Icing multi-dimensional prediction method and system based on multi-source data

Abstract

The invention discloses an icing multi-dimensional prediction method and system based on multi-source data, and the method comprises the steps: obtaining historical icing data, carrying out the preprocessing, obtaining the fused multi-source data, carrying out the optimization of an Attention-LSTM model through an improved travel mouse optimization algorithm, obtaining a current optimal solution, carrying out the training of the Attention-LSTM model based on the obtained data and the optimal solution, and carrying out the prediction of the current optimal solution. Obtaining a power transmission line icing prediction model, training the power transmission line icing prediction model based on the historical icing data, and obtaining an evaluated power transmission line icing prediction model; and obtaining real-time icing data of the target area, inputting the real-time icing data into the obtained model for prediction, and outputting a prediction result to generate decision auxiliary information. According to the method, the Attention-LSTM model is optimized through the improved Lavant mouse optimization algorithm, the energy factor attenuation rate is adjusted, and the global optimal discovery rate is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power transmission lines, and in particular relates to a multi-dimensional icing prediction method and system based on multi-source data. Background Art

[0002] In recent years, climate change has become a global environmental issue. Extreme weather has occurred frequently, and icing incidents on transmission lines have become increasingly frequent. Icing on the lines seriously threatens the normal operation of transmission lines and also brings great challenges to line maintenance. Therefore, how to accurately predict the ice thickness of the lines has become an urgent problem to be solved.

[0003] Currently, traditional icing prediction methods are primarily based on icing data. Using statistical theory and intelligent computing techniques, they analyze meteorological, environmental, and icing data to establish reliable and accurate fitting models. With the advancement and application of icing monitoring technology and equipment, more and more microclimate and icing monitoring data are being used in the construction of icing prediction models.

[0004] While traditional methods can predict ice thickness to a certain extent, they suffer from several significant drawbacks. First, traditional ice prediction methods rely too heavily on historical ice data, making them ineffective for predicting climate change and extreme weather. Second, micrometeorological data sources are often limited, making it difficult to cover all lines, and data quality cannot be guaranteed, resulting in insufficient model accuracy. Finally, the icing process on transmission lines is complex, with numerous influencing factors, many of which involve nonlinear and complex multidimensional relationships. This places high demands on algorithms, making it difficult for traditional machine learning algorithms to implement. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present invention provides a multi-dimensional ice cover prediction method and system based on multi-source data, which aims to solve the problems in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-dimensional ice cover prediction method based on multi-source data, comprising: Step S1: Obtain historical ice cover data of the target area; Step S2: pre-processing the acquired historical ice cover data to obtain fused multi-source data; Step S3: By introducing the elite archiving mechanism and energy factor into the Lemmings optimization algorithm, an improved Lemmings optimization algorithm is obtained. The improved Lemmings optimization algorithm is used to optimize the Attention-LSTM model to obtain the current optimal solution. Step S4: The Attention-LSTM model is trained using the fused multi-source data and the current optimal solution to obtain a transmission line icing prediction model; the historical icing data is divided into a training set and a test set, and the transmission line icing prediction model is trained and tested to obtain an evaluated transmission line icing prediction model; Step S5: Acquire real-time icing data of the target area, input the real-time icing data into the evaluated transmission line icing prediction model for prediction, and output prediction results of line icing thickness, icing type, risk level, and ice melting time in the target area; Step S6: generating decision-making auxiliary information based on the prediction results of the ice thickness, ice type, risk level and ice melting time of the line.

[0007] Furthermore, the historical ice cover data of the target area include micrometeorological data, weather station data, and high-resolution meteorological grid data; Weather station data comes from a weather station near an ice-covered tower. Observation equipment at the weather station records hourly temperature, humidity, wind speed, and rainfall during the ice-covered period. Weather station data is constructed based on the hourly observations recorded during the ice-covered period. Preprocess the observation values ​​to obtain the corresponding high-resolution meteorological grid data; The micro-meteorological data are the hourly temperature and humidity data collected by the micro-meteorological collection device during the ice-covered period of the ice-covered tower.

[0008] Furthermore, the specific process of fusing multi-source data is as follows: Calculate the distances between the ice-covered tower and the micrometeorological collection device, meteorological station, and the corresponding grid center in the high-resolution meteorological grid data, which are recorded as d1, d2, and d3 respectively; the inverse of the square of the distances d1, d2, and d3 is used as the spatial distance weight coefficient. ,express: (1); Where d is the distance; i is an index variable; i=1, 2, 3; is the distance from the ice-covered tower to the micro-meteorological collection device; is the distance from the ice-covered tower to the meteorological station; is the distance from the ice-covered tower to the corresponding grid center in the high-resolution meteorological grid data; Dynamically monitor the reliability scores of micrometeorological data, weather station data, and high-resolution meteorological grid data using a sliding window-based source data anomaly detection algorithm , real-time output reliability weight coefficient ,express: (2); Among them, the source data anomaly detection algorithm based on sliding window is realized by dynamically monitoring multiple quality dimensions to obtain reliability scores ;Multiple quality dimensions include data completeness, data mutation frequency and variance stability; Data integrity rate: The sliding window source data anomaly detection algorithm is used to calculate the ratio of the number of micrometeorological data, weather station data, and high-resolution meteorological grid data points actually received within the sliding time window to the total number of expected micrometeorological data, weather station data, and high-resolution meteorological grid data points. When the ratio of the actual number of received points to the total number of expected points is closer to 1, it means that the integrity of micrometeorological data, weather station data and high-resolution meteorological grid data is higher; Frequency of data mutation: Using a sliding window source data anomaly detection algorithm, the absolute difference between two adjacent time points in the micrometeorological data, weather station data, and high-resolution meteorological grid data within the statistical window is calculated. This absolute difference is then compared with a preset threshold. The higher the absolute difference exceeds the preset threshold, the lower the reliability of the micrometeorological data, weather station data, and high-resolution meteorological grid data. Variance stability: The source data anomaly detection algorithm of the sliding window is used to calculate the variance of the micrometeorological data, weather station data, and high-resolution meteorological grid data within the sliding window, and then compare it with the preset long-term benchmark variance. The closer the ratio of the calculated variance to the long-term benchmark variance is to 1, the more stable the fluctuation of the micrometeorological data, weather station data, and high-resolution meteorological grid data is. The overall reliability score of micrometeorological data, weather station data and high-resolution weather grid data within the time window is calculated by comprehensive data completeness, data mutation frequency and variance stability using weighted geometric mean or weighted arithmetic mean. ; Comprehensive spatial distance weight coefficient and reliability weight coefficient Obtain the weights ω of micrometeorological data, weather station data, and high-resolution meteorological grid data: (3); Where, for Adjustable hyperparameters of for Adjustable hyperparameters of The weight ω calculates the weighted average of micrometeorological data, weather station data, and high-resolution meteorological grid data to obtain fused multi-source data.

