An RFID-based power supply management method for a power tower inclination sensor

Abstract

The application provides an RFID-based power tower inclination sensor power supply management method, including: using a support vector machine and a decision tree to analyze historical operation data of a tower, establishing a tower fault prediction model, predicting a state of the tower through the tower fault prediction model, and obtaining a probability and a time of a tower fault; according to a tower fault prediction result, formulating a preventive maintenance plan, arranging priority maintenance for a tower with a fault probability greater than a preset threshold, and adjusting a power supply strategy in advance for a tower predicted to have a fault to reduce load pressure; using a tower group intelligent power supply management system to integrate state information and a power supply strategy of a single tower into an overall power grid, dynamically adjusting power supply loads of each tower through a load balancing algorithm, and reducing a tower power supply load for a tower in a dangerous state to transfer the load to a tower in a good state.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and particularly relates to a power tower tilt sensor power supply management method based on RFID. BACKGROUND

[0002] In the power tower tilt sensor power supply management method based on RFID, there is a key technical problem to be solved. Currently, the tower index and the tilt ratio are considered as two independent parameters, which leads to the failure to timely discover the tower abnormalities and the existing risks, resulting in one-sidedness of the power supply management decision, so as to fail to prevent power interruption and safety accidents caused by tower collapse. The tower index reflects the overall state of the tower, including material aging, structural integrity and other factors; and the tilt ratio directly reflects the tilting degree of the tower. However, there is a complex mutual influence relationship between the two parameters, and separate consideration of each parameter may lead to misjudgment of the actual condition of the tower. For example, a tower may have a high index, indicating that the overall condition is good, but at the same time, there may be a large tilt ratio. Conversely, a tower with a small tilt ratio may have a low index due to long-term environmental erosion. This complex relationship makes it difficult for the traditional power supply management method to accurately evaluate the actual condition of the tower, thereby affecting the formulation of the power supply strategy. SUMMARY

[0003] The present application provides a power tower tilt sensor power supply management method based on RFID, mainly comprising:

[0004] The tilt angle sensor with an RFID tag is used to collect the tower tilt angle data, the collected tilt angle data is preprocessed by a data fusion algorithm to filter out abnormal values and noise, and the real-time tilt angle data of the tower is obtained, the tilt angle of the tower is graded according to a preset tilt angle threshold, and the tower is divided into three levels of normal, warning and danger;

[0005] The historical operation data of the tower is obtained by the tilt angle sensor with an RFID tag, including the service life, material quality and bearing capacity of the tower, combined with the current environmental factors including wind speed, temperature and humidity, a tower index evaluation model is established, the comprehensive index of the tower is calculated according to the tower index evaluation model, and the overall state of the tower is reflected;

[0006] According to the tilt angle level and the comprehensive index of the tower, a tower state evaluation matrix is established, if the tilt angle level is normal and the comprehensive index is higher than the threshold, the tower state is determined to be good, if the tilt angle level is warning or the comprehensive index is lower than the threshold, the tower state is determined to be attention, and if the tilt angle level is dangerous or the comprehensive index is far lower than the threshold, the tower state is determined to be dangerous;

[0007] Different tower states are formulated with corresponding power supply management strategies, for the tower in good state, the conventional power supply mode is adopted, for the tower needing attention, the power supply monitoring frequency is increased, the power supply parameters are adjusted, for the tower in dangerous state, the emergency power supply mode is started, the power supply power is limited, and the early warning signal is sent out;

[0008] Support vector machine and decision tree are adopted to analyze the historical operation data of the tower, a tower fault prediction model is established, the state of the tower is predicted through the tower fault prediction model, and the probability and time of the tower fault are obtained;

[0009] According to the tower fault prediction result, a preventive maintenance plan is formulated, for the tower with a fault probability greater than a preset threshold, priority repair is arranged, and for the tower predicted to fail, the power supply strategy is adjusted in advance to reduce the load pressure;

[0010] The state information and power supply strategy of a single tower are integrated into the overall power grid by adopting the tower group intelligent power supply management system, the power supply load of each tower is dynamically adjusted through the load balancing algorithm, and for the tower in dangerous state, the tower power supply load is reduced, and the load is transferred to the tower in good state.

[0011] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0012] The power supply management method of the power tower inclination sensor based on RFID provided by the application can realize real-time monitoring of the tower inclination, timely identification of structural changes, prevention of large-scale power interruption caused by tower inclination or collapse, early identification of potential fault points through the fault prediction model, preventive maintenance, reduction of unplanned power outage, improvement of the stability and reliability of the power grid, reasonable arrangement of maintenance resources according to the tower state evaluation matrix, ensuring that limited maintenance resources are preferentially used for towers in poor state, improving the maintenance efficiency, predicting the probability and time of fault occurrence, making the maintenance plan more targeted, avoiding excessive maintenance or insufficient maintenance, saving costs and prolonging the service life of the tower, adopting differentiated power supply management strategies for towers in different states, maintaining the continuity and stability of power supply, improving user satisfaction, limiting the power supply power on the tower in dangerous state to avoid the expansion of the influence range caused by the fault, and ensuring the overall power supply balance of the power grid through the load balancing algorithm, improving the accuracy and comprehensiveness of data collection, providing high-quality basic data for intelligent power grid data analysis, integrating the state information and power supply strategy of a single tower, realizing intelligent management and optimization of the power grid level, promoting the digitalization and intelligentization transformation of the power system, combining environmental factors for comprehensive evaluation, improving the adaptability and response capability of the system to adverse weather conditions, and deeply analyzing the historical operation data to establish a tower index evaluation model that can better understand the performance of the tower under different environments and assist risk management decisions. BRIEF DESCRIPTION OF DRAWINGS

[0013] Fig. 1 A flow chart of a power supply management method for an RFID-based power tower inclination sensor of the present application.

[0014] Fig. 2 A schematic diagram of a power supply management method for an RFID-based power tower inclination sensor of the present application. DETAILED DESCRIPTION

[0015] The technical solutions of the present application will be described in detail below with reference to the embodiments, obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0016] As Figs. 1-2 , the power supply management method for an RFID-based power tower inclination sensor of the present embodiment can specifically include:

[0017] S101, collect tower inclination data through an inclination sensor with an RFID tag, preprocess the collected inclination data through a data fusion algorithm, filter out abnormal values and noise, and obtain real-time inclination data of the tower; according to a pre-set inclination threshold, classify the tower inclination into three levels: normal, warning and danger.

