Electric pole self-adaptive environment adjusting and reinforcing method

Abstract

The invention provides a self-adaptive environment adjusting and reinforcing method for an electric pole, and aims to improve the stability and durability of the electric pole under different environment conditions through an advanced monitoring and control technology. The method comprises the steps that a plurality of sensors are installed on an electric pole, and environment structure data such as temperature, humidity, stress and inclination are monitored in real time. Through the data, a pre-trained neural network prediction model can predict the maximum stress point possibly borne by the electric pole in a short period in the future and the environment change condition. And according to the prediction result, a reinforcement learning algorithm is used to generate a control strategy for the adjustable supporting device around the electric pole. The control strategies are then converted into control signals, and the supporting device is guided to adjust the supporting strength and direction of the supporting device so as to reinforce the electric pole structure. According to the method, the response capability of the electric pole is enhanced, and the safety and stability of the electric pole under extreme weather and unstable geological conditions are ensured by automatically adjusting the supporting system.

Description

Technical Field

[0001] The present invention relates to the technical field of electric poles, and in particular to a method for adaptively adjusting and reinforcing electric poles. Background Art

[0002] Utility poles play a vital role in power and communications infrastructure. They must withstand a variety of environmental stresses, such as extreme weather, temperature fluctuations, and geological activity. To maintain the structural integrity and functionality of utility poles, existing technologies employ a variety of monitoring devices, including sensors, to monitor the environmental and structural status of the poles in real time. These sensors measure parameters such as temperature, humidity, and tilt, and feed this data back to control centers for real-time monitoring and assessment of the pole's health.

[0003] However, while existing technologies can provide real-time data on the status of poles, they are limited in how to take preventive or corrective measures based on this data. Existing systems often lack the ability to translate this data into practical actions, especially when dynamic environmental conditions are rapidly changing. There is a lack of effective real-time response mechanisms to adjust the support structure to accommodate incoming environmental stresses or structural loads.

[0004] Therefore, it is urgent to develop a new method for adaptive environmental adjustment and reinforcement of electric poles. Summary of the Invention

[0005] The present application provides a method for adaptive environmental adjustment and reinforcement of electric poles to improve the stability and durability of electric poles under different environmental conditions.

[0006] This application provides a method for adaptive environmental adjustment and reinforcement of electric poles, comprising:

[0007] Multiple sensors are installed on the poles to monitor environmental structural data including temperature, humidity, stress and tilt data in real time;

[0008] An adjustable support device is arranged around the pole, and the adjustable support device is used to adjust the support force and direction according to the received control signal;

[0009] Using a pre-trained neural network prediction model, the system generates predictions based on real-time environmental structural data from sensors and meteorological, geological, and pole performance data from a historical database. These predictions include the maximum stress points and environmental changes that the poles may experience in the short term.

[0010] Based on the prediction results of the neural network prediction model and the current state of the adjustable support device, a reinforcement learning algorithm is used to generate a control strategy for the adjustable support device;

[0011] According to the control strategy, a control signal for adjustment and reinforcement is generated; and the adjustable support device reinforces the electric pole using the generated control signal.

[0012] Furthermore, the neural network prediction model includes:

[0013] A time series preprocessing module is used to extract features from real-time environmental structure data from sensors installed on power poles using a one-dimensional convolutional neural network structure to obtain feature data; and to process the feature data using a long short-term memory network to obtain preprocessed time series features;

[0014] The historical data fusion module is used to receive historical data from the historical database, including meteorological, geological and pole performance data, and perform weighted fusion of different types of historical data through the self-attention mechanism to obtain the fused historical feature representation;

[0015] The multimodal feature integration module is used to receive the preprocessed time series features provided by the time series preprocessing module and the fused historical feature representation provided by the historical data fusion module; interact and fuse the preprocessed time series features with the fused historical feature representation to generate an integrated multimodal feature representation;

[0016] The prediction output module is implemented using a multi-head attention mechanism and a feedforward neural network to receive the multimodal feature representation provided by the multimodal feature integration module; and through a multi-task learning method, it predicts the maximum stress point and environmental changes that the pole may withstand in the short term in the future. The environmental changes include changes in temperature, humidity and wind speed.

[0017] Furthermore, the historical data fusion module uses the following formula 1 to calculate the fused historical feature representation F:

[0018]

[0019] Among them, D i represents the i-th specific type of historical data; w i It represents the weight coefficient of the i-th specific type of historical data and is calculated by the following formula 2:

[0020]

[0021] Where H represents the high-dimensional feature vector set of all historical data extracted from the historical database; D i represents the i-th specific type of historical data; D jrepresents the jth specific type of historical data; score(H,D) is a scoring function used to evaluate the correlation between H and the specific type of historical data D, calculated using the following formula 3:

[0022] score(H,D)=V T tanh(W h H+W d D+b)(3)

[0023] Among them, V is the weight matrix used to transform the comprehensive historical feature representation; W h is the transformation matrix used to transform the historical feature matrix H; W d is the transformation matrix used to transform the specific historical data type D; b is the bias vector.

[0024] Furthermore, the multimodal feature integration module uses a graph neural network-based structure to process and fuse different feature sources, specifically including:

[0025] The time series features and historical features are represented as nodes in the graph, where each node represents a feature vector, V i Corresponding to the time series characteristics, V j Corresponding historical characteristics;

[0026] Information is passed between nodes through the graph neural network, and the following formula 4 is used as the update function to iteratively update the state of each node:

[0027]

[0028] in, represents the feature representation of node i at layer l+1; represents the feature representation of node i at layer l; represents the set of neighbor nodes of node i; c ij is a normalization constant, representing the number of neighbors of node i; W (l) The weight parameters learned in layer l; represents the feature representation of node j at layer l; B (l) is the bias parameter learned in layer l; σ is the nonlinear activation function sigmoid;

[0029] The updated features of all nodes are aggregated into a single graph-level representation using the following formula 5 for subsequent feature processing:

[0030]

[0031] Among them, F integratedis the integrated multimodal feature representation; READOUT is the aggregation function, which is implemented by summation; L represents the last layer of the graph network; G represents the entire graph; Represents a set including the feature representations of all nodes in the graph G at the last layer L.

[0032] Furthermore, the prediction output module is specifically used to:

[0033] Receiving the integrated multimodal feature representation provided by the multimodal feature integration module;

[0034] Processing the multimodal feature representation using a multi-head attention mechanism, specifically comprising dividing the input multimodal feature representation into n heads, each head corresponding to a subspace; for each head, respectively calculating a query vector, a key vector, and a value vector; within each head, calculating an attention weight using scaled dot product attention; and concatenating the outputs of all heads and performing a linear transformation to obtain an output of the multi-head attention;

[0035] The output of the multi-head attention is fed into two parallel feedforward neural networks, which are used to predict the maximum stress point and environmental changes respectively. The two parallel feedforward neural networks include:

[0036] The maximum stress point prediction network consists of two hidden layers, uses the ReLU activation function, and the output layer uses the linear activation function;

[0037] The environmental change prediction network consists of three parallel sub-networks that predict changes in temperature, humidity, and wind speed respectively.

[0038] Furthermore, the state space of the reinforcement learning algorithm includes real-time data obtained from multiple sensors and the maximum stress points and environmental changes that the poles may be subjected to in the short term in the future obtained from the prediction model.

[0039] Furthermore, the action space of the reinforcement learning algorithm includes support force adjustment and direction change of the adjustable support device.

[0040] Furthermore, the reinforcement learning algorithm uses the reward function R(s, a, s) provided by the following formula 6: ′ ):

[0041]

[0042] Where Δstability represents the change in pole stability after taking action a, specifically from state s to state s ′ The stability difference; Δtime represents the time from state s to state s ′The time required is used to measure the rapidity of the adjustment response; energy_cost represents the energy consumption of performing action a, including the energy required to adjust the strength and direction of the support device; maintenance_cost represents the long-term maintenance cost, reflecting the equipment wear and maintenance costs caused by frequent adjustments; predicted_deviation represents the deviation between the predicted results and the actual environmental changes, which is used to evaluate the accuracy of the prediction model; k1, k2, k3, and k4 are adjustment coefficients.