[0009] Furthermore, the specific process of the improved lemming optimization algorithm is as follows: The lemming optimization algorithm consists of six stages: initialization stage, long-distance migration stage, burrowing stage, foraging stage, and avoiding natural enemies stage; The initialization phase of the Lemmings optimization algorithm generates a random population in the search space as the initial population; Use Latin hypercube sampling to initialize the initial position of the initial population, expressed as: (4); Where, is the initial position of the i-th lemming in the j-th dimension; is the lower bound of the j-th dimension; is the random number generated by Latin hypercube sampling for the i-th lemming in dimension j, ensuring uniform distribution of the initial population; is the upper bound of the j-th dimension; The long-distance migration stage means: (5); Where, is the position of the i-th lemming after long-distance migration at the t+1th iteration; is the position of the optimal lemming in the initial population at the tth iteration; A sign to change the search direction; is the random number vector of the probability density function of Brownian motion; is a random number uniformly generated in the interval [0,1], balancing the position of the optimal lemming in the initial population; is the current position of the i-th lemming at the t-th iteration; is the position of the a-th lemming randomly selected from the initial population at the t-th iteration; A flag that changes the search direction, indicating: (6); Where, is a random number uniformly generated in the interval [0,1]. By introducing random numbers, the probability of switching the search direction is determined; The probability density function of Brownian motion is a random number vector, which represents: (7); Where, is the natural exponential function; is a random number vector; is the exponential term of Gaussian distribution; The burrowing phase means: (8); Where, It is a dynamically adjusted parameter that controls the step size or exploration range of the lemming’s digging behavior; is the position of the bth lemming randomly selected from the initial population at the tth iteration; Dynamically adjusted parameters, indicating: (9); Where, is a sine function; is the current time step half of The foraging phase means: (10); Where, are the parameters of the spiral winding mechanism, simulating the spiral movement path of lemmings in the foraging area; Parameters of the helical winding mechanism, representing: (11); (12); Where, is the radius of the spiral; is the cosine function; is the dimension of the search space; is the coordinate component of the best lemming in the initial population in the jth dimension at the tth iteration, where yes Coordinate components in the j dimension; is the coordinate component of the i-th lemming in the j-th dimension at the t-th iteration; Avoiding natural enemies stage means: (13); (14); Where, It is a dynamically adjusted parameter that controls the intensity or step length of the lemmings' avoidance behavior; is the random number vector of Levy Flight; is the maximum number of iterations; Lévy flight random number vector, representing: (15); Where, is a random number that obeys the normal distribution and is located in the numerator; is the standard deviation of the Levy distribution; is a random number that obeys the normal distribution and is located in the denominator; is the Levy index; Γ is the gamma function; Introducing elite archive mechanism: Establish an elite archive to save the positions of the best lemmings found so far. After each iteration, calculate the fitness value of the lemmings in the initial population according to the predefined fitness function, and sort the lemmings in the initial population according to the fitness value. Then, select the positions of the top k best lemmings after sorting; compare the positions of the best lemmings with the positions of the best lemmings saved in the elite archive, and select the position of the best lemming with the highest similarity value as the elite lemming. ; Will Joining the long-distance migration stage means: (16); Where, is the social learning factor; It is a random number in the interval [0,1], which is used to adjust the elite lemmings The degree of influence on the current lemming position update; Will Adding the digging phase means: (17); Will Adding the foraging phase means: (18); Will Adding the avoidance of natural enemies stage means: (19); Assume the energy factor E. When the energy factor E changes, the lemmings will enter different stages according to the size of the energy factor E. Energy Factor The calculation of , means: (20); Where, is the energy factor at the tth iteration; is the minimum value of the energy factor; is the maximum value of the energy factor; is the initial population diversity index at the tth iteration; The initial population diversity index at the tth iteration is expressed as: (twenty one); Where, is the initial population size, i.e., the number of lemmings in the initial population; is the dimension of the search space; is the average position coordinate of the initial population in the jth dimension at the tth iteration; is the initial position of the i-th lemming on the j-th dimension at the t-th iteration; According to formula (20), the energy factor at the iteration number t is calculated , determine the energy factor at iteration number t Whether the set energy threshold is reached, when the set energy threshold is reached, it enters the exploration stage, which includes the long-distance migration stage and the burrowing stage. When the set energy threshold is not reached, it enters the foraging stage. At the same time, it uses Determine the formula for updating the lemmings' positions; During the exploration phase, is less than 0.3, the lemmings are considered to be in the long-distance migration stage, and the position of the lemmings is updated using formula (16); when If it is greater than 0.3, the lemming is considered to be in the burrowing stage, and the position of the lemming is updated using formula (17); when If it is less than 0.5, the lemmings are considered to be in the foraging stage, and the position of the lemmings is updated using formula (18); when If it is greater than 0.5, the lemmings are considered to be in the stage of avoiding natural enemies, and the position of the lemmings is updated using formula (19); After completing the lemming position update in the exploration phase and the foraging phase, the fitness value of the lemming's new position after the update of equations (16), (17), (18) and (19) is evaluated, and the fitness values ​​of the lemmings after the update of equations (16), (17), (18) and (19) are compared to find the optimal lemming as the current optimal solution; Increase the number of iterations t by 1, that is ; is the current iteration number; judge Whether the maximum number of iterations has been reached ,when The maximum number of iterations was not reached , return to formula (20) and formula (21) to recalculate the energy factor ;when Reached the maximum number of iterations The optimization process ends.

