memory (ROM) 23, and a system bus 25 connecting the system memory 24 and the CPU 21. Computer device 20 also includes a basic input / output system (I / O system) 26 that facilitates information transfer between various devices within the computer, and a mass storage device 27 for storing the operating system 271.

[0159] The input / output system 26 may include a display for showing information and input devices such as a mouse and keyboard for user input. Both the display and the input devices are connected to the central processing unit 21 via an input / output controller connected to the system bus 25.

[0160] Mass storage device 27 is connected to central processing unit 21 via a mass storage controller (not shown) connected to system bus 25. Mass storage device 27 and its associated computer-readable media provide non-volatile storage for computer device 20. That is, mass storage device 27 may include computer-readable media (not shown) such as hard disk or CD-ROM (Compact Disc Read-Only Memory) drive.

[0161] Without loss of generality, computer-readable media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other solid-state storage devices, CD-ROM, DVD (Digital Video Disc) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that computer storage media are not limited to the above-mentioned types. The system memory 24 and mass storage device 27 described above can be collectively referred to as memory.

[0162] According to various embodiments of the present invention, the computer device 20 can also be connected to a remote computer on a network such as the Internet. That is, the computer device 20 can be connected to the network 29 via the network interface unit 28 connected to the system bus 25, or the network interface unit 28 can be used to connect to other types of networks or remote computer systems (not shown).

[0163] The above description is merely an exemplary embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for environmental monitoring of power transmission corridors based on micro-meteorological fusion data, characterized in that, include: Acquire micro-meteorological time-series data collected by micro-meteorological sensors deployed on power transmission towers within a continuous time window; The micro-meteorological time-series data is input into a trained conductor-ground wire state inference neural network. The forget gate dynamic value of the recursive processing layer inside the conductor-ground wire state inference neural network is dynamically modulated by the wind speed disturbance pattern features derived in real time from the wind speed time-series components. The forget gate dynamic value controls the retention and discard ratio of historical hidden state information by the recursive processing layer. The conductor-ground wire state inference neural network introduces physical mechanism constraints during the training phase. The training loss function embeds conductor-ground wire icing thermal balance loss term and wind deflection moment balance loss term to guide the network to converge within the physical feasible region. The conductor and ground wire state estimation neural network directly outputs the conductor and ground wire mechanical state parameters based on the input micro-meteorological time series data. The conductor and ground wire mechanical state parameters include conductor ice thickness characterization and conductor wind deflection angle characterization. The conductor ice thickness characterization reflects the continuous change of the degree of ice accumulation on the conductor surface, and the conductor wind deflection angle characterization reflects the change of the conductor's deviation angle under wind load. The mechanical state parameters of the conductor and ground wire are input into the pre-constructed recursive model of the physical state of the transmission corridor. The recursive model of the physical state of the transmission corridor performs a physical mechanism-driven recursive estimation of the mechanical state of the conductor and ground wire at the current moment based on the mechanical state parameters of the conductor and ground wire at the previous moment and the micro-meteorological time series data at the current moment, and generates the recursively estimated mechanical state parameters of the conductor and ground wire. When the recursive model of the physical state of the transmission corridor is running, it is simultaneously subject to the constraints of the ice-heat balance constraint and the wind deflection moment balance constraint. Based on the recursively estimated mechanical state parameters of the conductor and ground wire, and combined with the topological connection relationship of each tower segment in the transmission corridor and the line safety operation constraints, a control signal sequence containing line dynamic operation control instructions is generated. The line dynamic operation control instructions are used to initiate the line operation status adjustment operation of the transmission corridor.