[0018] Obtain tower inclination data through an inclination sensor with an RFID tag, and perform data fusion on the inclination data through a Kalman filtering algorithm to obtain real-time inclination data of the tower; according to a pre-set inclination threshold, classify the real-time inclination data, use a K-means clustering algorithm to classify the inclination data, and determine the current inclination state of the tower; for the inclination state of the tower, establish a warning model based on inclination rate of change, calculate the inclination trend through an exponential smoothing method, and determine whether the tower has a continuous inclination risk; according to the inclination state of the tower and the inclination trend, use a C4.5 decision tree algorithm to generate a tower inclination risk assessment result, and quantitatively assess the inclination risk from tower structure safety information and line operation factors; if the tower inclination risk assessment result exceeds a pre-set threshold, send warning information to relevant personnel through SMS and email, and realize real-time monitoring and warning of tower inclination.

[0019] Specifically, the inclination data of the tower is collected by the inclination sensor with RFID tag. The original inclination data is fused by Kalman filtering algorithm to filter out abnormal values and random noise, and the real-time inclination data of the tower is obtained. In the Kalman filtering process, the measurement noise covariance is determined by historical data statistical analysis, and the filtering accuracy is improved by dynamically adjusting the process noise covariance. According to the pre-set inclination threshold, the real-time inclination data is divided into three levels: normal, warning and dangerous. The K-means clustering algorithm is used to classify the inclination data and determine the current inclination state of each tower. In K-means clustering, the maximum inclination is selected as the initial clustering center, and the Euclidean distance is used to calculate the distance between data points and clustering centers. When the clustering center changes less than the pre-set threshold, the iteration is terminated. For the classified tower inclination state, a warning model based on inclination rate of change is established. The exponential smoothing method is used to calculate the inclination trend to determine whether the tower has a continuous inclination risk. In the exponential smoothing process, the appropriate smoothing parameter α is selected, where 0<α<1. The optimal α value is determined according to the characteristics of historical data, and the prediction period is set to 7 days. According to the tower inclination state and trend, the C4.5 decision tree algorithm is used to generate the tower inclination risk assessment results. The inclination risk is quantitatively evaluated from the aspects of tower structure safety, line operation influence, etc., and the tower inclination risk level is obtained. In building the C4.5 decision tree, the inclination trend is used as an important feature variable, and the information gain ratio is used as the feature selection standard. The pessimistic pruning method is used to optimize the decision tree. According to the risk assessment results, an automatic alarm mechanism is established. When the tower inclination risk level exceeds the pre-set threshold, warning information is sent to relevant personnel through various ways such as SMS and email, realizing the real-time monitoring and warning function of tower inclination. In the tower inclination monitoring system, the three-axis acceleration sensor and three-axis gyroscope sensor are used to collect the inclination data of the tower. The sampling frequency is set to 100 Hz. The original inclination data is fused by Kalman filtering algorithm to filter out abnormal values and random noise. In Kalman filtering, the measurement noise covariance is obtained by variance analysis of 1000 groups of historical data, and the initial value is set to 0.01. The initial value of process noise covariance is set to 0.001, and it is dynamically adjusted according to the residual sequence. The accuracy of the real-time inclination data obtained after filtering is improved to 0.01°. According to the pre-set inclination threshold, such as normal 0-1°, warning 1-3°, and dangerous >3°, the K-means clustering algorithm is used to classify the inclination data. In K-means clustering, the maximum inclination 3°, the median value 1.5° and the minimum value 0° are selected as the initial clustering center. The Euclidean distance is used to calculate the distance between data points and clustering centers. When the clustering center changes less than 0.01°, the iteration is terminated, usually 3-5 iterations are needed to converge. For the classified tower inclination state, the exponential smoothing method is used to calculate the inclination trend.The smoothness parameter a = 0.3 is selected, based on 30 days of historical data, to predict the inclination change in the next 7 days. If the predicted inclination change in the next 7 days exceeds 0.5°, it is determined that there is a risk of continuous inclination. The C4.5 decision tree algorithm is used to generate the tower inclination risk assessment results, with inclination, inclination change rate, tower height, foundation type, etc. as characteristic variables. The information gain ratio is used as the feature selection standard, the minimum sample size is set to 10, the maximum tree depth is set to 5, and the pessimistic pruning method is used to optimize the decision tree. According to the risk level output by the decision tree, including low, medium and high, when the risk level is "high", warning information is sent to the maintenance personnel through SMS and email, including tower number, current inclination, predicted inclination, etc. Key information is used to realize real-time monitoring and early warning of tower inclination.

[0020] S102, obtain the historical operation data of the tower according to the inclination sensor with the RFID tag, including the age, material and bearing capacity of the tower, combine the current environmental factors including wind speed, temperature and humidity, and establish a tower index evaluation model; calculate the comprehensive index of the tower according to the tower index evaluation model to reflect the overall state of the tower.

[0021] Obtain the historical operation data of the tower in the database and the real-time environmental factor data collected by the weather station, the historical operation data of the tower including the age, material and bearing capacity of the tower, and the real-time environmental factor data including wind speed, temperature and humidity; perform minimum-maximum normalization processing on the historical operation data of the tower and the real-time environmental factor data to obtain normalized data, the normalization processing using the formula: Xnormalized = (X-Xmin) / (Xmin-Xmax); construct a tower index evaluation model using a random forest algorithm, the number of decision trees of the random forest algorithm being a preset value, the random forest algorithm using Gini index as a feature selection standard, the maximum depth of each tree of the random forest algorithm being a preset depth value, and the minimum leaf node sample number of the random forest algorithm being a preset sample number; divide the tower state into three levels of normal, attention and danger according to the tower comprehensive index output by the tower index evaluation model, in combination with a preset normal threshold and a dangerous threshold, if the tower comprehensive index is less than the dangerous threshold, trigger a warning signal, and send the tower number, comprehensive index value and current environmental data to the monitoring center.