[0010] Furthermore, the specific process of obtaining the evaluated transmission line icing prediction model is as follows: The historical icing data is divided into a training set and a test set. The training set is used to train the transmission line icing prediction model, and the test set is used to test the accuracy of the trained transmission line icing prediction model. The coefficient of determination R2 and the minimum mean square error MSE were selected as performance indicators to evaluate the accuracy performance of the trained transmission line icing prediction model, and the evaluated transmission line icing prediction model was obtained.

[0011] Furthermore, the specific process of generating decision-making auxiliary information is as follows: Acquire real-time ice cover data of the target area, and pre-process the real-time ice cover data in step S2 to obtain real-time fused multi-source data; Input the real-time fused multi-source data of the target area into the evaluated transmission line icing prediction model to predict the icing situation of the lines in the target area and output the prediction results of the ice thickness, ice type, risk level and ice melting time of the lines in the target area; A decision rule base is constructed, and the prediction results of ice thickness, ice type, risk level and ice melting time of the output target area are input into the decision rule base to generate decision-making auxiliary information.

[0012] A multi-dimensional ice cover prediction system based on multi-source data is used to implement a multi-dimensional ice cover prediction method based on multi-source data, comprising: A data acquisition module is used to obtain historical ice cover data of the target area; The data processing module is used to pre-process the acquired historical ice cover data to obtain fused multi-source data; The model optimization module is used to improve the Lemmings optimization algorithm by introducing an elite archiving mechanism and an energy factor. This improves the Lemmings optimization algorithm and uses it to optimize the Attention-LSTM model to obtain the current optimal solution. The training module is used to train the Attention-LSTM model by fusing multi-source data and the current optimal solution to obtain a transmission line icing prediction model. The historical icing data is divided into a training set and a test set, and the transmission line icing prediction model is trained and tested to obtain an evaluated transmission line icing prediction model. The prediction module is used to obtain real-time icing data of the target area, input the real-time icing data into the evaluated transmission line icing prediction model for prediction, and output the prediction results of line icing thickness, icing type, risk level and ice melting time in the target area; The generation module is used to generate decision-making auxiliary information based on the prediction results of ice thickness, ice type, risk level and ice melting time of the line.

[0013] A computer device comprises: one or more processors; the processors are used to store one or more programs; when the one or more programs are executed by the one or more processors, the computer device is used to implement a multi-dimensional ice cover prediction method based on multi-source data.

[0014] A computer-readable storage medium includes a computer program stored thereon, wherein when the computer program is executed, the computer program is used to implement a multi-dimensional ice cover prediction method based on multi-source data.

[0015] Compared with the existing technology, the present invention has the following beneficial effects: (1) The present invention solves the problems of insufficient single data source and uneven data quality in traditional methods by integrating micrometeorological data, weather station data and high-resolution meteorological grid data and combining spatial distance weights with reliability weights, thus reducing the dependence on historical ice cover data and significantly improving the accuracy of ice cover thickness prediction.

[0016] (2) The improved lemming optimization algorithm proposed in this paper effectively solves the problem that the traditional lemming optimization algorithm is prone to falling into local optimality and slow convergence speed by introducing innovative means such as Latin hypercube sampling to initialize the population, elite archiving mechanism to save the position of the optimal lemming, and dynamic adjustment of the energy factor according to the diversity of the initial population. It enhances the global optimization ability, accelerates the convergence speed, provides better initial parameters for the Attention-LSTM model, and significantly improves the accuracy and stability of the evaluated transmission line icing prediction model.

[0017] (3) This invention breaks through the limitation of traditional ice cover prediction that is limited to thickness prediction, and realizes multi-dimensional prediction of ice cover thickness, ice cover type, risk level and ice melting time. It also generates decision-making auxiliary information by combining expert knowledge rule base, providing comprehensive ice cover situation awareness and accurate decision-making support for the operation and maintenance management of transmission lines, directly reducing line maintenance costs and risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Flow chart of the method of the present invention.

[0019] Figure 2 This is a comparison chart of the initial population diversity index of the present invention.

[0020] Figure 3 It is a comparison diagram of the fitness values ​​of the present invention. DETAILED DESCRIPTION

[0021] like Figure 1 As shown, the present invention provides a technical solution: a multi-dimensional ice cover prediction method based on multi-source data, comprising: Step S1: Obtain historical ice cover data of the target area; Step S2: pre-processing the acquired historical ice cover data to obtain fused multi-source data; Step S3: By introducing the elite archiving mechanism and energy factor into the Lemmings optimization algorithm, an improved Lemmings optimization algorithm is obtained. The improved Lemmings optimization algorithm is used to optimize the Attention-LSTM model to obtain the current optimal solution. Step S4: The Attention-LSTM model is trained using the fused multi-source data and the current optimal solution to obtain a transmission line icing prediction model; the historical icing data is divided into a training set and a test set, and the transmission line icing prediction model is trained and tested to obtain an evaluated transmission line icing prediction model; Step S5: Acquire real-time icing data of the target area, input the real-time icing data into the evaluated transmission line icing prediction model for prediction, and output prediction results of line icing thickness, icing type, risk level, and ice melting time in the target area; Step S6: generating decision-making auxiliary information based on the prediction results of the ice thickness, ice type, risk level and ice melting time of the line.

[0022] Among them, the historical ice cover data of the target area includes micrometeorological data, weather station data and high-resolution meteorological grid data; Weather station data comes from weather stations near ice-covered towers. Observation equipment at the weather stations (including thermometers, hygrometers, wind direction and anemometers, rain gauges, etc.) records hourly temperature, humidity, wind speed, and rainfall during the ice-covered period. Weather station data is generated based on the hourly observations during the ice-covered period. The observation values ​​are preprocessed to obtain high-resolution meteorological grid data of the 1km×1km grid points where the corresponding ice-covered towers are located; The micro-meteorological data are the hourly temperature and humidity data collected by the micro-meteorological collection device during the ice-covered period of the ice-covered tower.