2. The method according to claim 1, characterized in that, The step of inputting the micro-meteorological time-series data into a trained conductor-to-ground wire state estimation neural network, wherein the forget gate dynamic value of the recursive processing layer inside the conductor-to-ground wire state estimation neural network is dynamically modulated by the wind speed disturbance pattern features derived in real time from the wind speed time-series components, includes: Multi-time-granularity disturbance detection is performed on the wind speed time series components to generate wind speed disturbance pattern features including instantaneous disturbance amplitude sequence, continuous disturbance intensity sequence and disturbance intermittent frequency sequence. The instantaneous disturbance amplitude sequence captures short-term abrupt changes in wind speed, the continuous disturbance intensity sequence tracks the maintenance state of wind speed fluctuations, and the disturbance intermittent frequency sequence records the switching frequency between wind condition stable periods. The instantaneous disturbance amplitude sequence, the continuous disturbance intensity sequence, and the disturbance intermittent frequency sequence are input into the disturbance pattern encoding to generate a disturbance type distribution vector. The disturbance type distribution vector indicates the degree to which the current wind speed disturbance tends to be an isolated sudden disturbance, a continuous fluctuation, or an intermittent impact. Based on the perturbation type with the highest tendency in the perturbation type distribution vector, the adjustment strategy tendency of the forgetting gate dynamic value is determined. When the perturbation type is isolated sudden perturbation, the tendency is to suppress the forgetting amplitude to retain the historical state. When the perturbation type is intermittent impact, the tendency is to amplify the forgetting amplitude to accelerate the state update. The adjustment trajectory of the forget gate dynamic value output at the previous time step of the recursive processing layer is obtained. The adjustment trajectory records the trend of forget gate dynamic value changes over multiple past time steps. The adjustment strategy tendency of the current time step is inertially connected with the adjustment trajectory to generate the forget gate dynamic value of the current time step. The cell state vector passed from the previous time step to the current time step is gating filtered by the forget gate dynamic value of the current time step. The historical information in the cell state vector that is compatible with the current wind condition pattern is retained and the historical information that is mismatched is discarded, so as to obtain the filtered cell state vector. Based on the filtered cell state vector and the micro-meteorological time series data input to the recursive processing layer at the current time step, a hidden state vector for the current time step is generated. The hidden state vector is then output to the subsequent processing layer of the conductor state inference neural network and copied and passed to the recursive processing layer at the next time step.

3. The method according to claim 2, characterized in that, The step of performing multi-time-granularity disturbance detection on the wind speed time-series components generates wind speed disturbance pattern features including instantaneous disturbance amplitude sequences, continuous disturbance intensity sequences, and disturbance intermittent frequency sequences, including: The wind speed time series component is input into the disturbance separation structure, which separates the trend component and the fluctuation component of the wind speed time series component to generate a smooth trend sequence and a residual fluctuation sequence. The smooth trend sequence reflects the overall evolution direction of the wind speed, and the residual fluctuation sequence reflects the short-term fluctuation of the wind speed in the overall evolution direction. The residual fluctuation sequence is subjected to mutation detection. Points in the residual fluctuation sequence where the change amplitude of fluctuation value at adjacent time points exceeds the mutation response boundary are marked as mutation candidate points. Continuous mutation candidate points are merged into mutation event segments along the time axis. The peak amplitude and duration span of each mutation event segment are recorded to generate a mutation event record set. For each mutation event segment in the mutation event record set, its peak amplitude is extracted as the instantaneous perturbation amplitude of the event segment, and the peak amplitude is associated with each time point covered by the mutation event segment to obtain the instantaneous perturbation amplitude sequence; A fluctuation envelope analysis is performed on the residual fluctuation sequence to generate an upper envelope and a lower envelope. The fluctuation energy curve is determined based on the change of the envelope width between the upper and lower envelopes over time. The fluctuation energy curve is then subjected to sliding accumulation to generate the continuous disturbance intensity value at each time point. The continuous disturbance intensity sequence is obtained by arranging the values ​​in chronological order. The residual fluctuation sequence is used to identify the stationary period. The time period in the residual fluctuation sequence where the fluctuation value is continuously lower than the stationarity determination boundary is marked as a stationary segment. The time distance between the end time of each stationary segment and the start time of the next stationary segment is recorded as the interval duration. The frequency of the interval duration within the preset backtracking interval is taken as the disturbance interval frequency at the corresponding time point. The disturbance interval frequency sequence is obtained by arranging them by time. The instantaneous disturbance amplitude sequence, the continuous disturbance intensity sequence, and the disturbance intermittent frequency sequence are aligned and integrated on the time axis. The time offset introduced by the detection delay is eliminated based on the original timestamp of the wind speed time series component, and the wind speed disturbance pattern feature is generated.