[0043] Furthermore, generating a control signal for adjustment and reinforcement according to the control strategy includes:

[0044] Obtain the optimal adjustment parameters determined by the reinforcement learning algorithm, including specific values of support strength and support angle;

[0045] The optimal adjustment parameters are encoded into electronically or mechanically executable signals, and formatted into electronic pulses or wireless signals compatible with the interface of the adjustable support device.

[0046] Furthermore, the adjustable support device reinforces the electric pole by utilizing the generated control signal, including:

[0047] A control signal is received, and the supporting force and direction of the adjustable supporting device are adjusted according to the control signal.

[0048] This application has the following beneficial technical effects:

[0049] (1) By monitoring environmental and structural data (such as temperature, humidity, stress, and tilt) in real time and dynamically adjusting the support device based on this data, this method can significantly enhance the stability and safety of power poles in extreme weather conditions and unstable geological environments. This real-time response capability ensures that the power poles remain stable when facing environmental stresses such as strong winds, rainfall, or ground movement. (2) Using advanced neural network prediction models to analyze information obtained from sensors and historical data, it is possible to predict the maximum stress points and environmental changes that the power poles may encounter. (3) Through the precise control strategy generated by the reinforcement learning algorithm, the adjustable support device can operate in an optimal manner, making resources more efficiently utilized. This intelligent adjustment not only improves operational efficiency, but also significantly reduces long-term maintenance costs caused by excessive or improper use of resources. (4) The implementation of this method makes the power pole system highly adaptive, able to automatically adjust its structural configuration based on real-time monitoring data and continuously updated environmental forecasts. This adaptive capability makes the power pole system more intelligent and responsive, and can effectively cope with rapidly changing environmental conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1This is a flow chart of a method for adaptive environmental adjustment and reinforcement of electric poles provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0051] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0052] The first embodiment of the present application provides a method for adaptive environmental adjustment and reinforcement of electric poles. Figure 1 , which is a flow chart of the first embodiment of this application. Figure 1 A first embodiment of the present application provides a method for adaptively adjusting and reinforcing a pole in an environmentally friendly manner, which is described in detail.

[0053] S101: Multiple sensors are installed on the poles to monitor environmental structural data including temperature, humidity, stress and tilt data in real time.

[0054] Step S101 involves installing multiple sensors on the poles to monitor environmental structural data including temperature, humidity, stress, and tilt in real time. The implementation steps of this step are as follows:

[0055] First, select sensors suitable for installation on the poles. These sensors must be environmentally resistant to various weather conditions. Common sensor types include temperature sensors, humidity sensors, strain sensors, and tilt sensors. These sensors should provide highly accurate data to ensure reliable monitoring results. For example, temperature sensors can be thermocouples or thermistors, humidity sensors are often capacitive or resistive, strain sensors can be strain gauges, and tilt sensors can use inclinometers using microelectromechanical systems (MEMS) technology.

[0056] The sensor's installation location should be determined by the pole's structural characteristics and the nature of the required monitoring data. For example, a tilt sensor is typically mounted on the top of the pole to accurately measure tilt, while a strain sensor should be installed where the pole is likely to experience the greatest stress. Each sensor should be secured to the pole using an appropriate fixture, and its connection to the data collection system should be reliable and robust, such as using waterproof cables and connectors.

[0057] To achieve real-time data monitoring, the installed sensors need to be connected to a central processing unit (CPU), which is responsible for collecting and initially processing the data from each sensor. This typically involves converting analog signals to digital signals, and possibly filtering and adjusting gain. The data collection system should have wireless transmission capabilities to transmit data in real time to a remote monitoring center or directly to data processing equipment located near the pole.

[0058] Furthermore, to ensure system reliability, a regular maintenance and calibration program should be implemented to correct for sensor bias and drift. This includes physical inspection of the sensor, electrical performance testing, and replacement when necessary.

[0059] By precisely installing and configuring these sensors, the physical and environmental conditions of the poles can be continuously monitored, providing a scientific basis for subsequent data analysis and decision-making, ensuring the safety and stability of the pole structure.

[0060] S102: An adjustable support device is configured around the pole, wherein the adjustable support device is used to adjust the support force and direction according to the received control signal.

[0061] Adjustable support devices are key actuators for achieving adaptive environmental adjustment and reinforcement of power poles. These devices are typically installed around the perimeter of the pole to form an evenly distributed support system. Depending on the height of the pole and the expected environmental pressures it will withstand, three to six support devices are typically installed evenly around the pole. The bottom of each support device is connected to the ground through a solid foundation, which can be a concrete base or screw piles deep into the ground, depending on local soil conditions. The top of the support device is connected to different height positions of the pole, typically covering a range from 1 / 3 to 2 / 3 of the height from the ground, to provide comprehensive support and adjustment capabilities.

[0062] Each adjustable support device primarily consists of the following components: a base, telescopic mechanism, force adjustment system, directional adjustment system, connector, and control unit. The base provides a stable foundation for the entire device and is typically constructed of high-strength steel, ensuring stability in all weather conditions. The telescopic mechanism consists of two inner and outer sleeves, whose length is controlled by an electric or hydraulic system to adjust the support height. The force adjustment system includes a precision hydraulic cylinder or electric actuator, which precisely controls the pressure applied to the pole. The directional adjustment system consists of a multi-degree-of-freedom spherical joint, allowing the support device to rotate 360 degrees horizontally and adjust vertically within a certain angle, typically between 0 and 45 degrees. The connector, located at the top of the support device, features a specially designed curved structure that ensures a close fit with the pole surface while allowing a certain degree of sliding to accommodate slight movements. The control unit is the brains of the entire device, comprising a microprocessor, communication module, and various sensors. It receives control signals and coordinates the operation of various systems.

[0063] These adjustable support devices operate as follows: First, a control unit receives control signals from a central control system via a wired or wireless network. These signals contain information about the desired support force and direction. Upon receiving the signals, the control unit first evaluates the difference between the current state and the target state. It then simultaneously activates the force and direction adjustment systems to initiate adjustments.

[0064] For example, suppose the system detects that a strong wind is approaching and predicts that the pole may tilt to the southeast. The control system may send the following control signal: increase the support force of the support device on the northwest side of the pole to 2000N, and adjust its support direction to tilt 15 degrees to the southeast. After receiving this signal, the control unit of the support device will perform the following operations: First, the ball joint of the direction adjustment system will rotate so that the entire support arm tilts 15 degrees to the southeast. At the same time, the force adjustment system will increase the pressure of the hydraulic cylinder or extend the length of the electric push rod until the force sensor shows that the force applied to the pole reaches 2000N. During this process, the telescopic mechanism may also need to adjust its length appropriately to match the changes in force and direction.

[0065] Another example is that if stress concentration is predicted at the base of a pole, the system may send a signal to the bottom support device to increase its support force, while the top support device slightly reduces its force. Specifically, the control signal may instruct the bottom support device to increase its force from 1000N to 1500N, while the top support device reduces its force from 1000N to 800N. In this case, the control unit of each support device will precisely control its hydraulic system or electric actuator, gradually adjusting the force until the specified value is reached. At the same time, they may need to fine-tune the direction to ensure that the point of force application is always in the optimal position.

[0066] The entire adjustment process is continuous and smooth, with the support devices responding to changes in the control signal in real time. Each support device is equipped with high-precision force and angle sensors that monitor the actual support force and direction in real time. Through a feedback loop, the device continuously fine-tunes the control signal to ensure that the desired state is achieved. Furthermore, all support devices are coordinated via a communication network, ensuring their coordinated actions provide optimal support for the pole.

[0067] This design not only accommodates anticipated environmental changes but also enables rapid response to unexpected events. For example, if sensors suddenly detect an abnormal tilt in a pole, the system can adjust all support devices within seconds, providing emergency support and preventing the pole from collapsing. This flexible design also facilitates routine maintenance and repairs. Maintenance personnel can use the control system to temporarily relieve the load on a support device to allow for inspection or component replacement, while other support devices automatically adjust to compensate.