[0023] The specific process of obtaining the fused multi-source data is as follows: Calculate the distances between the ice-covered tower and the micrometeorological collection device, meteorological station, and the corresponding grid center in the high-resolution meteorological grid data, which are recorded as d1, d2, and d3 respectively; the inverse of the square of the distances d1, d2, and d3 is used as the spatial distance weight coefficient. ,express: (1); Where d is the distance; i is an index variable; i=1, 2, 3; is the distance from the ice-covered tower to the micro-meteorological collection device; is the distance from the ice-covered tower to the meteorological station; is the distance from the ice-covered tower to the corresponding grid center in the high-resolution meteorological grid data; Dynamically monitor the reliability scores of micrometeorological data, weather station data, and high-resolution meteorological grid data using a sliding window-based source data anomaly detection algorithm , real-time output reliability weight coefficient ,express: (2); Among them, the source data anomaly detection algorithm based on sliding window is realized by dynamically monitoring multiple quality dimensions to obtain reliability scores ; Multiple quality dimensions include data completeness, data mutation frequency, and variance stability: Data integrity rate: The sliding window source data anomaly detection algorithm is used to calculate the ratio of the number of micrometeorological data, weather station data, and high-resolution meteorological grid data points actually received within the sliding time window to the total number of expected micrometeorological data, weather station data, and high-resolution meteorological grid data points. When the ratio of the actual number of received points to the total number of expected points is closer to 1, it means that the integrity of micrometeorological data, weather station data and high-resolution meteorological grid data is higher; Frequency of data mutation: Using a sliding window source data anomaly detection algorithm, the absolute difference between two adjacent time points in the micrometeorological data, weather station data, and high-resolution meteorological grid data within the statistical window is calculated. This absolute difference is then compared with a preset threshold. The higher the absolute difference exceeds the preset threshold, the lower the reliability of the micrometeorological data, weather station data, and high-resolution meteorological grid data. Variance stability: Using a sliding window source data anomaly detection algorithm, the variance of the micrometeorological data, weather station data, and high-resolution meteorological grid data within the sliding window is calculated and compared with a preset long-term benchmark variance (such as the average variance of the past 30 days). The closer the ratio of the calculated variance to the long-term benchmark variance is to 1, the more stable the fluctuations of the micrometeorological data, weather station data, and high-resolution meteorological grid data are. The overall reliability score of micrometeorological data, weather station data and high-resolution weather grid data within the time window is calculated by comprehensive data completeness, data mutation frequency and variance stability using weighted geometric mean or weighted arithmetic mean. ; Comprehensive spatial distance weight coefficient and reliability weight coefficient Obtain the weights ω of micrometeorological data, weather station data, and high-resolution meteorological grid data: (3); Where, for Adjustable hyperparameters of for Adjustable hyperparameters of The weighted average of micrometeorological data, weather station data and high-resolution meteorological grid data is calculated according to the obtained weight ω to obtain fused multi-source data.