4. The method according to claim 3, characterized in that, The step of determining the adjustment strategy tendency of the forget gate dynamic value based on the perturbation type with the highest tendency in the perturbation type distribution vector includes: Perform a primary type determination on the disturbance type distribution vector, compare the magnitudes of the tendency values ​​of isolated sudden disturbance, continuous fluctuation, and intermittent impact in the disturbance type distribution vector, mark the type with the largest tendency value as the primary disturbance type, and mark the other two types as auxiliary disturbance types; When the main control disturbance type is marked as isolated sudden disturbance, the instantaneous disturbance amplitude corresponding to the current time step is extracted from the instantaneous disturbance amplitude sequence. It is determined whether the instantaneous disturbance amplitude is within the preset isolated sudden disturbance response sensitive range. If it is within the sensitive range, the adjustment strategy tendency direction is set to the forgetting suppression direction. If it is not within the sensitive range, the adjustment strategy tendency direction is reversed to the neutral direction. When the main control disturbance type is marked as intermittent impact type, the disturbance interval frequency of the current time step is extracted from the disturbance interval frequency sequence, and the disturbance interval frequency is compared with the intermittent impact frequency reference value. If the disturbance interval frequency is greater than the intermittent impact frequency reference value, the adjustment strategy tendency direction is set to the forgetting acceleration direction; otherwise, the adjustment strategy tendency direction of the previous time step is maintained. When the main control disturbance type is marked as continuous fluctuation type, the continuous disturbance intensity of the current time step is extracted from the continuous disturbance intensity sequence, and the continuous disturbance intensity is compared with the continuous fluctuation intensity reference value. If the continuous disturbance intensity is greater than the continuous fluctuation intensity reference value, the adjustment strategy tendency direction is set to the forgetting fine-tuning direction, and the adjustment step size of the forgetting gate dynamic value is limited to a narrow step size range. The tendency value and its direction of change of the auxiliary disturbance type are obtained. When the tendency value of the auxiliary disturbance type shows a unidirectional increasing trend in a continuous time step and the auxiliary disturbance type and the main disturbance type are conflict types, a conflict warning signal of the adjustment strategy tendency is generated, and the execution strength of the current adjustment strategy tendency is weakened according to the conflict warning signal. The adjustment strategy tendency weakened by the enforcement intensity is fused with the adjustment strategy tendency of the previous time step to generate the adjustment strategy tendency of the current time step, and then output to the inertial connection processing step.

5. The method according to claim 1, characterized in that, The method for obtaining the conductor and ground wire state estimation neural network directly outputs the conductor and ground wire mechanical state parameters based on the input micro-meteorological time-series data. These mechanical state parameters include a conductor icing thickness characterization parameter and a conductor wind deflection angle characterization parameter, comprising: Extract the hidden state vector sequence generated by the recursive processing layer inside the conductor state estimation neural network on all time steps covered by the micro-meteorological time series data. Each hidden state vector in the hidden state vector sequence is a fusion representation of the micro-meteorological information and historical state information of the corresponding time step. The hidden state vector sequence is input to the conductor state mapping structure connected to the output of the recursive processing layer. The conductor state mapping structure performs dimensional projection transformation on the hidden state vector at each time step to generate conductor icing-related projection components and conductor wind deflection-related projection components. The conductor icing-related projection components and conductor wind deflection-related projection components are generated by mutually independent branch mapping channels. The time-series memory accumulation is performed on the ice-related projection components of the conductor, and an ice accumulation memory chain is maintained. The ice-related projection components of the conductor at each time step are recursively superimposed along the ice accumulation memory chain to generate a sequence of conductor ice thickness characterization quantities. The time-series response dynamics of the wind deflection-related projection components of the conductor are performed, and the inertial hysteresis characteristics of the conductor wind deflection angle response process are simulated by combining the wind speed time-series components of the corresponding time step in the micro-meteorological time-series data, thereby generating a sequence of conductor wind deflection angle characterization quantities. The sequence of conductor ice thickness and the sequence of conductor wind deflection are time-registered according to their respective associated timestamps, so that the conductor ice thickness and conductor wind deflection at the same timestamp form a state pair. The state pairs are organized into the conductor-ground wire mechanical state parameters according to the timestamp order, and the conductor-ground wire mechanical state parameters are output to the transmission corridor physical state recursive model.