[0022] Specifically, historical operation data of the tower, including the age, material, and load-bearing capacity of the tower, are obtained from a database, and current environmental factors such as wind speed, temperature, and humidity data are collected in real time through a weather station. The obtained historical operation data and real-time environmental data are subjected to minimum-maximum normalization processing to unify different dimensional data to the range of 0 to 1, with the calculation formula being Xnormalized = (X-Xmin) / (Xmax-Xmin). A random forest algorithm is used to construct a tower index evaluation model, with the number of decision trees being set to 100, the Gini index being used as the feature selection standard, the maximum depth of each tree being set to 10, and the minimum leaf node sample number being 5. The normalized data are input into the tower index evaluation model, and the comprehensive index of the tower is predicted through the combination of multiple decision trees. K-fold cross-validation is used to evaluate the performance of the tower index evaluation model, with K being set to 5, the mean square error and the determination coefficient being calculated, and the performance of the tower index evaluation model being determined to be good if the mean square error is less than 0.1 and the determination coefficient is greater than 0.8. According to the tower comprehensive index output by the random forest algorithm, the tower state is divided into normal, attention, and danger levels in combination with preset thresholds, with the normal threshold being 0.7 and the danger threshold being 0.3. These thresholds are obtained through statistical analysis of historical data. For towers in the danger level, a warning signal is triggered, and the tower number, comprehensive index value, and current environmental data are sent to the monitoring center. The tower comprehensive index is updated every 1 hour in combination with real-time monitoring data to realize dynamic evaluation and warning functions. In the tower index evaluation system, tower operation data, including tower age such as 20 years, material such as Q345 steel, and load-bearing capacity such as 50 tons, are first extracted from a historical database. At the same time, environmental data such as wind speed 10 m / s, temperature 25℃, and humidity 60% are collected in real time through weather stations installed around the tower. These data are subjected to minimum-maximum normalization processing, for example, the 20-year tower age is normalized to 0.4, and the 50-ton load-bearing capacity is normalized to 0.8. Subsequently, a random forest algorithm is used to construct a tower index evaluation model, with 100 decision trees being set, the maximum depth of each tree being 10, and the minimum leaf node sample number being 5. The model input includes normalized tower attributes and environmental data, and the output is the tower comprehensive index. Through 5-fold cross-validation, the model performance is evaluated, the mean square error is calculated to be 0.08, and the determination coefficient is calculated to be 0.85, proving that the model performance is good. According to the tower comprehensive index output by the model, such as 0.65, the tower state is judged against preset thresholds, with normal being >0.7 and danger being <0.3. If a dangerous state is detected, the system automatically sends warning information to the monitoring center, including the tower number such as TW-001, the comprehensive index value 0.25, and the current environmental data. The system updates the tower comprehensive index every hour, for example, when the wind speed suddenly increases to 15 m / s, the newly calculated comprehensive index may decrease to 0.55, which has not yet reached the danger level but has entered the attention state, and the system will record this change and continue to monitor.The dynamic evaluation method can timely reflect the change of the tower state and provide timely and accurate early warning information for maintenance personnel.

[0023] S103, according to the tower inclination grade and the comprehensive index, a tower state evaluation matrix is established, if the inclination grade is normal and the comprehensive index is higher than the threshold value, it is determined that the tower state is good, if the inclination grade is warning or the comprehensive index is lower than the threshold value, it is determined that the tower state needs attention, and if the inclination grade is dangerous or the comprehensive index is much lower than the threshold value, it is determined that the tower state is dangerous.

[0024] A two-dimensional state evaluation matrix is established, the horizontal axis of the two-dimensional state evaluation matrix represents the inclination grade, the vertical axis represents the comprehensive index interval, and the corresponding tower state determination result is filled in the matrix cell; according to the two-dimensional state evaluation matrix, the comprehensive index threshold value is set, the tower comprehensive index data of the last year is extracted from the database, the mean and standard deviation of the comprehensive index data are calculated, the mean is subtracted by 1.5 times the standard deviation to obtain the comprehensive index threshold value; a tower state determination model is constructed by using C4.5 decision tree algorithm, the input variables of the tower state determination model are inclination grade and comprehensive index, the output variable is tower state, the maximum tree depth of the decision tree is set to 5, the minimum leaf node sample number is set to 10, and the information gain ratio is used as the splitting criterion; the real-time monitored tower inclination grade and comprehensive index are input into the tower state determination model to obtain the current tower state evaluation result, the tower current state evaluation result is divided into three levels of good, need attention and dangerous.

[0025] Specifically, a two-dimensional status assessment matrix is ​​established based on tower tilt angle data and comprehensive index data. The horizontal axis represents the tilt angle level, categorized as normal, warning, and dangerous. The vertical axis represents the comprehensive index range, with each cell filled with the corresponding tower status judgment result. A comprehensive index threshold is set. The most recent year's comprehensive index data is extracted from the database, and its mean and standard deviation are calculated. The mean minus 1.5 times the standard deviation is set as the threshold; values ​​below this threshold are considered abnormal. A C4.5 decision tree algorithm is used to construct the tower status judgment model. The input variables are tilt angle level and comprehensive index, and the output variable is tower status. The maximum tree depth is set to 5, and the minimum number of leaf node samples is 10. Information gain ratio is used as the splitting criterion, and optimal pruning parameters are selected through cross-validation. The real-time monitored tower tilt angle level and comprehensive index are input into the status judgment model to obtain the current tower status assessment result, which is divided into three levels: good, requiring attention, and dangerous. If the status is abnormal, it is checked every hour; if it is severely abnormal, it is checked every 10 minutes, and the emergency plan is immediately activated. In the tower condition assessment system, a 3x3 two-dimensional condition assessment matrix is ​​first constructed. The horizontal axis represents the tilt angle level (normal: 0-1°, warning: 1-3°, danger: >3°), and the vertical axis represents the comprehensive index range (high: >0.8, medium: 0.6-0.8, low: <0.6). The comprehensive index data for the most recent year is extracted from the database, and the mean is calculated to be 0.75 with a standard deviation of 0.1. A threshold of 0.6 is set. The C4.5 decision tree algorithm is used to construct the condition determination model, with a maximum tree depth of 5, a minimum leaf node sample size of 10, and information gain ratio as the splitting criterion. Optimal pruning parameters are selected using 10-fold cross-validation, with an alpha value set to 0.01. The hold-out method is used to divide 10,000 historical data points into a training set of 8,000 and a test set of 2,000. The model performance is evaluated on the test set, yielding an accuracy of 95%, precision of 93%, and recall of 94%. Real-time monitoring data showed that a certain tower had a tilt angle of 2.5° and a comprehensive index of 0.72, which was identified as an "abnormal" state after being input into the model. The system automatically adjusted the detection frequency to once every 30 minutes and sent a warning message to the monitoring center, including the tower number, current tilt angle, comprehensive index, and predicted cause of the anomaly. Simultaneously, the system activated pre-set emergency plans, such as increasing the monitoring of surrounding towers and preparing emergency repair teams. This dynamic assessment and multi-level early warning mechanism can promptly identify potential risks, providing strong protection for the safe operation of the power grid.