[0024] Among them, the specific process of the improved lemming optimization algorithm is: The traditional lemming optimization algorithm mainly includes six stages: initialization stage, long-distance migration (exploration) stage, burrowing (exploration) stage, foraging (exploitation) stage, avoiding natural enemies (exploitation) stage, and transition from exploration to exploitation; The initialization phase of the traditional Lemmings optimization algorithm generates a random population as the initial population in the search space, thus starting the optimization process of the algorithm; Random initialization in the initialization phase may lead to uneven distribution of the initial population, affecting the convergence efficiency. The present invention uses Latin hypercube sampling to initialize the initial position of the initial population, which can significantly improve the performance of the lemming optimization algorithm, indicating that: (4); Where, is the initial position of the i-th lemming in the j-th dimension; is the lower bound of the j-th dimension; is the random number generated by Latin hypercube sampling for the i-th lemming in dimension j, ensuring uniform distribution of the initial population; is the upper bound of the j-th dimension; The long-distance migration phase refers to when lemmings are overpopulated and food is scarce, and lemmings will randomly migrate long distances. Lemmings will explore and search the space based on their current location and the location of random lemmings in the initial population, looking for habitats rich in food resources to obtain better living conditions and resources. It is also worth noting that the direction and distance of lemmings' migration are not fixed and are affected by various factors such as the ecological environment. The long-distance migration stage means: (5); Where, is the position of the i-th lemming after long-distance migration at the t+1th iteration; is the position of the optimal lemming in the initial population at the tth iteration; A sign to change the search direction; is the random number vector of the probability density function of Brownian motion; is a random number uniformly generated in the interval [0,1], balancing the position of the optimal lemming in the initial population; is the current position of the i-th lemming at the t-th iteration; is the position of the a-th lemming randomly selected from the initial population at the t-th iteration; A flag that changes the search direction, indicating: (6); Where, is a random number uniformly generated in the interval [0,1]. By introducing random numbers, the probability of switching the search direction is determined; The probability density function of Brownian motion is a random number vector, which represents: (7); Where, is the natural exponential function; is a random number vector; is the exponential term of Gaussian distribution; The burrowing phase involves lemmings digging holes in their habitat, forming complex tunnels that provide them with safe shelter and a place to store food. Lemmings will randomly dig new holes based on the current location of their burrow and the location of random lemmings in the initial population. This design helps lemmings quickly escape the threat of predators and find food more efficiently. The burrowing phase means: (8); Where, It is a dynamically adjusted parameter that controls the step size or exploration range of the lemming’s digging behavior; is the position of the bth lemming randomly selected from the initial population at the tth iteration; Dynamically adjusted parameters, indicating: (9); Where, is a sine function; is the current time step half of The foraging phase refers to the extensive and random movement of lemmings within their burrows, relying on their keen sense of smell and hearing to locate food sources. Lemmings typically establish a relatively small foraging area within their habitat, depending on the abundance and availability of food. To ingest as much food as possible, lemmings will wander randomly within the foraging area. To simulate the random foraging behavior of lemmings, a spiral winding mechanism was used to simulate the foraging phase, indicating: (10); Where, are the parameters of the spiral winding mechanism, which are used to simulate the spiral movement path of lemmings in the foraging area; Parameters of the helical winding mechanism, representing: (11); (12); Where, is the radius of the spiral; is the cosine function; is the dimension of the search space; is the coordinate component of the best lemming in the initial population in the jth dimension at the tth iteration, where yes Coordinate components in the j dimension; is the coordinate component of the i-th lemming in the j-th dimension at the t-th iteration; The predator avoidance stage refers to the avoidance and protective behavior of lemmings when they encounter danger. The burrows in their habitat are the lemmings' refuge. Once they discover an enemy, lemmings will use their special running ability to escape back to their burrows. At the same time, lemmings will also make deceptive movements to evade predators. The predator avoidance stage means: (13); (14); Where, It is a dynamically adjusted parameter used to control the intensity or step length of lemmings’ avoidance behavior; is the random number vector of Levy Flight; is the maximum number of iterations; Lévy flight random number vector, representing: (15); Where, is a random number that obeys the normal distribution and is located in the numerator; is the standard deviation of the Levy distribution; is a random number that obeys the normal distribution and is located in the denominator; is the Levy index; Γ is the gamma function; Introducing elite archive mechanism: Establish an elite archive to save the positions of the best lemmings found so far. After each iteration, calculate the fitness value of the lemmings in the initial population according to the predefined fitness function, and sort the lemmings in the initial population according to the fitness value. Then, select the positions of the top k best lemmings after sorting; compare the positions of the best lemmings with the positions of the best lemmings saved in the elite archive, and select the position of the best lemming with the highest similarity value as the elite lemming. ; Will Joining the long-distance migration stage means: (16); is the social learning factor; It is a random number in the interval [0,1], which is used to adjust the elite lemmings The degree of influence on the current lemming position update; Adding the digging phase means: (17); Will Adding the foraging phase means: (18); Will Adding the avoidance of natural enemies stage means: (19); In order to keep lemmings iterating between different stages, an energy factor E is designed. When the energy factor changes, lemmings will enter different stages according to the size of the energy factor E. When lemmings have enough energy factor E, they will selectively migrate or dig holes; otherwise, they will forage for food everywhere to avoid predators. The traditional lemming optimization algorithm uses a linearly decreasing energy factor E. This approach cannot adapt to nonlinear optimization requirements and does not consider the actual search state of the initial population, which can easily lead to premature convergence or over-exploration. To this end, the present invention proposes an adaptive energy factor E calculation method, which adjusts the energy factor E decay rate according to the real-time state dynamics of the initial population (initial population diversity). When the initial population diversity is high, the energy factor E decay is slowed down and the exploration time is extended. When the initial population converges too quickly, the energy factor E is increased and the exploration is restarted. When the initial population stagnates, the energy factor E decay is accelerated and the search returns to the foraging stage. The adaptive energy factor E calculation method proposed in the present invention can significantly improve the convergence speed and improve the global optimal discovery rate, which is expressed as: (20); Where, is the energy factor at the tth iteration; is the minimum value of the energy factor; is the maximum value of the energy factor; is the initial population diversity index at the tth iteration; The initial population diversity index at the tth iteration is expressed as: (twenty one); Where, is the initial population size, i.e., the number of lemmings in the initial population; is the dimension of the search space; is the average position coordinate of the initial population in the jth dimension at the tth iteration; is the initial position of the i-th lemming on the j-th dimension at the t-th iteration; According to formula (20), the energy factor at the iteration number t is calculated , determine the energy factor at iteration number t Whether the set energy threshold is reached, when the set energy threshold is reached, it enters the exploration stage, which includes the long-distance migration stage and the burrowing stage. When the set energy threshold is not reached, it enters the foraging stage. At the same time, it uses Determine the formula for updating the lemmings' positions; During the exploration phase, is less than 0.3, the lemmings are considered to be in the long-distance migration stage, and the position of the lemmings is updated using formula (16); when If it is greater than 0.3, the lemming is considered to be in the burrowing stage, and the position of the lemming is updated using formula (17); when If it is less than 0.5, the lemmings are considered to be in the foraging stage, and the position of the lemmings is updated using formula (18); when If it is greater than 0.5, the lemmings are considered to be in the stage of avoiding natural enemies, and the position of the lemmings is updated using formula (19); After completing the lemming position update in the exploration phase and the foraging phase, the fitness value of the lemming's new position after the update of equations (16), (17), (18) and (19) is evaluated, and the fitness values ​​of the lemmings after the update of equations (16), (17), (18) and (19) are compared to find the optimal lemming as the current optimal solution; Increase the number of iterations t by 1, that is ; is the current iteration number; judge Whether the maximum number of iterations has been reached ,when The maximum number of iterations was not reached , return to formula (20) and formula (21) to recalculate the energy factor ;when Reached the maximum number of iterations The optimization process ends.

[0025] The historical icing data is divided into a training set and a test set. The training set is used to train the transmission line icing prediction model, and the test set is used to test the accuracy of the trained transmission line icing prediction model. The coefficient of determination R2 and the minimum mean square error MSE were selected as performance indicators to evaluate the accuracy performance of the trained transmission line icing prediction model, and the evaluated transmission line icing prediction model was obtained.

[0026] Among them, based on the long short-term memory network (LSTM), the attention mechanism is introduced to construct the Attention-LSTM model; Micrometeorological data, weather station data, and high-resolution weather grid data are input into the Attention-LSTM model. These data are processed through a long short-term memory network to obtain a hidden state sequence. The hidden state sequence is then calculated through the attention calculation layer to obtain attention weights. The attention weights of the hidden states are weighted summed to generate a fixed context vector. This context vector is passed through a fully connected output layer to obtain the corresponding output result. Among them, the improved Lemmings optimization algorithm is used to optimize the learning rate, number of hidden layer units, attention weight matrix, and context vector weight of the Attention-LSTM model to obtain the current optimal solution. The specific process is as follows: Learning rate: The learning rate is a hyperparameter in the improved Lemmings optimization algorithm. It is used to control the step size of the Attention-LSTM model parameter update. If the learning rate is too large, the hyperparameter update amplitude will be too large, resulting in oscillation. If the learning rate is too small, the hyperparameter update amplitude will be too small, and the convergence speed will become slow. Number of hidden layer units: The number of hidden layer units is a hyperparameter of the Attention-LSTM model. It represents the characteristic dimension of the hidden state in each long short-term memory layer and determines the capacity of the Attention-LSTM model. Too few hidden layer units will lead to underfitting, while too many units will easily lead to overfitting and high computational cost. Attention weight matrix: The attention weight matrix is ​​a hyperparameter of the Attention-LSTM model. It is used to map hidden states to attention weights. Optimizing the attention weight matrix enables the Attention-LSTM model to more accurately assess the importance of micrometeorological data, weather station data, and high-resolution weather grid data at different historical moments. Context vector weight: The context vector weight is also a hyperparameter of the Attention-LSTM model. It is used to calculate the context vector. Optimizing the context vector weight enables the Attention-LSTM model to more accurately assess the importance of micrometeorological data, weather station data, and high-resolution weather grid data at different historical moments.