[0068] Overall, this adjustable support system provides comprehensive, adaptive support for utility poles through precise force control, flexible directional adjustment, and intelligent collaboration. It proactively adjusts support based on environmental changes and forecasts, significantly improving the stability and safety of the poles and providing a strong guarantee for the reliable operation of the power system.

[0069] S103: Using a pre-trained neural network prediction model, a prediction result is obtained based on real-time environmental structural data from sensors and meteorological, geological and pole performance data in a historical database; wherein the prediction result includes the maximum stress point and environmental changes that the pole may withstand in the short term in the future.

[0070] The pre-trained neural network prediction model plays a crucial role as the core intelligent unit of the entire adaptive environmental conditioning and reinforcement system. Its primary task is to analyze and predict data from multiple sources, providing a key basis for subsequent decision-making. Specifically, the model receives two main inputs: environmental and structural data collected in real time from sensors installed on the poles, including parameters such as temperature, humidity, stress, and tilt; and long-term meteorological records, geological information, and pole performance data stored in a historical database.

[0071] This neural network prediction model adopts a deep learning architecture and is mainly composed of four key modules: a data preprocessing module, a feature extraction module, a time series analysis module, and a multi-task prediction module. The data preprocessing module first cleans, standardizes, and aligns the input real-time data and historical data to ensure data quality and consistency. The feature extraction module uses a convolutional neural network (CNN) layer to capture spatial features and local patterns in the data, such as the spatial correlation of stress distribution. The time series analysis module uses a long short-term memory (LSTM) network, which enables the model to understand and predict long-term dependencies in time series data, such as temperature trends or periodic patterns. The multi-task prediction module is the output of the model, which simultaneously predicts multiple target variables, including the location and size of the maximum stress point that the pole may bear in the short term in the future, as well as changes in environmental parameters (such as temperature, humidity, wind speed, etc.).

[0072] The model training process utilizes a large amount of historical data, including monitoring data from power poles across diverse geographic locations, seasons, and weather conditions. Advanced optimization algorithms, such as the Adam optimizer, are used for training, along with techniques such as batch normalization and dropout to improve the model's generalization and robustness. To address potential class imbalances in the data, such as the rarity of extreme weather events, the model also employs weighted sampling and focal loss.

[0073] During actual operation, the model receives the latest sensor data at regular intervals (for example, every 5 minutes). Each time new data is received, the model immediately makes a prediction, outputting hourly forecasts for the next 24 hours. The forecast consists of two main components: first, the point of maximum stress the pole may experience, including the magnitude (in Newtons), location (relative to the base of the pole), and estimated time of occurrence. Second, environmental variations, including maximum and minimum temperatures, humidity ranges, peak wind speeds, and wind direction trends.

[0074] A key feature of the model is its adaptive learning capability. It not only relies on pre-trained knowledge but also continuously learns and updates from new data. For example, if the model detects a significant deviation between its predictions and the ground truth, it automatically triggers an online learning process to fine-tune the model parameters using the new data. This mechanism ensures that the model can adapt to both the peculiarities of the local environment and long-term climate change.

[0075] To enhance forecast reliability, the model also incorporates an uncertainty estimation module. This module uses Monte Carlo dropout techniques to estimate confidence intervals for forecasts through multiple sampling. This means the model not only provides point estimates but also provides a confidence range for the forecast, which is particularly valuable for subsequent decision making.

[0076] Another important feature of the model is its interpretability. By integrating visualization tools based on the attention mechanism, system administrators can intuitively understand why the model makes a certain prediction. For example, it can show which historical data has the greatest impact on the current prediction, or which sensor data plays a decisive role in the prediction.

[0077] In practice, the model's predictions are directly fed into the next processing module, the reinforcement learning algorithm, to generate specific control strategies. The predictions are also displayed in easy-to-understand charts on the system's monitoring interface for human operator reference and oversight.

[0078] Overall, this pre-trained neural network prediction model provides a highly accurate and reliable prediction foundation for adaptive environmental adjustment and reinforcement of power poles by integrating multi-source data, leveraging advanced deep learning techniques, and incorporating innovative features such as adaptive learning and interpretability. It not only foresees potential risks but also provides precise guidance for proactive preventive measures, significantly improving the safety and reliability of power infrastructure.

[0079] Furthermore, the neural network prediction model includes:

[0080] A time series preprocessing module is used to extract features from real-time environmental structure data from sensors installed on power poles using a one-dimensional convolutional neural network structure to obtain feature data; and to process the feature data using a long short-term memory network to obtain preprocessed time series features;

[0081] The historical data fusion module is used to receive historical data from the historical database, including meteorological, geological and pole performance data, and perform weighted fusion of different types of historical data through the self-attention mechanism to obtain the fused historical feature representation;

[0082] The multimodal feature integration module is used to receive the preprocessed time series features provided by the time series preprocessing module and the fused historical feature representation provided by the historical data fusion module; interact and fuse the preprocessed time series features with the fused historical feature representation to generate an integrated multimodal feature representation;

[0083] The prediction output module is implemented using a multi-head attention mechanism and a feedforward neural network to receive the multimodal feature representation provided by the multimodal feature integration module; and through a multi-task learning method, it predicts the maximum stress point and environmental changes that the pole may withstand in the short term in the future. The environmental changes include changes in temperature, humidity and wind speed.

[0084] Crucial is the neural network prediction model, which complexly combines multiple data processing and machine learning techniques to extract useful information from real-time and historical data to predict stress points and environmental changes on the poles.

[0085] The main task of the time series preprocessing module is to extract key features from the real-time data collected by various sensors installed on the poles. This includes data such as temperature, humidity, stress and tilt. The data is first processed by a one-dimensional convolutional neural network (CNN), which can effectively capture the spatial correlation between various sensor data. The CNN layer helps identify features that may have a significant impact on the stability of the pole by learning local patterns in the data. The processed data is then passed to a long short-term memory network (LSTM), which is a network specially designed to process time series data. LSTM is able to remember long-term dependencies through its internal gating mechanism, thereby analyzing the temporal dynamics in the data, such as the changing trends of temperature and stress over time.

[0086] The historical data fusion module receives data stored in a historical database, such as past meteorological, geological, and pole performance data. This data is processed using a self-attention mechanism, a technique that performs weighted information fusion within a sequence, enabling the model to focus on the historical information most relevant to the current prediction task. For example, if certain meteorological conditions have caused pole damage in the past, the model will learn to be more alert when it encounters similar conditions.

[0087] The multimodal feature integration module integrates the features output by the time series preprocessing module and the historical data fusion module. This integration is accomplished through feature layer concatenation and further processing, ensuring that the current state of real-time data and the deeper insights from historical data contribute to the final prediction results.

[0088] The prediction output module utilizes a multi-head attention mechanism and a feedforward neural network. These techniques, combined, extract key information from multimodal feature representations for the final prediction. The multi-head attention mechanism allows the model to consider information from multiple perspectives when processing input features, enhancing its expressive power. Ultimately, this module outputs the maximum stress point that the pole is likely to experience in the short term, along with predicted environmental changes such as temperature, humidity, and wind speed.