6. The method according to claim 5, characterized in that, The step involves performing temporal memory accumulation on the icing-related projection components of the conductor, maintaining an icing accumulation memory chain, and recursively superimposing the icing-related projection components of each time step along the icing accumulation memory chain to generate a sequence of conductor icing thickness characterization parameters, including: Initialize the icing accumulation memory chain, where the accumulated state quantity of the icing accumulation memory chain at the initial time step is a preset icing reference quantity; Read the conductor icing-related projection component at the current time step, and extract the temperature time series component corresponding to the current time step from the micro-meteorological time series data. Generate a temperature modulation coefficient based on the temperature time series component, and scale the amplitude of the conductor icing-related projection component using the temperature modulation coefficient to obtain the temperature effect corrected projection component. Read the humidity time series component corresponding to the current time step in the micro-meteorological time series data, generate a humidity modulation coefficient based on the humidity time series component, and perform amplitude scaling again on the temperature effect-corrected projection component using the humidity modulation coefficient to obtain the humidity-heat co-corrected projection component. The cumulative state quantity of the previous time step is read from the icing accumulation memory chain, and the projected component after the wet and heat co-correction is superimposed with the cumulative state quantity of the previous time step in a directed recursive manner. The directed recursive superposition process restricts the direction of decrease of the state quantity according to the unidirectional accumulation characteristics of the icing process. The superimposed state variables are truncated to a lower bound. When the superimposed state variables are lower than the icing reference value, the state variables are restored to the icing reference value to obtain the conductor icing thickness characterization value at the current time step. The conductor icing thickness characterization value at the current time step is written into the icing accumulation memory chain as the accumulated state value for the next time step, and the conductor icing thickness characterization values ​​at each time step are arranged in chronological order to generate the conductor icing thickness characterization value sequence.

7. The method according to claim 6, characterized in that, The step of generating a temperature modulation coefficient based on the temperature time-series component, and then scaling the amplitude of the icing-related projection component of the conductor using the temperature modulation coefficient to obtain a temperature-effect-corrected projection component includes: The temperature time series component of the current time step is read from the micro-meteorological time series data, and the historical temperature sequence of the temperature time series component within the preset backtracking interval is extracted. The historical temperature sequence is arranged in chronological order to form a temperature evolution trajectory. The temperature evolution trajectory is divided into temperature change stages, the heating segment, cooling segment and constant temperature segment in the temperature evolution trajectory are identified, and the stage type label of the current time step in the temperature evolution trajectory is extracted, while the duration of the current stage is recorded. The temperature evolution trajectory is analyzed to determine the degree of temperature fluctuation, and a temperature fluctuation activity characterization quantity is generated. The temperature fluctuation activity characterization quantity reflects the frequency and range of temperature value fluctuations in the historical temperature sequence. The stage type label, the duration of the current stage, and the temperature fluctuation activity characterization quantity are jointly input into the temperature modulation decision. The temperature modulation decision process outputs the modulation tendency type, which includes one of the following: enhancement tendency, suppression tendency, and neutral maintenance tendency. The modulation coefficient generation path is determined according to the modulation tendency type. When the modulation tendency type is enhancement tendency, an amplification-type temperature modulation coefficient is generated along the enhancement mapping path. When the modulation tendency type is suppression tendency, an attenuation-type temperature modulation coefficient is generated along the suppression mapping path. When the modulation tendency type is neutral maintenance tendency, a neutral temperature modulation coefficient is generated. The generated temperature modulation coefficient is combined with the amplitude scaling of the conductor icing-related projection component at the current time step to obtain the temperature effect-corrected projection component, and the temperature effect-corrected projection component is output to the step of damp-heat co-correction.