[0026] S104. Develop corresponding power supply management strategies for different tower conditions. For towers in good condition, adopt the conventional power supply mode; for towers that require attention, increase the power supply monitoring frequency and adjust the power supply parameters; for towers in dangerous condition, activate the emergency power supply mode, limit the power supply, and issue a warning signal.

[0027] The tower state level and power supply parameters are acquired, and a power supply management strategy matrix is established; according to the power supply management strategy matrix, a power supply parameter adaptive adjuster is designed by using a Mamdani fuzzy control algorithm, input variables of the power supply parameter adaptive adjuster are tower state indexes, and output variables are power supply voltage, current and power factor; according to an output result of the power supply parameter adaptive adjuster, a C4.5 decision tree is constructed as a power supply mode selector, input variables of the power supply mode selector include tower state, load demand and power grid operating condition, and an output variable is a power supply mode; the output result of the power supply mode selector is linked with actual power supply equipment through a SCADA system, and automatic power supply management is realized; according to the tower state and the power supply mode, a multi-level early warning signal release mechanism is designed, the multi-level early warning signal release mechanism determines an early warning level based on tower state indexes, power supply parameter deviation values and load change rates, and early warning information is pushed in real time through a short message, an email and a monitoring platform.

[0028] Specifically, a 3x3 power supply management strategy matrix is established according to the tower state evaluation results, with the horizontal axis representing tower state levels as good, attention needed, and dangerous, and the vertical axis representing power supply parameters including voltage, current, and power factor. The corresponding power supply management strategy is filled in the matrix cell. A Mamdani fuzzy control algorithm is used to design a power supply parameter adaptive adjuster, with tower state indicators as input variables and power supply voltage, current, and power factor as output variables. A triangular membership function is designed, 15 fuzzy rules are formulated, and the centroid method is used for defuzzification to achieve dynamic adjustment of power supply parameters. A C4.5 decision tree is constructed as a power supply mode selector, with tower state, load demand, and grid operating conditions as input variables and power supply modes including normal, adjustment, and emergency as output variables. The maximum tree depth is set to 5, the minimum leaf node sample size is set to 10, the information gain ratio is used as the splitting criterion, and the optimal pruning parameter is selected through cross-validation. The power supply mode selection results are sent to the SCADA system to control the actual power supply equipment, realizing automatic power supply management. A multi-level early warning signal release mechanism is designed to automatically generate early warning information based on tower state and power supply mode. The early warning levels are divided into four levels: green for normal, yellow for slight abnormality, orange for moderate abnormality, and red for serious abnormality. The early warning level is determined based on tower state indicators, power supply parameter deviation values, and load change rates. Early warning information, including early warning level, impact range, and response measures, is pushed in real time through SMS, email, and monitoring platforms. In the tower power supply management system, a 3x3 power supply management strategy matrix is first established, with the horizontal axis representing tower state levels such as good (0-0.3), attention needed (0.3-0.7), and dangerous (0.7-1), and the vertical axis representing power supply parameters such as voltage (±5%), current (±10%), and power factor (0.9-1). A Mamdani fuzzy control algorithm is used to design a power supply parameter adjuster, with tower state indicators as input variables and power supply parameter adjustments as output variables, where the tower state index ranges from 0 to 1. A triangular membership function is designed, with input divided into low, medium, and high levels, output divided into decrease, maintain, and increase levels, and 15 fuzzy rules formulated. For example, when the tower state is 0.6, the voltage adjustment is -2%, the current adjustment is -5%, and the power factor adjustment is 0.95 through fuzzy reasoning. A C4.5 decision tree is constructed as a power supply mode selector, with tower state, load demand, and grid operating conditions as input variables and power supply modes as output variables. 10,000 historical data are used for training, with the maximum tree depth set to 5, the minimum leaf node sample size set to 10, and the optimal pruning parameter selected through 5-fold cross-validation. The accuracy of the trained decision tree on the test set reaches 92%. The power supply mode selection results are sent to the SCADA system to send control instructions, such as switching to emergency power supply mode, which automatically reduces the power supply power to 70% of the rated value.The early warning signal issuing mechanism generates a red early warning signal according to the comprehensive judgment of the tower state index being 0.8, the power supply voltage deviation being -4%, and the load change rate being +15% / hour, pushes the early warning signal to the operation and maintenance personnel through a short message and an email, and displays an affected line diagram and recommended countermeasures on a monitoring platform.

[0029] In S105, a support vector machine and a decision tree are used to analyze historical operation data of the tower, and a tower fault prediction model is established. The state of the tower is predicted through the tower fault prediction model to obtain a probability and a time of a tower fault.

[0030] The historical operation data of the tower are obtained from a database, and the historical operation data include a tower inclination angle, a vibration frequency, a stress distribution, and environmental factors. Data denoising is performed through five-layer wavelet transform, a db4 wavelet basis is selected, and denoised tower operation data are obtained. Principal component analysis is performed according to the denoised tower operation data, a characteristic vector with a cumulative contribution rate reaching a preset threshold is selected, and a key characteristic vector of the tower state is determined. A support vector machine algorithm is used to construct a tower fault prediction model, a radial basis kernel function is selected, kernel function parameters and a penalty factor are optimized through a grid search method, an optimal model parameter is selected using a cross-validation method, importance analysis of the key characteristic vector is performed using a CART decision tree algorithm, a maximum tree depth and a minimum leaf node sample number are set, an optimal split feature is selected through a Gini index, and a feature importance ranking is obtained. The historical operation data are segmented using a sliding time window method, a window size and a step size are set, a multilayer LSTM network is used to construct a time series prediction model, an activation function is selected, a batch size and an initial value of a learning rate are set, an optimizer is used, a future state of the tower is predicted, and time series prediction results of tower indexes are obtained.