[0027] The improved Lemmings optimization algorithm is used to optimize hyperparameters such as the learning rate, number of hidden layer units, attention weight matrix, and context vector weight, so that the prediction performance parameters of the Attention-LSTM model become the best, which is the current optimal solution.

[0028] The specific process of generating decision-making auxiliary information is as follows: The staff retrieves the real-time ice cover data of the target area in the past or next 12 hours from different sources, and pre-processes the real-time ice cover data in step S2 to obtain real-time fused multi-source data; The real-time fusion multi-source data of the target area is input into the evaluated transmission line icing prediction model to predict the line icing situation in the target area. Based on the corresponding output results, further analysis is performed to output the prediction results of the line icing thickness, icing type, risk level and ice melting time in the target area; Line ice thickness: Output continuous line ice thickness values ​​in a regression manner. The evaluated transmission line ice prediction model calculates the prediction results based on real-time fusion of multi-source data. Icing Type: Outputs the probability distribution of major icing types such as rime, rime, and mixed rime in a classified manner; represents the probability assessment of different types of icing by the evaluated transmission line icing prediction model; Risk level: Output the preset risk level in a classified manner (such as Level I: Mild; Level II: Moderate; Level III: Severe; Level IV: Extreme); Ice melting time: Outputs the estimated ice melting time using regression method; the estimated ice melting time is based on the forecast of meteorological grid data (temperature, wind speed, solar radiation) for at least the next 24 hours; A decision rule library was built, which encapsulates industry knowledge and expert experience in the form of "if-then" rules. For example, if the risk level is Level III (severe), the icing type is rime, and the line ice thickness is high, the warning level is orange, and the decision-making support information recommends activating the DC ice melting device four hours before the predicted ice melting time. The prediction results of line ice thickness, ice type, risk level and ice melting time are input into the decision rule library, and decision-making auxiliary information is automatically matched and generated; the decision-making auxiliary information is visualized through the human-computer interaction interface or automatically sent to the operation and maintenance personnel's mobile terminal and the ice melting device control system through the communication interface.

[0029] A system for predicting ice thickness based on multi-source data, for implementing the multi-dimensional ice thickness prediction method based on multi-source data, comprising: A data acquisition module is used to obtain historical ice cover data of the target area; The data processing module is used to pre-process the acquired historical ice cover data to obtain fused multi-source data; The model optimization module is used to improve the Lemmings optimization algorithm by introducing an elite archiving mechanism and an energy factor. This improves the Lemmings optimization algorithm and uses it to optimize the Attention-LSTM model to obtain the current optimal solution. The training module is used to train the Attention-LSTM model by fusing multi-source data and the current optimal solution to obtain a transmission line icing prediction model. The historical icing data is divided into a training set and a test set, and the transmission line icing prediction model is trained and tested to obtain an evaluated transmission line icing prediction model. The prediction module is used to obtain real-time icing data of the target area, input the real-time icing data into the evaluated transmission line icing prediction model for prediction, and output the prediction results of line icing thickness, icing type, risk level and ice melting time in the target area; The generation module is used to generate decision-making auxiliary information based on the prediction results of ice thickness, ice type, risk level and ice melting time of the line.

[0030] A computer device comprises: one or more processors; the processors are used to store one or more programs; when the one or more programs are executed by the one or more processors, a multi-dimensional ice cover prediction method based on multi-source data is implemented.

[0031] A computer-readable storage medium stores a computer program, which, when executed, implements a multi-dimensional ice cover prediction method based on multi-source data.

[0032] Figure 2 The changing trends of the initial population diversity before and after the Lemmings algorithm optimization are shown. The original Lemmings algorithm uses a random method to initialize the population, which can easily lead to initial population aggregation and insufficient diversity, causing the Attention-LSTM model to fall into a local optimum. The improved Lemmings optimization algorithm uses Latin hypercube sampling in the initialization phase, which significantly improves the uniformity and diversity of the initial population. from Figure 2 It can be seen that in the early stage, the initial population diversity index of the original Lemmings algorithm and the improved Lemmings optimization algorithm are both at a high level. However, as the number of iterations increases, the initial population diversity index of the original Lemmings algorithm decreases rapidly, dropping to 0.3 after 10 iterations and almost to 0 after 50 iterations. The initial population diversity index of the improved lemming optimization algorithm has a more gradual downward trend, remaining above 0.6 after 10 iterations and remaining around 0.3 after 50 iterations, with an even more gradual downward trend. This indicates that the improved lemming optimization algorithm can better maintain the initial population diversity, enhance global search capabilities, and effectively avoid local optimal solutions.

[0033] Figure 3The comparison of fitness curves is shown. The original Lemmings algorithm does not introduce an elite archiving mechanism, which makes the iteration number of the original Lemmings algorithm converge faster. The number of iterations tends to stabilize around the 25th iteration, and the fitness value of the original Lemmings algorithm is large, indicating that the Attention-LSTM model is prone to falling into local optimal solutions. After the improved Lemmings optimization algorithm introduced the elite archiving mechanism, although the number of iterations converged slightly slower and reached stability at the 39th iteration, the fitness value of the improved Lemmings optimization algorithm was ultimately reduced to 0.46, reflecting a stronger global optimization capability. In summary, the improved Lemmings optimization algorithm not only maintains good convergence efficiency, but also significantly improves the search performance for the optimal solution.