[0089] The following is the reference implementation code of the neural network prediction model:

[0090] import tensorflow as tf

[0091] from tensorflow.keras.models import Model

[0092] from tensorflow.keras.layers import Input,Conv1D,LSTM,Dense,Flatten,Attention,Concatenate

[0093] #Assume that the input data has been preprocessed according to the following dimensions

[0094] #Real-time environment structure data dimensions: (batch_size, time_steps, sensor_features)

[0095] #Historical data dimensions: (batch_size, history_steps, history_features)

[0096] #Time series preprocessing module

[0097] def create_time_series_model(input_shape):

[0098] """Build a time series preprocessing module to process real-time sensor data using 1D convolution and LSTM"""

[0099] inputs=Input(shape=input_shape)#Input layer

[0100] x = Conv1D(32, kernel_size = 3, activation = 'relu')(inputs) #1D convolutional layer

[0101] x=LSTM(64,return_sequences=True)(x)#LSTM layer

[0102] model = Model(inputs,x)

[0103] return model

[0104] #Historical data fusion module

[0105] def create_history_model(input_shape):

[0106] """Build a historical data fusion module and use the self-attention mechanism to process historical data"""

[0107] inputs=Input(shape=input_shape)#Input layer

[0108] #Self-attention layer

[0109] query=Dense(64,activation='relu')(inputs)

[0110] value=Dense(64,activation='relu')(inputs)

[0111] attention=Attention()([query,value])#Attention mechanism

[0112] model=Model(inputs,attention)

[0113] return model

[0114] #Prediction output module

[0115] def create_prediction_model(time_series_shape,history_shape):

[0116] """Build the entire prediction model, including multimodal feature integration and multi-task learning prediction output"""

[0117] #Time series model and historical data model

[0118] time_series_input=Input(shape=time_series_shape)

[0119] history_input = Input(shape = history_shape)

[0120] time_series_model = create_time_series_model(time_series_shape)

[0121] history_model = create_history_model(history_shape)

[0122] # Model integration

[0123] time_series_features = time_series_model(time_series_input)

[0124] history_features = history_model(history_input)

[0125] # Feature fusion

[0126] combined_features = Concatenate()([Flatten()(time_series_features), Flatten()(history_features)])

[0127] # Predict the maximum stress point and environmental changes in the short - term future

[0128] stress_prediction = Dense(1, activation = 'linear', name ='stress_output')(combined_features)

[0129] environment_prediction = Dense(3, activation = 'linear', name = 'environment_output')(combined_features)

[0130] model = Model(inputs = [time_series_input, history_input], outputs = [stress_prediction, environment_prediction])

[0131] return model

[0132] #Build and compile the model

[0133] model=create_prediction_model((100,4),(30,3))

[0134] model.compile(optimizer='adam',loss='mse',metrics=['accuracy'])

[0135] #Model Overview

[0136] model.summary()

[0137] The key to training a neural network prediction model, especially one designed to predict stress points and environmental changes on utility poles, lies in precisely executing several steps, from data preparation to model training, evaluation, and ultimately deployment. The following details this process.

[0138] First, data needs to be collected and prepared. For time series data, this means collecting data from sensors on the poles, such as temperature, humidity, strain, and tilt sensors. This data needs to be acquired in real time, and its quality and integrity must be guaranteed. For historical data, historical meteorological, geological, and pole performance data must be extracted from storage systems. This historical data provides essential context, helping the model understand long-term trends and cyclical patterns.

[0139] After data preparation is complete, the next step is data preprocessing. This includes normalization to make data from different sources comparable, as well as filling in missing values or removing outliers to ensure data quality. Time series data may also require windowing, which involves dividing the continuous data into a series of time windows based on the time scale required for prediction.

[0140] This is followed by the design and implementation of the model. The model consists of four main parts:

[0141] 1. Time Series Preprocessing Module: This module uses a one-dimensional convolutional neural network (CNN) to extract spatial features within each time window, and then uses a long short-term memory network (LSTM) to capture temporal dependencies. This not only extracts key information from the sensor data but also maintains the time series characteristics of the data.

[0142] 2. Historical Data Fusion Module: This module uses the self-attention mechanism to process long-term historical data. The self-attention mechanism allows the model to identify and focus on the most relevant parts of the historical data, thereby better integrating this information.

[0143] 3. Multimodal Feature Integration Module: In this step, the model combines the time series preprocessed data with the output of the historical data fusion module. This integration approach ensures that information from both real-time and long-term data sources is considered simultaneously.

[0144] 4. Prediction Output Module: Finally, using a multi-head attention mechanism and a feedforward neural network, the model predicts future environmental changes and stress points on the poles. Multi-task learning allows the model to simultaneously predict multiple outputs, such as the maximum stress point and related environmental variables (temperature, humidity, and wind speed).

[0145] When training a model, choosing the right loss function and optimizer is crucial. Because the model outputs multiple predictions, a loss function is often needed to evaluate the accuracy of all predictions simultaneously. You can use mean squared error (MSE) as the loss function and the Adam optimizer for parameter optimization.

[0146] During the training process, the validation dataset is used to evaluate the model performance, adjust hyperparameters, and make necessary model modifications to prevent overfitting. Once the model performance is satisfactory, the test set can be used for final evaluation.

[0147] Finally, after model training is complete, it is deployed and integrated into the power pole monitoring system. This ensures that the model can continuously receive real-time data and output prediction results in a timely manner, supporting adaptive environmental adjustment and reinforcement of the power poles.

[0148] Furthermore, the historical data fusion module uses the following formula 1 to calculate the fused historical feature representation F:

[0149]

[0150] Among them, D i represents the i-th specific type of historical data; w i It represents the weight coefficient of the i-th specific type of historical data and is calculated by the following formula 2:

[0151]

[0152] Where H represents the high-dimensional feature vector set of all historical data extracted from the historical database; D i represents the i-th specific type of historical data; D j represents the jth specific type of historical data; score(H,D) is a scoring function used to evaluate the correlation between H and the specific type of historical data D, calculated using the following formula 3:

[0153] score(H,D)=V T tanh(W hH+W d D+b)(3)

[0154] Among them, V is the weight matrix used to transform the comprehensive historical feature representation; W h is the transformation matrix used to transform the historical feature matrix H; W d is the transformation matrix used to transform the specific historical data type D; b is the bias vector.

[0155] The historical data fusion module plays a crucial role in the adaptive environmental conditioning and reinforcement method for power poles. It is used to synthesize comprehensive feature representations of historical data, which is crucial for the overall system's predictive capabilities. This module operates through a series of computational steps, using specially designed formulas to calculate the fused historical features. The following details the components of each formula and their functions to ensure that those skilled in the art can understand and implement this module.

[0156] First, Formula 1 is used to calculate the fused historical feature representation F. This feature representation is obtained by combining different types of historical data D i The weighted sum is obtained, where the weight w of each data type is i Indicates the importance of this type of data in the prediction model. Specifically, D i It refers to the i-th specific type of historical data, such as past temperature records, wind speed or geological data.

[0157] Weight coefficient w i It is calculated using Formula 2, which uses a softmax function to ensure that all weights sum to 1, thus forming a probability distribution. This weighting method allows the model to automatically adjust the influence of each data type based on the relevance of historical data. In Formula 2, score(H,D) is a scoring function that quantifies the correlation between the overall historical data feature vector set H and the specific historical data type D.

[0158] Formula 3 defines how to calculate score(H,D). This scoring function uses bilinear transformation and hyperbolic tangent activation function. Specifically, W h and W d are transformation matrices used to transform the historical feature matrix H and the specific type of historical data D into a common feature space. Nonlinearity is then added using the hyperbolic tangent function tanh, enabling the model to capture complex relationships. The bias vector b adjusts the output of the activation function, while the vector V converts this activation result into the final score. These matrices and the bias vector can be obtained through training.

[0159] These calculation steps ensure that the historical data fusion module can dynamically adjust the weights of the data types based on their actual contribution to the prediction task. In this way, the model can not only utilize all available historical data, but also ensure that more critical information has a greater impact on the prediction results.

[0160] Furthermore, the multimodal feature integration module uses a graph neural network-based structure to process and fuse different feature sources, specifically including:

[0161] The time series features and historical features are represented as nodes in the graph, where each node represents a feature vector, V i Corresponding to the time series characteristics, V j Corresponding historical characteristics;

[0162] Information is passed between nodes through the graph neural network, and the following formula 4 is used as the update function to iteratively update the state of each node:

[0163]

[0164] in, represents the feature representation of node i at layer l+1; represents the feature representation of node i at layer l; represents the set of neighbor nodes of node i; c ij is a normalization constant, representing the number of neighbors of node i; W (l) The weight parameters learned in layer l; represents the feature representation of node j at layer l; B (l) is the bias parameter learned in layer l; σ is the nonlinear activation function sigmoid;

[0165] The updated features of all nodes are aggregated into a single graph-level representation using the following formula 5 for subsequent feature processing:

[0166]

[0167] Among them, F integrated is the integrated multimodal feature representation; READOUT is the aggregation function, which is implemented by summation; L represents the last layer of the graph network; G represents the entire graph; Represents a set including the feature representations of all nodes in the graph G at the last layer L.