[0031] Specifically, the historical operating data of the tower, including the tower inclination, vibration frequency, stress distribution, and environmental factors, are obtained from the database. The data is denoised by five-layer wavelet transform, and the db4 wavelet basis is selected. Principal component analysis is performed on the denoised data, and the characteristic vectors with a cumulative contribution rate of 95% are selected to obtain the key feature vectors of the tower state. A support vector machine algorithm is used to construct a tower fault prediction model, and a radial basis kernel function is selected. The kernel function parameters γ are optimized by grid search method, with a range of 10^-3 to 10^3 and a step of 10, and the penalty factor C is in the range of 0.1 to 1000 with a step of 10. The optimal model parameters are selected using 5-fold cross-validation method. The CART decision tree algorithm is used for importance analysis of the features, with a maximum tree depth of 10 and a minimum leaf node sample size of 5. The optimal split feature is selected by Gini index, and the feature importance ranking is obtained. The historical data is segmented using the sliding time window method, with a window size of 30 days and a step of 1 day. A time series prediction model is constructed, using a three-layer LSTM network with 64 neurons in each layer and a ReLU activation function. The batch size is set to 32, and the initial learning rate is 0.001. The Adam optimizer is used to predict the future state of the tower, and the time series prediction results of each indicator of the tower are obtained. The outputs of the support vector machine and LSTM models are combined, and the probability distribution of the tower failure is calculated by Bayesian inference. The prior probability is set as the historical failure rate, and the posterior probability is updated by the likelihood function. When the predicted failure probability exceeds 0.8, the warning mechanism is triggered, and the time interval of the possible failure is estimated based on the time series prediction results. In the tower fault prediction system, the tower operating data for the past 5 years is extracted from the database, including the inclination 0-5°, vibration frequency 0-100 Hz, stress distribution 0-500 MPa, and environmental factors temperature -40-60℃, humidity 0-100%, and wind speed 0-50 m / s recorded every hour. The data is denoised by five-layer wavelet transform, and the db4 wavelet basis is selected. The signal-to-noise ratio of the denoised data is improved by about 20%. Through principal component analysis, the first five principal components are selected as the key feature vectors, with a cumulative contribution rate of 96.8%. The support vector machine is used to build the fault prediction model, and the kernel function parameters γ and the penalty factor C are optimized by grid search, with γ = 0.1 and C = 100. The 5-fold cross-validation accuracy is 94.3%. The CART decision tree is then used for feature importance analysis, with a tree depth of 8 and a minimum leaf node sample size of 10. The results show that the inclination change rate, vibration frequency, and wind speed are the three most important features affecting the tower failure, with importance scores of 0.35, 0.28, and 0.20, respectively. The sliding window method with a window size of 30 days and a step of 1 day is used to process the data, and a three-layer LSTM network with 64 neurons in each layer and a ReLU activation function is used. The batch size is set to 32, the initial learning rate is 0.001, and the Adam optimizer is used for training for 100 epochs. The root mean square error on the test set is 0.15.Finally, the fault probability is calculated by Bayesian inference combining the outputs of the support vector machine and the LSTM model, the prior probability is set as the historical average annual failure rate 0.05, and the pre-warning is triggered when the posterior fault probability of a certain tower exceeds 0.8. For example, for the tower numbered TW-001, the system predicts that the probability of its failure in the next 7-10 days is 0.85, and immediately sends a pre-warning message to the maintenance personnel, including the tower location, the predicted failure time interval and the recommended maintenance items.

[0032] S106, according to the tower failure prediction result, a preventive maintenance plan is made, for the tower with a failure probability greater than a preset threshold, a priority maintenance is arranged, and for the tower predicted to fail, a power supply strategy is adjusted in advance to reduce the load pressure.

[0033] A preventive maintenance priority matrix is constructed according to the fault probability interval and the predicted failure occurrence time, the horizontal axis of the matrix represents the fault probability interval, and the vertical axis represents the predicted failure occurrence time; a fuzzy comprehensive evaluation method is used to quantitatively evaluate the tower maintenance priority, the evaluation includes establishing an evaluation index system, determining the weight of each index by the analytic hierarchy process, constructing a fuzzy relationship matrix, and using the weighted average method for fuzzy synthesis operation; an optimal preventive maintenance plan is made by using the 0-1 knapsack dynamic programming algorithm, the algorithm takes the maintenance priority score as the value and the maintenance time as the weight to solve the optimal maintenance combination that meets the manpower resource and time window constraints; according to the preventive maintenance plan, for towers with different priorities, an adaptive load control algorithm is used to dynamically adjust the power supply strategy, the algorithm automatically calculates the load adjustment amount by real-time monitoring the tower load condition combined with weather forecast data using a fuzzy PID controller; if a tower is predicted to fail within a preset time range, the tower is additionally increased by the predicted load reduction amount, and the monitoring frequency is increased.