[0034] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A multi-dimensional ice cover prediction method based on multi-source data, characterized in that: include: Step S1: Obtain historical ice cover data of the target area; Step S2: pre-processing the acquired historical ice cover data to obtain fused multi-source data; Step S3: By introducing the elite archiving mechanism and energy factor into the Lemmings optimization algorithm, an improved Lemmings optimization algorithm is obtained. The improved Lemmings optimization algorithm is used to optimize the Attention-LSTM model to obtain the current optimal solution. Step S4: The Attention-LSTM model is trained using the fused multi-source data and the current optimal solution to obtain a transmission line icing prediction model; the historical icing data is divided into a training set and a test set, and the transmission line icing prediction model is trained and tested to obtain an evaluated transmission line icing prediction model; Step S5: Acquire real-time icing data of the target area, input the real-time icing data into the evaluated transmission line icing prediction model for prediction, and output prediction results of line icing thickness, icing type, risk level, and ice melting time in the target area; Step S6: generating decision-making auxiliary information based on the prediction results of the ice thickness, ice type, risk level and ice melting time of the line.

2. The multi-dimensional ice cover prediction method based on multi-source data according to claim 1, characterized in that: The historical ice cover data of the target area includes micrometeorological data, weather station data and high-resolution meteorological grid data; Weather station data comes from a weather station near an ice-covered tower. Observation equipment at the weather station records hourly temperature, humidity, wind speed, and rainfall during the ice-covered period. Weather station data is constructed based on the hourly observations recorded during the ice-covered period. Preprocess the observation values ​​to obtain the corresponding high-resolution meteorological grid data; The micro-meteorological data are the hourly temperature and humidity data collected by the micro-meteorological collection device during the ice-covered period of the ice-covered tower.

3. The multi-dimensional ice cover prediction method based on multi-source data according to claim 2, characterized in that: The specific process of obtaining fused multi-source data is as follows: Calculate the distances between the ice-covered tower and the micrometeorological collection device, meteorological station, and the corresponding grid center in the high-resolution meteorological grid data, which are recorded as d1, d2, and d3 respectively; the inverse of the square of the distances d1, d2, and d3 is used as the spatial distance weight coefficient. ,express: （1）； Where d is the distance; i is an index variable; i=1, 2, 3; is the distance from the ice-covered tower to the micro-meteorological collection device; is the distance from the ice-covered tower to the meteorological station; is the distance from the ice-covered tower to the corresponding grid center in the high-resolution meteorological grid data; Dynamically monitor the reliability scores of micrometeorological data, weather station data, and high-resolution meteorological grid data using a sliding window-based source data anomaly detection algorithm , real-time output reliability weight coefficient ,express: （2）； Among them, the source data anomaly detection algorithm based on sliding window is realized by dynamically monitoring multiple quality dimensions to obtain reliability scores ;Multiple quality dimensions include data completeness, data mutation frequency and variance stability; Data integrity rate: The sliding window source data anomaly detection algorithm is used to calculate the ratio of the number of micrometeorological data, weather station data, and high-resolution meteorological grid data points actually received within the sliding time window to the total number of expected micrometeorological data, weather station data, and high-resolution meteorological grid data points. When the ratio of the actual number of received points to the total number of expected points is closer to 1, it means that the integrity of micrometeorological data, weather station data and high-resolution meteorological grid data is higher; Frequency of data mutation: Using a sliding window source data anomaly detection algorithm, the absolute difference between two adjacent time points in the micrometeorological data, weather station data, and high-resolution meteorological grid data within the statistical window is calculated. This absolute difference is then compared with a preset threshold. The higher the absolute difference exceeds the preset threshold, the lower the reliability of the micrometeorological data, weather station data, and high-resolution meteorological grid data. Variance stability: The source data anomaly detection algorithm of the sliding window is used to calculate the variance of the micrometeorological data, weather station data, and high-resolution meteorological grid data within the sliding window, and then compare it with the preset long-term benchmark variance. The closer the ratio of the calculated variance to the long-term benchmark variance is to 1, the more stable the fluctuation of the micrometeorological data, weather station data, and high-resolution meteorological grid data is. The overall reliability score of micrometeorological data, weather station data and high-resolution weather grid data within the time window is calculated by comprehensive data completeness, data mutation frequency and variance stability using weighted geometric mean or weighted arithmetic mean. ; Comprehensive spatial distance weight coefficient and reliability weight coefficient Obtain the weights ω of micrometeorological data, weather station data, and high-resolution meteorological grid data: （3）； Where, for Adjustable hyperparameters of for Adjustable hyperparameters of The weighted average of micrometeorological data, weather station data and high-resolution meteorological grid data is calculated according to the obtained weight ω to obtain fused multi-source data.