[0168] The design of a multimodal feature integration module is crucial in the adaptive environmental adjustment and reinforcement method for power poles. It utilizes a graph neural network (GNN)-based architecture to process and fuse multiple features from time series and historical data. This architecture enables the module to effectively leverage complex relationships and dependencies, leading to more accurate predictions of potential risks and environmental changes facing power poles.

[0169] First, we represent time series and historical features as nodes in a graph, with each node representing a feature vector. In this graph structure, time series and historical feature nodes are connected by edges, indicating the underlying relationship between them. This representation allows graph neural networks to capture and exploit the complex interactions between features through the connections between nodes.

[0170] The core of graph neural networks is the node state update mechanism, implemented through an iterative update function that takes into account information about each node and its neighbors. Specifically, the new state of a node is calculated based on a weighted sum of its previous state and the states of its neighbors. The contribution of each neighbor node is adjusted using a weight parameter and a normalization constant to ensure that the update is not affected by uneven numbers of neighbors.

[0171] The update function's specific expression involves the node's own feature representation and that of its neighbors. Neighborhood information is aggregated onto the current node through a weight matrix, and then adjusted using a bias parameter. This process also includes a nonlinear activation function, typically a sigmoid function, to increase the model's expressiveness and introduce nonlinear processing capabilities.

[0172] At each layer of a graph neural network, the features of all nodes are updated in the same manner as described above, until the predetermined number of layers is reached. At the final layer of the network, an aggregation function, typically a summation function, is used to aggregate the final features of all nodes into a graph-level representation. This integrated feature representation encompasses comprehensive information from all nodes in the entire graph, providing comprehensive data support for the final prediction task.

[0173] Implementing this multimodal feature integration module requires careful design of the graph structure, selection of appropriate weight parameters and activation functions, and adjustment of the network depth and width to suit specific application requirements. Furthermore, training the entire graph neural network requires a large amount of data and computing resources to ensure that the model can accurately capture and utilize the complex relationships between various features.

[0174] Furthermore, the prediction output module is specifically used to:

[0175] Receiving the integrated multimodal feature representation provided by the multimodal feature integration module;

[0176] Processing the multimodal feature representation using a multi-head attention mechanism, specifically comprising dividing the input multimodal feature representation into n heads, each head corresponding to a subspace; for each head, respectively calculating a query vector, a key vector, and a value vector; within each head, calculating an attention weight using scaled dot product attention; and concatenating the outputs of all heads and performing a linear transformation to obtain an output of the multi-head attention;

[0177] The output of the multi-head attention is fed into two parallel feedforward neural networks, which are used to predict the maximum stress point and environmental changes respectively. The two parallel feedforward neural networks include:

[0178] The maximum stress point prediction network consists of two hidden layers, uses the ReLU activation function, and the output layer uses the linear activation function;

[0179] The environmental change prediction network consists of three parallel sub-networks that predict changes in temperature, humidity, and wind speed respectively.

[0180] The prediction output module first receives the integrated multimodal feature representations provided by the multimodal feature integration module. These feature representations contain information extracted and integrated from time series data and historical data, providing a comprehensive data foundation for the final prediction.

[0181] Next, these multimodal features are fed into a multi-head attention mechanism for processing. The multi-head attention mechanism is designed to enable the model to simultaneously acquire information from multiple subspaces, thereby enhancing the model's ability to integrate information. Specifically, the input feature representation is first divided into several heads, each of which processes a subset of the features, and each head independently calculates the query vector, key vector, and value vector. These vectors are weighted by calculating the attention weights using scaled dot product attention, which involves taking the dot product of the query vector and the key vector, and then weighting the value vector based on the obtained score. In this way, each head can capture different types of dependencies.

[0182] After all heads have completed their calculations, their outputs are concatenated and subjected to a linear transformation. This step integrates the information from each head to form the final output of the multi-head attention. This output contains feature information comprehensively considered from multiple perspectives, providing rich input for the subsequent prediction task.

[0183] Finally, the output of the multi-head attention is fed into two parallel feedforward neural networks, each responsible for a different prediction task. The first network is the maximum stress point prediction network, which consists of two hidden layers using the ReLU activation function and an output layer using a linear activation function. This network focuses on predicting the maximum physical stress points that the pole may experience, and its output is the location and intensity of the stress points. The second network is the environmental change prediction network, which consists of three sub-networks running in parallel, predicting changes in temperature, humidity, and wind speed. Each sub-network may also include several hidden layers and an output layer specialized for its respective environmental variable.

[0184] With such a structure, the prediction output module can effectively process and utilize the integrated multimodal features, and simultaneously make accurate predictions on the physical safety and environmental adaptability of the poles.

[0185] S104: Based on the prediction results of the neural network prediction model and the current state of the adjustable support device, a reinforcement learning algorithm is used to generate a control strategy for the adjustable support device.

[0186] In the pole adaptive environmental adjustment and reinforcement method, step S104 involves using a reinforcement learning algorithm to generate a control strategy based on the results of the neural network prediction model and the current state of the adjustable support device. The goal of this step is to optimize the pole's response so that it remains stable and safe in the face of different environmental challenges.

[0187] First, a reinforcement learning environment needs to be set up, which typically involves defining a state space, an action space, and a reward function. In this application, the state space includes the predictions from the neural network model in step S103, such as the maximum stress point on the pole, environmental changes, and real-time data from sensors on the support device, such as the device's current position and stress level.

[0188] The action space defines all possible actions that the adjustable support can perform, such as adjusting support force and changing support angle. These actions are the specific parameters for how they affect the stability of the pole, and each action is translated into actual mechanical adjustments by the control system.

[0189] The reward function is a key component of reinforcement learning, measuring the effectiveness of a particular action. In this case, the reward function might be designed based on the degree to which the action improves the stability of the pole, such as reducing the deviation from the predicted stress point or maintaining or improving the verticality of the pole. Cost factors could also be considered, such as the energy consumption of the action and potential long-term maintenance costs.

[0190] Next, we need to select and implement a suitable reinforcement learning algorithm. Given the complexity of the pole system and the continuous action space, a suitable algorithm might be Deep Deterministic Policy Gradient (DDPG) or similar algorithms, which are well suited to handling continuous action spaces and can learn how to generate the optimal action from the current state.

[0191] In implementation, a reinforcement learning algorithm learns an optimal policy by interacting with its environment. This typically involves performing a series of actions, observing the outcomes, and adjusting the policy based on feedback from a reward function. Specifically, the algorithm repeatedly tests different support adjustment strategies in a simulated environment, evaluating the change in pole stability after each adjustment. In this way, it gradually learns how to optimize the pole's response under various environmental conditions.

[0192] Finally, after algorithm training, the generated control strategies are used to adjust the pole's support in real time to address various challenges in the real world. These strategies are converted into control signals and directly input into the support's control system, enabling precise force and direction adjustments.

[0193] In the context of adaptive pole reinforcement, the generated control strategy is a set of specific operational instructions that tells the adjustable supports around the pole how to adjust to maximize the structure's stability and respond to impending environmental changes. This strategy is learned from historical and real-time data using a reinforcement learning algorithm, designed to anticipate and address future challenges.

[0194] To illustrate this with a practical example, consider a power pole located in a windy area. A neural network prediction model, based on current data collected from sensors (such as tilt and wind speed) and historical weather patterns, predicts the likelihood of high wind speeds in the coming hours. A reinforcement learning algorithm then processes this information to generate a control policy, which might involve increasing the pressure on specific supports or adjusting the angles of certain supports to counteract the predicted wind forces.