[0034] Specifically, a preventive maintenance priority matrix is constructed according to the fault probability and the predicted occurrence time output by the tower fault prediction model. The horizontal axis represents the fault probability interval, which includes three intervals: 0-0.3, 0.3-0.6, and 0.6-1. The vertical axis represents the predicted fault occurrence time, which includes three time ranges: 0-7 days, 7-30 days, and more than 30 days. The corresponding maintenance priority score, such as 1-9 points, is filled in the matrix cell. The fuzzy comprehensive evaluation method is used to quantitatively evaluate the tower maintenance priority. An evaluation index system is established, including fault probability, predicted fault time, tower importance, and maintenance cost. The analytic hierarchy process is used to determine the weight of each index, a fuzzy relationship matrix is constructed, and a weighted average method is used for fuzzy synthesis operation to calculate the comprehensive score of tower maintenance priority. The 0-1 knapsack dynamic programming algorithm is used to develop the optimal preventive maintenance plan, with the maintenance priority score as the value and the maintenance time as the weight. The optimal maintenance combination that meets the manpower resource and time window constraints is solved. The state transition equation is f[i][j] = max(f[i-1][j], f[i-1][j-w[i]]+v[i]), where i represents the tower number, j represents the available time, w[i] represents the time required to maintain tower i, and v[i] represents the maintenance priority score of tower i. According to the preventive maintenance plan, adaptive load control algorithm is used to dynamically adjust the power supply strategy for towers with different priorities. By monitoring the tower load in real time and combining with the 24-hour weather forecast data, a fuzzy PID controller is used to automatically calculate the load adjustment amount. For towers that may fail within 7 days, an additional 20% load reduction is added, and the monitoring frequency is increased to once every hour. In the tower preventive maintenance system, a 3x3 maintenance priority matrix is first constructed based on the output of the fault prediction model. For example, a tower with a fault probability of 0.7 and a predicted failure within 7 days scores 9 points. Then, the fuzzy comprehensive evaluation method is used to quantify the priority, with the fault probability weight set to 0.4, the predicted time weight set to 0.3, the tower importance weight set to 0.2, and the maintenance cost weight set to 0.1. The comprehensive score is obtained through fuzzy synthesis. For example, a tower with an index score of [0.8, 0.9, 0.7, 0.6] has a comprehensive score of 0.79. Then, the 0-1 knapsack dynamic programming algorithm is used to develop the maintenance plan. Assuming there are 10 towers to be maintained, the total available time is 100 hours, and the maintenance time of each tower varies from 8 to 15 hours. The optimal maintenance combination is solved through the state transition equation, and the number and order of the 6 towers that can be maintained within the time limit are obtained. Finally, adaptive load control is implemented for towers with different priorities. For example, for a tower with a comprehensive score of 0.79, based on the current load rate of 70% and the 24-hour high temperature forecast of 35°C, the fuzzy PID controller calculates that the load needs to be reduced by 15%.Considering that the tower predicts possible failure within 7 days, an additional 20% load reduction is added, a final 35% load reduction is performed, and the monitoring frequency is adjusted to once an hour to continuously track the tower state changes.

[0035] S107, adopt the tower group intelligent power supply management system, integrate the state information and power supply strategy of a single tower into the whole power grid, dynamically adjust the power supply load of each tower through the load balancing algorithm, and reduce the power supply load of the tower in danger, and transfer the load to the tower in good condition.

[0036] Collect real-time state information of the tower, the real-time state information including inclination, vibration, stress distribution and environmental factors; According to the real-time state information, multi-source data fusion is carried out through Dempster-Shafer evidence theory module to obtain the health index of the tower; According to the tower health index and the current load condition, a fuzzy control algorithm is used to formulate the power supply strategy of the tower, including setting the triangular membership function of the health index and the load rate, formulating the fuzzy rule base, using Mamdani inference mechanism, and obtaining the power supply power adjustment amount by solving the fuzzy through the barycenter method; Construct a power grid topology model based on graph theory, the tower as a node and the transmission line as an edge in the power grid topology model, and realize the minimum spanning tree by using Kruskal algorithm, the edge weight of the minimum spanning tree is defined as the weighted sum of line loss and tower health index; According to the path obtained by the minimum spanning tree, realize the load transfer of the tower in danger, including: calculating the maximum transmission capacity of each path, distributing the load to be transferred in proportion, and adjusting the load distribution through intelligent switch step by step; Adopt genetic algorithm to optimize the load balancing strategy of the whole power grid, the genetic algorithm including using real number coding to represent the load proportion of each tower in chromosome coding, generating fitness function as the inverse of the load variance between towers, roulette selection, selection operation, using arithmetic crossover and uniform mutation, and solving to obtain the globally optimal load distribution scheme.

[0037] Specifically, real-time state information of each tower is collected, including inclination, vibration, stress distribution and environmental factors, multi-source data fusion is carried out through Dempster-Shafer evidence theory module, basic probability distribution is calculated, combined with evidence synthesis rule, unified tower health index is obtained, and tower state information database is established. According to the tower health index and the current load condition, the fuzzy control algorithm is used to formulate the power supply strategy of single tower, the triangular membership function of health index and load rate is set, the fuzzy rule base is formulated, the Mamdani reasoning mechanism is adopted, and the appropriate power supply power adjustment amount is obtained by solving the fuzzy through the barycenter method. The power grid topology model based on graph theory is constructed, the tower is taken as the node and the transmission line as the edge, the Kruskal algorithm is used to realize the minimum spanning tree, the edge weight is defined as the weighted sum of line loss and tower health index, and the optimal allocation scheme of load among different towers is determined through depth first search. According to the path obtained by the minimum spanning tree, the load transfer of the tower in poor state is realized, the maximum transmission capacity of each path is calculated, the load to be transferred is allocated in proportion, and the load distribution is adjusted through intelligent switch step by step. Genetic algorithm is used to optimize the load balancing strategy of the whole power grid, real number coding is used to encode the load proportion of each tower, the fitness function is designed as the inverse of the load variance between towers, roulette wheel selection, selection operation, arithmetic crossover and uniform mutation are adopted, the population size is set to 100, and the iteration is 500 generations. The global optimal load allocation scheme is obtained. In the tower group intelligent power supply management system, firstly, the real-time state information of 100 towers is collected, including inclination 0-5°, vibration frequency 0-100Hz, stress distribution 0-500MPa and environmental factors including temperature-40-60℃, humidity 0-100%, and wind speed 0-50m / s. The Dempster-Shafer evidence theory module is used to fuse these data, and the basic probability of each index is allocated, such as inclination 0.3, vibration 0.2, stress 0.3, and environment 0.2. The tower health index is obtained through orthogonal operation. For example, the health index of a tower after fusion is 0.72. Then, the fuzzy control algorithm is used to formulate the power supply strategy, the triangular membership function of health index and load rate is set, such as health index "low" 0-0.4, "medium" 0.3-0.7, "high" 0.6-1, and load rate "low" 0-50%, "medium" 40%-80%, "high" 70%-100%. Through Mamdani reasoning, it is concluded that the tower should reduce the power supply power by 5%. Then, the power grid topology graph is constructed, which contains 100 nodes and 150 edges, the Kruskal algorithm is used to generate the minimum spanning tree, and the edge weight formula is 0.6×line loss+0.4×(1-tower health index). For the tower with health index of 0.72, three paths for load transfer are calculated, the maximum transmission capacity is 10MW, 8MW and 5MW respectively. According to the proportion of 6:5:3, 4MW of load to be transferred is allocated, and the load distribution is adjusted step by step through intelligent switch.Finally, the genetic algorithm is run to optimize load balancing, the chromosome length is 100, representing the load proportion of each tower, the population size is 100, the iteration is 500 generations, the crossover probability is 0.8, and the mutation probability is 0.1. The final scheme reduces the health index of the tower with 80% load from 0.72 to 65%, and increases the load of the tower with a higher health index from 60% to 70%, achieving load balancing of the overall power grid.