4. The multi-dimensional ice cover prediction method based on multi-source data according to claim 3 is characterized by: The specific process of the improved lemming optimization algorithm is as follows: The lemming optimization algorithm consists of six stages: initialization stage, long-distance migration stage, burrowing stage, foraging stage, and avoiding natural enemies stage; The initialization phase of the Lemmings optimization algorithm generates a random population in the search space as the initial population; Use Latin hypercube sampling to initialize the initial position of the initial population, expressed as: （4）； Where, is the initial position of the i-th lemming in the j-th dimension; is the lower bound of the j-th dimension; is the random number generated by Latin hypercube sampling for the i-th lemming in dimension j, ensuring uniform distribution of the initial population; is the upper bound of the j-th dimension; The long-distance migration stage means: （5）； Where, is the position of the i-th lemming after long-distance migration at the t+1th iteration; is the position of the optimal lemming in the initial population at the tth iteration; A sign to change the search direction; is the random number vector of the probability density function of Brownian motion; is a random number uniformly generated in the interval [0,1], balancing the position of the optimal lemming in the initial population; is the current position of the i-th lemming at the t-th iteration; is the position of the a-th lemming randomly selected from the initial population at the t-th iteration; A flag that changes the search direction, indicating: （6）； Where, is a random number uniformly generated in the interval [0,1]. By introducing random numbers, the probability of switching the search direction is determined; The probability density function of Brownian motion is a random number vector, which represents: （7）； Where, is the natural exponential function; is a random number vector; is the exponential term of Gaussian distribution; The burrowing phase means: （8）； Where, It is a dynamically adjusted parameter that controls the step size or exploration range of the lemming’s digging behavior; is the position of the bth lemming randomly selected from the initial population at the tth iteration; Dynamically adjusted parameters, indicating: （9）； Where, is a sine function; is the current time step half of The foraging phase means: （10）； Where, are the parameters of the spiral winding mechanism, simulating the spiral movement path of lemmings in the foraging area; Parameters of the helical winding mechanism, representing: （11）； （12）； Where, is the radius of the spiral; is the cosine function; is the dimension of the search space; is the coordinate component of the best lemming in the initial population in the jth dimension at the tth iteration, where yes Coordinate components in the j dimension; is the coordinate component of the i-th lemming in the j-th dimension at the t-th iteration; Avoiding natural enemies stage means: （13）； （14）； Where, It is a dynamically adjusted parameter that controls the intensity or step length of the lemmings' avoidance behavior; is the random number vector of Levy Flight; is the maximum number of iterations; Lévy flight random number vector, representing: （15）； Where, is a random number that obeys the normal distribution and is located in the numerator; is the standard deviation of the Levy distribution; is a random number that obeys the normal distribution and is located in the denominator; is the Levy index; Γ is the gamma function; Introducing elite archive mechanism: Establish an elite archive to save the positions of the best lemmings found so far. After each iteration, calculate the fitness value of the lemmings in the initial population according to the predefined fitness function, and sort the lemmings in the initial population according to the fitness value. Then, select the positions of the top k best lemmings after sorting; compare the positions of the best lemmings with the positions of the best lemmings saved in the elite archive, and select the position of the best lemming with the highest similarity value as the elite lemming. ; Will Joining the long-distance migration stage means: （16）； Where, is the social learning factor; It is a random number in the interval [0,1], which is used to adjust the elite lemmings The degree of influence on the current lemming position update; Will Adding the digging phase means: （17）； Will Adding the foraging phase means: （18）； Will Adding the avoidance of natural enemies stage means: （19）； Assume the energy factor E. When the energy factor E changes, the lemmings will enter different stages according to the size of the energy factor E. Energy Factor The calculation of , means: （20）； Where, is the energy factor at the tth iteration; is the minimum value of the energy factor; is the maximum value of the energy factor; is the initial population diversity index at the tth iteration; The initial population diversity index at the tth iteration is expressed as: （21）； Where, is the initial population size, i.e., the number of lemmings in the initial population; is the dimension of the search space; is the average position coordinate of the initial population in the jth dimension at the tth iteration; is the initial position of the i-th lemming on the j-th dimension at the t-th iteration; According to formula (20), the energy factor at the iteration number t is calculated , determine the energy factor at iteration number t Whether the set energy threshold is reached, when the set energy threshold is reached, it enters the exploration stage, which includes the long-distance migration stage and the burrowing stage. When the set energy threshold is not reached, it enters the foraging stage. At the same time, it uses Determine the formula for updating the lemmings' positions; During the exploration phase, is less than 0.3, the lemmings are considered to be in the long-distance migration stage, and the position of the lemmings is updated using formula (16); when If it is greater than 0.3, the lemming is considered to be in the burrowing stage, and the position of the lemming is updated using formula (17); when If it is less than 0.5, the lemmings are considered to be in the foraging stage, and the position of the lemmings is updated using formula (18); when If it is greater than 0.5, the lemmings are considered to be in the stage of avoiding natural enemies, and the position of the lemmings is updated using formula (19); After completing the lemming position update in the exploration phase and the foraging phase, the fitness value of the lemming's new position after the update of equations (16), (17), (18) and (19) is evaluated, and the fitness values ​​of the lemmings after the update of equations (16), (17), (18) and (19) are compared to find the optimal lemming as the current optimal solution; Increase the number of iterations t by 1, that is ; is the current iteration number; judge Whether the maximum number of iterations has been reached ,when The maximum number of iterations was not reached , return to formula (20) and formula (21) to recalculate the energy factor ;when Reached the maximum number of iterations The optimization process ends.

5. The multi-dimensional ice cover prediction method based on multi-source data according to claim 4, characterized in that: The specific process of obtaining the evaluated transmission line icing prediction model is as follows: The historical icing data is divided into a training set and a test set. The training set is used to train the transmission line icing prediction model, and the test set is used to test the accuracy of the trained transmission line icing prediction model. The coefficient of determination R2 and the minimum mean square error MSE were selected as performance indicators to evaluate the accuracy performance of the trained transmission line icing prediction model, and the evaluated transmission line icing prediction model was obtained.

6. The multi-dimensional ice cover prediction method based on multi-source data according to claim 5, characterized in that: The specific process of generating decision-making auxiliary information is as follows: Acquire real-time ice cover data of the target area, and pre-process the real-time ice cover data in step S2 to obtain real-time fused multi-source data; Input the real-time fused multi-source data of the target area into the evaluated transmission line icing prediction model to predict the icing situation of the lines in the target area and output the prediction results of the ice thickness, ice type, risk level and ice melting time of the lines in the target area; A decision rule base is constructed, and the prediction results of ice thickness, ice type, risk level and ice melting time of the output target area are input into the decision rule base to generate decision-making auxiliary information.

7. A multi-dimensional icing prediction system based on multi-source data, used to implement the multi-dimensional icing prediction method based on multi-source data according to any one of claims 1 to 6, characterized in that: include: A data acquisition module is used to obtain historical ice cover data of the target area; The data processing module is used to pre-process the acquired historical ice cover data to obtain fused multi-source data; The model optimization module is used to improve the Lemmings optimization algorithm by introducing an elite archiving mechanism and an energy factor. This improves the Lemmings optimization algorithm and uses it to optimize the Attention-LSTM model to obtain the current optimal solution. The training module is used to train the Attention-LSTM model by fusing multi-source data and the current optimal solution to obtain a transmission line icing prediction model. The historical icing data is divided into a training set and a test set, and the transmission line icing prediction model is trained and tested to obtain an evaluated transmission line icing prediction model. The prediction module is used to obtain real-time icing data of the target area, input the real-time icing data into the evaluated transmission line icing prediction model for prediction, and output the prediction results of line icing thickness, icing type, risk level and ice melting time in the target area; The generation module is used to generate decision-making auxiliary information based on the prediction results of ice thickness, ice type, risk level and ice melting time of the line.

8. A computer device, characterized in that: include: one or more processors; The processor is configured to store one or more programs; When the one or more programs are executed by the one or more processors, they are used to implement the multi-dimensional ice cover prediction method based on multi-source data as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that include: A computer program is stored thereon, and when the computer program is executed, it is used to implement the multi-dimensional ice cover prediction method based on multi-source data as described in any one of claims 1-6.

Images