[0195] The specific operation might be as follows: If the forecast model indicates that northerly winds will exert greater pressure on the pole, the control strategy might instruct the north-facing support device to increase its support force, while simultaneously reducing the south-facing support force to maintain overall balance. These operational instructions are converted into electronic signals and sent directly to the support device's control system. Once the control system receives the instructions, the hydraulic or electric manipulator arm will adjust accordingly to ensure that the pole remains stable under the influence of wind.

[0196] This strategy is based not only on predictions of future conditions but also on the current state of the pole and similar events in historical data. This makes every operational decision information-based and data-driven, improving the efficiency and effectiveness of responding to unexpected environmental changes. By continuously monitoring environmental changes and adjusting control strategies in real time, the safety and functional performance of the poles are significantly enhanced.

[0197] Furthermore, the state space of the reinforcement learning algorithm includes real-time data obtained from multiple sensors and the maximum stress points and environmental changes that the poles may be subjected to in the short term in the future obtained from the prediction model.

[0198] Furthermore, the action space of the reinforcement learning algorithm includes support force adjustment and direction change of the adjustable support device.

[0199] Reinforcement learning algorithms play a central role in adaptive pole conditioning and reinforcement, dynamically adjusting support structures to respond to changing environmental and structural stresses. This approach involves a complex decision-making process in which the algorithm optimizes the pole's support configuration based on real-time data and predictions.

[0200] First, the state space of a reinforcement learning algorithm defines the basis for its decision-making. In this case, this state space comprises two main types of data: real-time data from multiple sensors on the poles, such as temperature, humidity, tilt, and stress; and data from predictive models, primarily forecasting the potential short-term maximum stress points on the poles and impending environmental changes, such as wind speed fluctuations and temperature fluctuations. This data provides the algorithm with the necessary input to assess the current state of the poles and potential future risks.

[0201] Next, the definition of an action space guides the algorithm in how to respond to identified risks and stresses. In this pole reinforcement method, the action space includes manipulation of adjustable support devices, specifically adjustments to support strength and changes to support orientation. For example, if the prediction model indicates that the pole will experience higher wind pressure on the north side, the reinforcement learning algorithm might decide to increase the strength of the north support device or adjust its angle to better withstand the wind pressure.

[0202] Implementing this reinforcement learning strategy requires precise control of the mechanical adjustments of the support device and requires the algorithm to make fast and accurate decisions based on the data in the state space. In addition, the algorithm should be able to continuously learn from the results of its operations and optimize its decision model to deal with future situations more effectively. This usually involves using one or more reinforcement learning methods, such as Q-learning or deep deterministic policy gradient (DDPG), which can handle complex state and action spaces and iteratively improve the policy through interaction with the environment.

[0203] In practice, ensuring that data from all sensors and predictive systems is accurately and updated in real time to the reinforcement learning algorithm is crucial. This requires a highly integrated system monitoring and data processing platform. Furthermore, the generation and transmission of control signals must be efficient and reliable to ensure that each adjustment precisely complies with the algorithm's instructions.

[0204] Furthermore, the reinforcement learning algorithm uses the reward function R(s, a, s) provided by the following formula 6: ′ ):

[0205]

[0206] Where Δstability represents the change in pole stability after taking action a, specifically from state s to state s ′ The stability difference; Δtime represents the time from state s to state s ′ The time required is used to measure the rapidity of the adjustment response; energy_cost represents the energy consumption of performing action a, including the energy required to adjust the strength and direction of the support device; maintenance_cost represents the long-term maintenance cost, reflecting the equipment wear and maintenance costs caused by frequent adjustments; predicted_deviation represents the deviation between the predicted results and the actual environmental changes, which is used to evaluate the accuracy of the prediction model; k1, k2, k3, and k4 are adjustment coefficients.

[0207] The reinforcement learning algorithm used in the adaptive environmental adjustment and reinforcement method for power poles includes a reward function, which aims to optimize the stability of the pole while taking into account cost and efficiency. The design of the reward function reflects the comprehensive consideration of multiple key factors. The following is a detailed explanation of each component of this reward function and its specific meaning.

[0208]

[0209] Δstability refers to the change in the stability of the pole after taking a specific action. This means that the pole changes from the current state s to the next state s ′ The specific change in pole stability can be any positive or negative change.

[0210] When discussing pole stability calculation methods, Δstability (the change in pole stability) is a key metric used to measure the degree of change in pole stability after a specific control action is performed. This calculation involves comparing the stability of the pole between different states, which are defined by a series of influencing factors such as structural stress, tilt angle, and force distribution. The following are the basic steps and considerations for calculating Δstability:

[0211] First, it is necessary to define quantitative indicators of pole stability, which may include:

[0212] Structural stress analysis: By calculating the stress distribution of the pole under load, it is determined whether the pole will exceed its physical bearing limit.

[0213] Tilt measurement: Measure the tilt angle of the pole to determine if there is any tilt beyond a safe range.

[0214] Vibration characteristics: Analyze the response of the pole to wind loads and other dynamic loads, such as vibration frequency and amplitude.

[0215] Then, measure the initial and post-response states:

[0216] Initial state(s): Before taking any control action, collect data such as stress, tilt, and vibration of the pole.

[0217] Post-response state (s'): After the control action is performed, the same stability indicator is re-evaluated.

[0218] Next, calculate the change in stability:

[0219] Δstability is obtained by comparing the stability index of the pole before and after the control action. The specific calculation method may be:

[0220] Difference calculation: For each stability indicator, calculate its difference between state s and state s'. For example, if the tilt decreases from 5 degrees to 3 degrees, this indicates that the stability has improved.

[0221] Comprehensive evaluation: Combine the changes in all relevant stability indicators and use arithmetic or weighted average methods to obtain a comprehensive stability change.

[0222] This method can quantify the changes in pole stability after adopting a specific control strategy, providing a basis for further adjustment and optimization. This quantitative analysis is a key technical means to ensure that the pole system can maintain stability and safety under various environmental conditions.

[0223] Δtime represents the time required to go from state s to state s ′ The time required for the process to complete. This parameter measures the speed of the adjustment response. Faster responses are generally preferred because they can more quickly stabilize the structure and respond to possible emergencies. The coefficient k1 in this section adjusts the importance of stability improvement relative to time efficiency, and its value determines the influence of this term in the total reward function.

[0224]

[0225] energy_cost represents the energy consumed to perform action a, including the energy required to adjust the strength and direction of the support mechanism. This may involve electrical energy, hydraulic energy, or other forms of energy consumption. The square root is used here to ensure that the increase in energy cost does not linearly affect the reward, and the negative impact of the increase in cost on the reward gradually decreases at high costs. The coefficient k2 is used to balance the importance of energy cost in the reward function; a higher k2 value means that energy cost has a higher weight in the decision.

[0226] 3.k3·log(1+maintenance_cost)

[0227] Maintenance_cost reflects the wear and tear of equipment and the cost of maintaining it due to frequent adjustments. This includes long-term maintenance expenses, which may involve replacing parts or repairing damaged support systems. A logarithmic function is used for this cost to mitigate the extreme impact of higher maintenance costs on the reward, so that the marginal impact of cost increases on the reward gradually decreases. The coefficient k3 determines the weight of maintenance costs in the reward function.

[0228] 4.|predicted_deviation|

[0229] Predicted_deviation represents the deviation between the predicted result and the actual environmental changes. This is a measure of the accuracy of the prediction model; smaller deviations indicate more accurate predictions. This directly impacts the reward value; larger deviations reduce the reward, incentivizing the algorithm to reduce errors and improve prediction accuracy. The coefficient k4 adjusts the influence of prediction deviation on the reward.

[0230] By properly setting these coefficients k1, k2, k3, and k4, the reinforcement learning algorithm can achieve a balance between multiple factors (such as response speed, energy consumption, maintenance costs, and prediction accuracy), thereby optimizing the overall stability and operational efficiency of the pole. This multifaceted reward function design is a crucial step in implementing adaptive environmental conditioning and reinforcement methods for poles, requiring careful tuning and testing to ensure effectiveness in real-world applications.