[0038] The above embodiment is only one of the preferred embodiments of the present application and should not be used to limit the protection scope of the present application, but any modification or polishing without substantial meaning made within the main design idea and spirit of the present application, which still solves the technical problems consistent with the present application, should be included in the protection scope of the present application.

Claims

1. A power supply management method for an RFID-based electric power tower inclination sensor, characterized by, The method comprises: collecting tower inclination data by an inclination sensor with an RFID tag, preprocessing the collected inclination data by a data fusion algorithm, filtering out abnormal values and noise to obtain real-time inclination data of the tower, and classifying the tower inclination into three levels of normal, warning and danger according to a preset inclination threshold; obtaining historical operation data of the tower including the age, material and bearing capacity of the tower by the inclination sensor with the RFID tag, combining current environmental factors including wind speed, temperature and humidity, establishing a tower index evaluation model, calculating the comprehensive index of the tower according to the tower index evaluation model to reflect the overall state of the tower; establishing a tower state evaluation matrix according to the tower inclination level and the comprehensive index, determining that the tower state is good if the inclination level is normal and the comprehensive index is higher than a threshold, determining that the tower state needs attention if the inclination level is warning or the comprehensive index is lower than the threshold, and determining that the tower state is dangerous if the inclination level is dangerous or the comprehensive index is much lower than the threshold; formulating corresponding power supply management strategies for different tower states, adopting a conventional power supply mode for the tower in good state, increasing the power supply monitoring frequency and adjusting the power supply parameters for the tower needing attention, and starting an emergency power supply mode for the tower in dangerous state, limiting the power supply power and sending a warning signal; analyzing the historical operation data of the tower by support vector machines and decision trees, establishing a tower fault prediction model, predicting the state of the tower by the tower fault prediction model to obtain the probability and time of tower failure; formulating a preventive maintenance plan according to the tower fault prediction result, arranging priority repair for the tower with a failure probability greater than a preset threshold, and adjusting the power supply strategy in advance for the tower predicted to fail to reduce the load pressure; integrating the state information and power supply strategy of a single tower into the overall power grid by using a tower group intelligent power supply management system, dynamically adjusting the power supply load of each tower by a load balancing algorithm, and reducing the power supply load of the tower in dangerous state to transfer the load to the tower in good state, including: collecting real-time state information of the tower, the real-time state information including inclination, vibration, stress distribution and environmental factors; obtaining a health index of the tower by a Dempster-Shafer evidence theory module for multi-source data fusion according to the real-time state information; formulating a power supply strategy of the tower by a fuzzy control algorithm according to the tower health index and the current load condition, including setting a triangular membership function of the health index and load rate, formulating a fuzzy rule base, using a Mamdani inference mechanism, and obtaining a power supply power adjustment amount by defuzzification through a barycenter method; constructing a power grid topology model based on graph theory, taking the tower as a node and the transmission line as an edge in the power grid topology model, realizing a minimum spanning tree by Kruskal algorithm, and defining the edge weight of the minimum spanning tree as the weighted sum of line loss and tower health index.According to the path obtained by the minimum spanning tree, load transfer of the state dangerous tower is realized, including: calculating the maximum transmission capacity of each path, proportionally distributing the load to be transferred, and adjusting the load distribution through intelligent switches step by step; the genetic algorithm is used to optimize the load balancing strategy of the whole power grid, the genetic algorithm includes that the chromosome coding adopts real number coding to represent the load proportion of each tower, a fitness function is generated for the inverse of the load variance between towers, roulette is selected, selection operation is adopted, arithmetic crossover and uniform mutation are adopted, and the globally optimal load distribution scheme is obtained.

2. The method of claim 1, wherein, The inclination sensor with an RFID tag is used to collect the inclination data of the tower, the collected inclination data is preprocessed by a data fusion algorithm to filter out abnormal values and noise, and the real-time inclination data of the tower is obtained, the inclination of the tower is classified according to a preset inclination threshold, and the inclination data is classified by using a K-means clustering algorithm, and the current inclination state of the tower is determined; an early warning model based on the inclination change rate is established for the inclination state of the tower, the inclination change trend is calculated by an exponential smoothing method, and it is judged whether the tower has a continuous inclination risk; According to the inclination state of the tower and the inclination change trend, a C4.5 decision tree algorithm is used to generate a tower inclination risk assessment result, and the inclination risk is quantitatively evaluated from the tower structure safety information and the line operation factors; If the tower inclination risk assessment result exceeds a preset threshold, an early warning information is sent to relevant personnel through SMS and email, and real-time monitoring and early warning of tower inclination are realized.