8. The method according to claim 1, characterized in that, The step involves inputting the conductor and ground wire mechanical state parameters into a pre-constructed recursive model of the transmission corridor's physical state. Based on the conductor and ground wire mechanical state parameters from the previous moment and the current moment's micro-meteorological time-series data, the recursive model performs a physical mechanism-driven recursive estimation of the conductor and ground wire's mechanical state at the current moment, generating the recursively estimated conductor and ground wire mechanical state parameters, including: Extract the conductor ice thickness and conductor wind deflection from the conductor mechanical state parameters of the previous time step as the recursive initial state of the recursive model of the power transmission corridor physical state. The wind speed time series component, temperature time series component, and humidity time series component from the micro-meteorological time series data of the current time step are input into the recursive model of the physical state of the transmission corridor. The icing recursive branch of the physical state recursive model of the transmission corridor calculates the recursive estimate of the conductor icing thickness of the current time step based on the conductor icing thickness characterization quantity of the previous time step in the recursive initial state and the temperature time series component and humidity time series component of the current time step. The recursive branch of the physical state recursive model of the power transmission corridor calculates the recursive estimate of the conductor wind deflection angle at the current time step based on the conductor wind deflection angle characterization quantity of the previous time step in the recursive initial state and the wind speed time series component of the current time step, combined with the mechanical damping characteristics of the conductor and the response delay of the wind load. The recursive estimates of the conductor icing thickness and the conductor wind deflection angle are compared with the pre-set icing thermal balance constraints and wind deflection moment balance constraints in the recursive model of the physical state of the transmission corridor, respectively, to determine whether the recursive estimates fall within the physically feasible range allowed by the constraints. If the recursive estimate exceeds the physical feasible interval, the excess portion of the recursive estimate is projected back into the physical feasible interval along the constraint boundary to obtain the constraint-corrected recursive estimate of the conductor icing thickness and the constraint-corrected recursive estimate of the conductor wind deflection angle. The recursive estimate of the conductor icing thickness and the recursive estimate of the conductor wind deflection angle after constraint correction are output as the recursive estimated conductor-ground wire mechanical state parameters. At the same time, the recursive estimated conductor-ground wire mechanical state parameters are fed back as the recursive initial state for the next time step. The wind deflection recursive branch of the physical state recursive model of the transmission corridor calculates the recursive estimate of the conductor wind deflection angle for the current time step based on the conductor wind deflection angle characterization quantity of the previous time step in the recursive initial state and the wind speed time series component of the current time step, combined with the conductor-ground wire mechanical damping characteristics and the response delay of wind load, including: The wind speed time-series component at the current time step is converted into wind pressure to generate the wind pressure load time-series quantity acting on the conductor and ground wire. The wind pressure conversion process converts the wind speed into the pressure exerted on the surface of the conductor and ground wire. Based on the mechanical damping characteristics and inertial parameters of the conductor and ground wire, the wind deflection angle characterization quantity and its rate of change of the conductor in the previous time step are damped and inertially continued to obtain the residual oscillation state quantity of the previous time step, which includes the residual angular component and the residual angular velocity component. Obtain the wind deflection response time constant of the conductor and determine the dynamic response transfer ratio of the wind pressure load to the conductor swing response based on the wind deflection response time constant and the time interval between the current time step and the previous time step. The wind pressure load time series is converted into the wind deflection driving torque acting on the conductor and ground wire. Combined with the residual oscillation state quantity and dynamic response transfer ratio, the initial recursive amount of the wind deflection angle at the current time step is obtained by recursively solving the oscillation motion equation of the conductor and ground wire according to the residual oscillation state quantity and dynamic response transfer ratio. Read the torque balance deviation record generated during the wind deflection recursive estimation process of the previous time step, determine the angle compensation amount based on the torque balance deviation record, perform deviation compensation on the initial recursive amount of wind deflection angle, and obtain the compensated wind deflection angle recursive amount. The compensated wind deflection angle recursive value is output as the recursive estimate of the conductor wind deflection angle for the current time step, and the torque balance deviation record for the current time step is generated for use in the next time step.

9. A server, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program, which is loaded and executed by the processor to implement the power transmission corridor environmental monitoring method based on micro-meteorological fusion data as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which is loaded and executed by a processor to implement the power transmission corridor environmental monitoring method based on micro-meteorological fusion data as described in any one of claims 1 to 8.

Images