[0231] S105: generating a control signal for adjustment and reinforcement according to the control strategy; and using the generated control signal, the adjustable support device reinforces the pole.

[0232] In the final, critical step of the adaptive pole adjustment and reinforcement method, the system accurately calculates and generates control signals for adjustment and reinforcement based on the previously generated control strategy. These signals are then transmitted to the adjustable support device, achieving dynamic reinforcement of the pole. This process, the execution phase of the entire system, directly impacts the stability and safety of the pole.

[0233] Generating control signals is a complex and sophisticated process. First, the system analyzes the control policy output by the reinforcement learning algorithm. This policy typically includes a series of action instructions tailored to different environmental conditions and predicted outcomes. The system then selects the most appropriate set of actions from this policy based on the current situation. It then converts these abstract action instructions into concrete numerical signals.

[0234] These numerical signals primarily encompass two aspects: support force and support orientation. For support force, the system calculates the precise force required for each support device, typically in Newtons. This calculation takes into account factors such as the weight of the pole, predicted wind speeds, and potential snow and ice loads. For support orientation, the system calculates the angle for each support device, typically expressed in polar coordinates, including both the horizontal azimuth and the vertical elevation.

[0235] The generated control signals aren't simple static instructions, but rather a dynamic sequence that changes over time. For example, if a severe storm is predicted to last 12 hours, the system might generate a sequence of adjustment instructions every 10 minutes for the next 12 hours. This dynamic adjustment capability allows the system to proactively respond to environmental changes rather than reactively reacting to situations that have already occurred.

[0236] The format and encoding of the control signals must match the hardware specifications of the adjustable support device. Typically, these signals are encoded into a standard digital communication protocol, such as Modbus or Industrial Ethernet. Each signal packet may contain information such as the support device's unique identifier, target force value, target direction angle, and execution timestamp. The system also incorporates a checksum or error detection code into the signal to ensure accurate and reliable signal transmission.

[0237] Once the signals are generated, the system transmits them to each adjustable support device via a pre-defined communication channel. This communication may be wired (e.g., industrial Ethernet cable) or wireless (e.g., industrial-grade Wi-Fi or 4G / 5G networks), depending on the site environment and reliability requirements. The system ensures that each support device accurately receives its control signal and requests a confirmation message from the support device to verify successful signal reception and interpretation.

[0238] When the adjustable support devices receive a control signal, they immediately begin to perform the corresponding adjustment action. Each support device is equipped with a precision actuator, usually including an electrically or hydraulically driven telescopic rod and a rotating mechanism. These mechanisms can accurately adjust the support force and direction according to the received signal. For example, if a support device receives a signal to increase the force from 1000N to 1500N and deflect the support angle 15 degrees from the vertical, its execution process may be as follows: First, the telescopic mechanism will slowly and accurately extend, while the internal force sensor will monitor the applied force in real time until it reaches 1500N. Then, the rotating mechanism will precisely rotate the support arm until the tilt angle reaches exactly 15 degrees. The whole process may only take a few seconds, but every tiny adjustment is carefully calculated.

[0239] While making adjustments, each support device continuously reports its current status to the central control system, including the actual force applied, the current support angle, and any possible anomalies. This real-time feedback mechanism allows the system to confirm whether the adjustment is performed as expected and make real-time fine-tuning when necessary.

[0240] It's worth noting that the adjustments of all support devices are coordinated. The system ensures that the movements of each support device coordinate with each other to achieve optimal support for the pole. For example, when a support device on one side increases its support force, the support device on the opposite side may correspondingly reduce its force to maintain the overall balance of the pole. This coordinated behavior is achieved through unified command from a central control system.

[0241] The entire adjustment and reinforcement process is continuous and dynamic. The system continuously generates new control signals based on the latest sensor data and predictions, enabling the support device to promptly respond to environmental changes. This continuous adaptive process ensures that the pole is always optimally stressed, significantly improving its stability and safety in a variety of harsh environments.

[0242] Through this precise control signal generation and execution mechanism, this method can achieve intelligent and automated reinforcement of poles, effectively respond to various complex environmental challenges, and significantly improve the reliability and lifespan of power infrastructure.

[0243] Furthermore, generating a control signal for adjustment and reinforcement according to the control strategy includes:

[0244] Obtain the optimal adjustment parameters determined by the reinforcement learning algorithm, including specific values of support strength and support angle;

[0245] The optimal adjustment parameters are encoded into electronically or mechanically executable signals, and formatted into electronic pulses or wireless signals compatible with the interface of the adjustable support device.

[0246] In the adaptive environmental adjustment and reinforcement method for poles, generating control signals for adjusting and reinforcing the poles is a key step in achieving stability regulation. This process ensures an effective transition from theoretical decisions to practical execution, precisely controlling the support device to respond to predicted and real-time monitored environmental and structural changes.

[0247] First, based on the output of the reinforcement learning algorithm, the optimal parameters for adjusting the pole's support devices are determined. These parameters include specific values for support force and support angle, which the algorithm arrives at by evaluating multiple possible actions and selecting the one it ultimately believes will maximize the long-term reward. For example, the algorithm might determine that, under specific wind speed and tilt conditions, increasing the force of the north support and adjusting the angle of the east support are most effective in stabilizing the pole.

[0248] After determining the optimal adjustment parameters, the next task is to convert them into control signals that the pole support device can understand and execute. This step involves encoding the parameter values into electronically or mechanically executable signals, such as electronic pulses or wireless signals. The encoding process ensures that the signal format is compatible with the support device's interface so that the device can correctly interpret and respond to these signals.

[0249] During the encoding process, specific hardware or software tools may be required to convert these parameter values. For example, if the support device receives wireless signals, a transmitter is required to convert the digital parameters into wireless frequency signals; if the device operates based on wired electrical control, the parameters must be converted into corresponding voltage or current changes.

[0250] The generated control signal is sent to the support device via a predetermined communication method. During the signal transmission process, the accuracy of the signal and the real-time transmission must be ensured to prevent delays or errors from affecting the adjustment effect.

[0251] After receiving the control signal, the support device adjusts its force and angle according to the received instructions. These adjustments are made in real time, requiring the support system to have high response speed and precise control capabilities.

[0252] Through the above steps, the theoretically optimal adjustment strategy derived from the reinforcement learning algorithm can be translated into specific operational measures to effectively support and reinforce the poles. The successful implementation of this process relies on accurate data processing, signal encoding technology, and the efficient execution of the support device. Ensuring the precise coordination of these steps is key to implementing the adaptive environmental adjustment and reinforcement method for poles.

[0253] Furthermore, the adjustable support device reinforces the electric pole by utilizing the generated control signal, including:

[0254] A control signal is received, and the supporting force and direction of the adjustable supporting device are adjusted according to the control signal.

[0255] In the pole adaptive environmental adjustment and reinforcement method, the correct implementation of the control signal is crucial to ensure that the adjustable support device can effectively adjust and reinforce the pole according to the predetermined strategy.

[0256] The control system for an adjustable support device must first be able to receive and interpret control signals. These control signals may be transmitted in a variety of forms, including but not limited to wired signals, radio waves, or data packets transmitted over a network. Each signal transmission method requires corresponding hardware interfaces and software protocols to ensure accurate signal reception.

[0257] The receiving module of the support device should be equipped with corresponding sensors and receivers, such as wireless receivers or cable connection ports. These devices must be able to handle high-frequency data transmission to meet the needs of immediate adjustment when the environment changes rapidly.

[0258] Once the control signal is received, the built-in microprocessor or control unit must immediately decode the signal. During this process, the control unit will analyze the specific adjustment instructions for the support force and direction.

[0259] Adjusting the support force may involve changing the hydraulic pressure, motor torque, or other mechanical force output in the support device. Adjusting the direction may involve changing the physical position of the support point or adjusting the angle between the support device and the pole.