3. The method of claim 1, wherein, The historical operation data of the tower with an RFID tag is obtained, including the age, material and bearing capacity of the tower, and combined with the current environmental factors including wind speed, temperature and humidity, a tower index evaluation model is established, and the comprehensive index of the tower is calculated according to the tower index evaluation model to reflect the overall state of the tower, including: obtaining the historical operation data of the tower in the database and the real-time environmental factor data collected by the weather station, the historical operation data of the tower including the age, material and bearing capacity of the tower, and the real-time environmental factor data including wind speed, temperature and humidity; the minimum-maximum normalization processing is performed according to the tower historical operation data and the real-time environmental factor data, and the normalized data is obtained, the normalization processing uses the formula: Xnormalized=(X-Xmin) / (Xmax-Xmin); a random forest algorithm is used to construct a tower index evaluation model, the number of decision trees of the random forest algorithm is a preset value, the Gini index is used as the feature selection standard of the random forest algorithm, the maximum depth of each tree of the random forest algorithm is a preset depth value, and the minimum leaf node sample number of the random forest algorithm is a preset sample number; according to the tower comprehensive index output by the tower index evaluation model, the tower state is divided into three levels of normal, attention and danger according to the preset normal threshold and danger threshold, and if the tower comprehensive index is less than the danger threshold, an early warning signal is triggered, and the tower number, comprehensive index value and current environmental data are sent to the monitoring center.

4. The method of claim 1, wherein, The tower state evaluation matrix is established according to the tower inclination level and the comprehensive index, if the inclination level is normal and the comprehensive index is higher than the threshold value, the tower state is determined to be good, if the inclination level is warning or the comprehensive index is lower than the threshold value, the tower state is determined to be attention, if the inclination level is dangerous or the comprehensive index is much lower than the threshold value, the tower state is determined to be dangerous, including: a two-dimensional state evaluation matrix is established, the horizontal axis of the two-dimensional state evaluation matrix represents the inclination level, the vertical axis represents the comprehensive index interval, and the corresponding tower state determination result is filled in the matrix cell; the comprehensive index threshold value is set according to the two-dimensional state evaluation matrix, the tower comprehensive index data of the last year is extracted from the database, the mean value and the standard deviation of the comprehensive index data are calculated, and the mean value is reduced by 1.5 times the standard deviation to obtain the comprehensive index threshold value; a tower state determination model is constructed by using a C4.5 decision tree algorithm, the input variables of the tower state determination model are the inclination level and the comprehensive index, the output variable is the tower state, the maximum tree depth of the decision tree is set to 5, the minimum leaf node sample number is set to 10, and the information gain ratio is used as the splitting criterion; the tower inclination level and the comprehensive index monitored in real time are input into the tower state determination model to obtain the current tower state evaluation result, and the tower current state evaluation result is divided into three levels of good, attention and danger.

5. The method of claim 1, wherein, Different power supply management strategies are formulated for different tower states, for the tower in good state, the conventional power supply mode is adopted, for the tower needing attention, the power supply monitoring frequency is increased, the power supply parameters are adjusted, for the tower in dangerous state, the emergency power supply mode is started, the power supply power is limited, and the warning signal is sent, including: the tower state level and the power supply parameters are obtained, and a power supply management strategy matrix is established; according to the power supply management strategy matrix, a Mamdani fuzzy control algorithm is used to design a power supply parameter adaptive adjuster, the input variable of the power supply parameter adaptive adjuster is the tower state index, and the output variable is the power supply voltage, current and power factor; according to the output result of the power supply parameter adaptive adjuster, a C4.5 decision tree is constructed as a power supply mode selector, the input variables of the power supply mode selector include the tower state, the load demand and the power grid operating condition, and the output variable is the power supply mode; The output result of the power supply mode selector is linked with the actual power supply equipment through the SCADA system to realize automatic power supply management; according to the tower state and the power supply mode, a multi-level warning signal release mechanism is designed, the multi-level warning signal release mechanism determines the warning level based on the tower state index, the power supply parameter deviation value and the load change rate, and the warning information is pushed in real time through SMS, email and monitoring platform.

6. The method of claim 1, wherein, The support vector machine and the decision tree are used to analyze historical operation data of the tower, a tower fault prediction model is established, the state of the tower is predicted through the tower fault prediction model, and the probability and time of tower failure are obtained, including: obtaining tower historical operation data from a database, the historical operation data including tower inclination, vibration frequency, stress distribution and environmental factors; data denoising is performed through five-layer wavelet transform, a db4 wavelet basis is selected, and denoised tower operation data is obtained; principal component analysis is performed according to the denoised tower operation data, a characteristic vector with a cumulative contribution rate reaching a preset threshold is selected, and a key characteristic vector of the tower state is determined; a support vector machine algorithm is used to construct a tower fault prediction model, a radial basis kernel function is selected, a grid search method is used to optimize kernel function parameters and a penalty factor, and a cross-validation method is used to select optimal model parameters; the key characteristic vector is analyzed for importance using a CART decision tree algorithm, a maximum tree depth and a minimum leaf node sample number are set, an optimal split feature is selected through a Gini index, and a feature importance ranking is obtained; The historical operation data is segmented using a sliding time window method, a window size and a step size are set, a multi-layer LSTM network is used to construct a time series prediction model, an activation function is selected, a batch size and a learning rate initial value are set, an optimizer is used, and a future state of the tower is predicted to obtain time series prediction results of tower indicators.

7. The method of claim 1, wherein, According to the tower fault prediction result, a preventive maintenance plan is formulated, for a tower with a fault probability greater than a preset threshold, priority maintenance is arranged, and for a tower predicted to fail, power supply strategy is adjusted in advance to reduce load pressure, including: a preventive maintenance priority matrix is constructed according to the fault probability interval and the predicted failure time, the horizontal axis of the matrix represents the fault probability interval, and the vertical axis represents the predicted failure time; The tower maintenance priority is quantitatively evaluated using a fuzzy comprehensive evaluation method, the evaluation including establishing an evaluation index system, determining the weight of each index through an analytic hierarchy process, constructing a fuzzy relationship matrix, and performing fuzzy synthesis operation using a weighted average method; an optimal preventive maintenance plan is formulated using a 0-1 knapsack dynamic programming algorithm, the algorithm taking the maintenance priority score as the value and the maintenance time as the weight to solve the optimal maintenance combination that meets the manpower resource and time window constraints; according to the preventive maintenance plan, an adaptive load control algorithm is used to dynamically adjust the power supply strategy for towers with different priorities, the algorithm automatically calculates the load adjustment amount by real-time monitoring of the tower load condition, combining weather forecast data, and using a fuzzy PID controller; if the tower is predicted to fail within a preset time range, the tower is additionally increased by a predicted load reduction amount, and the monitoring frequency is increased.

Images