[0260] After adjustments are made, the support device should be able to monitor its status in real time and provide feedback to the central control system. This feedback mechanism ensures that the effectiveness of the adjustment measures can be monitored in real time and allows the system to make fine adjustments when necessary to adapt to changes in the environment or load.

[0261] Feedback data should include, but is not limited to, adjusted support forces and angles, as well as pole stability data. This information is crucial for evaluating the effectiveness of current measures and making timely adjustments.

[0262] By implementing the detailed steps above, the adaptive environmental adjustment and reinforcement method for poles ensures the stability and safety of poles in a variety of environmental conditions. Implementing these steps requires precise control system design and efficient signal processing algorithms, as well as a high degree of mechanical precision and reliability in the support device. Proper execution of these operations significantly improves the pole system's responsiveness and adjustment efficiency, effectively reducing potential risks caused by environmental factors.

[0263] A second embodiment of the present application provides an electronic device, comprising:

[0264] processor;

[0265] The memory is used to store a program, and when the program is read and executed by the processor, the program executes the pole adaptive environmental adjustment and reinforcement method provided in the first embodiment of the present application.

[0266] The third embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for adaptive environmental adjustment and reinforcement of electric poles provided in the first embodiment of the present application is executed.

[0267] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A method for adaptive environmental adjustment and reinforcement of electric poles, characterized in that: include: Multiple sensors are installed on the poles to monitor environmental structural data including temperature, humidity, stress and tilt data in real time; An adjustable support device is arranged around the pole, and the adjustable support device is used to adjust the support force and direction according to the received control signal; Using a pre-trained neural network prediction model, the system generates predictions based on real-time environmental structural data from sensors and meteorological, geological, and pole performance data from a historical database. These predictions include the maximum stress points and environmental changes that the poles may experience in the short term. Based on the prediction results of the neural network prediction model and the current state of the adjustable support device, a reinforcement learning algorithm is used to generate a control strategy for the adjustable support device; According to the control strategy, a control signal for adjustment and reinforcement is generated; and the adjustable support device reinforces the electric pole using the generated control signal.

2. The method for adaptive environmental adjustment and reinforcement of electric poles according to claim 1, characterized in that: The neural network prediction model includes: A time series preprocessing module is used to extract features from real-time environmental structure data from sensors installed on power poles using a one-dimensional convolutional neural network structure to obtain feature data; and to process the feature data using a long short-term memory network to obtain preprocessed time series features; The historical data fusion module is used to receive historical data from the historical database, including meteorological, geological and pole performance data, and perform weighted fusion of different types of historical data through the self-attention mechanism to obtain the fused historical feature representation; The multimodal feature integration module is used to receive the preprocessed time series features provided by the time series preprocessing module and the fused historical feature representation provided by the historical data fusion module; interact and fuse the preprocessed time series features with the fused historical feature representation to generate an integrated multimodal feature representation; The prediction output module is implemented using a multi-head attention mechanism and a feedforward neural network to receive the multimodal feature representation provided by the multimodal feature integration module; and through a multi-task learning method, it predicts the maximum stress point and environmental changes that the pole may withstand in the short term in the future. The environmental changes include changes in temperature, humidity and wind speed.

3. The method for adaptive environmental adjustment and reinforcement of electric poles according to claim 2, characterized in that: The historical data fusion module uses the following formula 1 to calculate the fused historical feature representation F: Among them, D i represents the i-th specific type of historical data; w i It represents the weight coefficient of the i-th specific type of historical data and is calculated by the following formula 2: Where H represents the high-dimensional feature vector set of all historical data extracted from the historical database; D i represents the i-th specific type of historical data; D j represents the jth specific type of historical data; score(H,D) is a scoring function used to evaluate the correlation between H and the specific type of historical data D, calculated using the following formula 3: score(H,D)=V T tanh(W h H+W d D+b)(3) Among them, V is the weight matrix used to transform the comprehensive historical feature representation; W h is the transformation matrix used to transform the historical feature matrix H; W d is the transformation matrix used to transform the specific historical data type D; b is the bias vector.

4. The method for adaptive environmental adjustment and reinforcement of electric poles according to claim 2, characterized in that: The multimodal feature integration module uses a graph neural network-based structure to process and fuse different feature sources, specifically including: The time series features and historical features are represented as nodes in the graph, where each node represents a feature vector, V i Corresponding to the time series characteristics, V j Corresponding historical characteristics; Information is passed between nodes through the graph neural network, and the following formula 4 is used as the update function to iteratively update the state of each node: in, represents the feature representation of node i at layer l+1; represents the feature representation of node i at layer l; represents the set of neighbor nodes of node i; c ij is a normalization constant, representing the number of neighbors of node i; W (l) The weight parameters learned in layer l; represents the feature representation of node j at layer l; B (l) is the bias parameter learned in layer l; σ is the nonlinear activation function sigmoid; The updated features of all nodes are aggregated into a single graph-level representation using the following formula 5 for subsequent feature processing: Among them, F integrated is the integrated multimodal feature representation; READOUT is the aggregation function, which is implemented by summation; L represents the last layer of the graph network; G represents the entire graph; Represents a set including the feature representations of all nodes in the graph G at the last layer L.

5. The method for adaptive environmental adjustment and reinforcement of electric poles according to claim 2, characterized in that: The prediction output module is specifically used for: Receiving the integrated multimodal feature representation provided by the multimodal feature integration module; Processing the multimodal feature representation using a multi-head attention mechanism, specifically comprising dividing the input multimodal feature representation into n heads, each head corresponding to a subspace; for each head, respectively calculating a query vector, a key vector, and a value vector; within each head, calculating an attention weight using scaled dot product attention; and concatenating the outputs of all heads and performing a linear transformation to obtain an output of the multi-head attention; The output of the multi-head attention is fed into two parallel feedforward neural networks, which are used to predict the maximum stress point and environmental changes respectively. The two parallel feedforward neural networks include: The maximum stress point prediction network consists of two hidden layers, using the ReLU activation function, and the output layer uses a linear activation function; The environmental change prediction network consists of three parallel sub-networks that predict changes in temperature, humidity, and wind speed respectively.

6. The method for adaptive environmental adjustment and reinforcement of electric poles according to claim 1, characterized in that: The state space of the reinforcement learning algorithm includes real-time data obtained from multiple sensors and the maximum stress points and environmental changes that the pole may be subjected to in the short term in the future obtained from the prediction model.

7. The method for adaptively adjusting and reinforcing a power pole according to claim 6, characterized in that: The action space of the reinforcement learning algorithm includes support force adjustment and direction change of the adjustable support device.

8. The method for adaptively adjusting and reinforcing a power pole according to claim 7, characterized in that: The reinforcement learning algorithm uses the reward function R(s, a, s) provided by the following formula 6 ′ ): Where Δstability represents the change in pole stability after taking action a, specifically from state s to state s ′ The stability difference; Δtime represents the time from state s to state s ′ The time required is used to measure the rapidity of the adjustment response; energy_cost represents the energy consumption of performing action a, including the energy required to adjust the strength and direction of the support device; maintenance_cost represents the long-term maintenance cost, reflecting the equipment wear and maintenance costs caused by frequent adjustments; predicted_deviation represents the deviation between the predicted results and the actual environmental changes, which is used to evaluate the accuracy of the prediction model; k1, k2, k3, and k4 are adjustment coefficients.

9. The method for adaptively adjusting and reinforcing a power pole according to claim 1, characterized in that: Generating a control signal for adjustment and reinforcement according to the control strategy includes: Obtain the optimal adjustment parameters determined by the reinforcement learning algorithm, including the specific values of support strength and support angle; The optimal adjustment parameters are encoded into electronically or mechanically executable signals, and formatted into electronic pulses or wireless signals compatible with the interface of the adjustable support device.

10. The method for adaptively adjusting and reinforcing a power pole according to claim 1, characterized in that: The adjustable support device reinforces the electric pole by utilizing the generated control signal, including: A control signal is received, and the supporting force and direction of the adjustable supporting device are adjusted according to the control signal.

Images