Deep neural network-based distribution line icing pole falling and breaking probability prediction method and system

Abstract

The invention relates to the technical field of power system risk assessment, in particular to a deep neural network-based distribution line icing pole falling and breaking probability prediction method and a deep neural network-based distribution line icing pole falling and breaking probability prediction system. Acquiring historical meteorological data and pole and tower back-off information of the icing disaster area of the distribution line; constructing a deep neural network prediction model and optimizing the deep neural network prediction model; and training the deep neural network prediction model by adopting a simulated annealing whale algorithm to obtain a probability prediction result of rod falling and breaking under an icing condition. According to the deep neural network prediction model, an input layer is provided with four input neurons which are respectively corresponding to a wind direction, a wind speed, an icing thickness and a temperature; the hidden layers comprise 16 nodes in the first hidden layer, 8 nodes in the second hidden layer and 4 nodes in the third hidden layer; and the output layer adopts a Sigmoid activation function. By combining the whale optimization algorithm with the simulated annealing algorithm, effective combination of global search and local optimization is realized, and the training efficiency and prediction accuracy of the prediction model are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system risk assessment, and in particular to a method and system for predicting the probability of ice-covered and broken poles of distribution lines using a deep neural network. Background Art

[0002] In power distribution systems, icing disasters are a common weather phenomenon, especially in cold climates. Icing may cause the collapse of poles on distribution lines, which will have a serious impact on the stability and reliability of the power supply system. Therefore, accurately predicting the probability of pole collapse is of great significance for formulating effective prevention and maintenance strategies.

[0003] The prediction of pole collapse and breakage in power system icing disaster is very important to improve the power grid's ability to defend against icing disasters and ensure the safe and stable operation of the power grid. At present, most domestic and foreign scholars use the finite element model of the tower-line system to simulate and analyze the stress of transmission line towers. The stability of the tower structure is judged by the structural statics analysis and characteristic / non-characteristic buckling analysis of the transmission tower under a certain load. There are few studies on the structural stability of distribution towers under icing conditions. Therefore, it is particularly important to carry out risk assessment of pole collapse and breakage under icing conditions.

[0004] In the quantitative analysis of the failure risk of distribution lines, some scholars have conducted relevant research on the prediction of the power outage area, power outage duration, and the probability of pole tower collapse in the distribution system under typhoon disasters. However, there are few reports on the prediction of the probability of pole collapse under icing conditions. Therefore, it is very important to establish a new, professional, and reliable method for predicting the probability of pole collapse based on various indicators of distribution network towers under icing conditions. Summary of the invention

[0005] In view of the problems existing in the prior art, the inventor proposed the present invention.

[0006] Therefore, the problem to be solved by the present invention is how to solve the problem that there is little research on the structural stability of distribution poles under icing conditions, and provide a method for predicting the probability of distribution line poles collapsing and breaking based on deep neural networks. Combined with historical data, a corresponding database is established, and a deep neural network prediction model is constructed. Through simulated annealing whale algorithm training and optimization, the probability of distribution line poles collapsing and breaking under icing disasters can be accurately predicted.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a method for predicting the probability of ice-covered and broken poles of distribution lines using a deep neural network, which comprises:

[0009] As a preferred solution of the method for predicting the probability of ice-covered and broken poles of distribution lines using a deep neural network of the present invention, the method comprises: obtaining historical meteorological data and tower broken information of the distribution line ice-covered disaster area;

[0010] Constructing a deep neural network prediction model and optimizing the deep neural network prediction model;

[0011] The simulated annealing whale algorithm is used to train the deep neural network prediction model to obtain the prediction results of the probability of pole breaking under icing conditions.

[0012] As a preferred solution of the method for predicting the probability of ice-covered and broken poles in distribution lines using a deep neural network according to the present invention, the historical meteorological data includes wind speed data, wind direction data, temperature data and ice thickness data at standard height.

[0013] As a preferred solution of the method for predicting the probability of ice-covered and broken poles in distribution lines using the deep neural network of the present invention, the deep neural network prediction model includes: an input layer, which is provided with 4 input neurons, corresponding to wind direction, wind speed, ice thickness and temperature respectively; a hidden layer, which is set to 3 layers, with 16 nodes in the first hidden layer, 8 nodes in the second hidden layer, and 4 nodes in the third hidden layer; an output layer, which uses a Sigmoid activation function to output the probability of broken poles.

[0014] As a preferred solution of the method for predicting the probability of ice-covered and broken poles of distribution lines using the deep neural network of the present invention, the simulated annealing whale algorithm includes: initializing whale group parameters, including convergence factor, vector coefficient, probability determination coefficient and annealing speed; using the optimal position of individual whales as the initial temperature of the annealing algorithm; and using the temperature after annealing as the initial weight and threshold of the deep neural network model.

[0015] As a preferred solution of the method for predicting the probability of ice-covered and broken poles of distribution lines using the deep neural network of the present invention, the simulated annealing whale algorithm also includes: judging the behavior pattern of the whale according to a random number, and when the random number is less than the probability determination coefficient, executing a spiral attack to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is less than 1, executing encirclement of the prey to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is greater than or equal to 1, executing a random search to update the position.

[0016] As a preferred solution of the method for predicting the probability of ice-covered and broken poles of distribution lines using the deep neural network of the present invention, the spiral attack updates the position in the following manner: calculating the distance between the current whale position and the optimal position; updating the whale position through the spiral equation, wherein the spiral parameters are used to define the spiral shape.

[0017] As a preferred solution of the method for predicting the probability of ice-covered and broken poles of distribution lines using the deep neural network of the present invention, it also includes: calculating and comparing the fitness values ​​of whale individuals in the updated population and the original population; determining whether to accept the new solution according to the MetroPolis criterion; and adjusting the weights and biases of the deep neural network using the temperature value updated by the simulated annealing whale algorithm.

[0018] In a second aspect, an embodiment of the present invention provides a distribution line icing and pole-failure probability prediction system based on a deep neural network, which includes a data acquisition module for acquiring historical meteorological data and pole-failure information of the distribution line icing disaster area;

[0019] A construction module is used to construct a deep neural network prediction model and optimize the deep neural network prediction model;

[0020] The result prediction module uses the simulated annealing whale algorithm to train the deep neural network prediction model to obtain the prediction result of the probability of pole breaking under ice conditions.

[0021] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the method for predicting the probability of ice-covered distribution line poles being broken using a deep neural network as described in the first aspect of the present invention are implemented.

[0022] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the method for predicting the probability of ice-covered and broken poles in distribution lines using a deep neural network as described in the first aspect of the present invention are implemented.

[0023] The beneficial effects of the present invention are as follows: the present invention combines historical data, establishes a corresponding database, constructs a deep neural network prediction model, and trains and optimizes the model through a simulated annealing whale algorithm, which can accurately predict the probability of distribution line poles collapsing under ice disasters. It has the following advantages:

[0024] Combining the global and local search capabilities of the whale optimization algorithm and the simulated annealing algorithm, a better initial solution can be found in a larger range, and then fine-tuned locally to improve search efficiency and optimization effect;

[0025] The search speed of the simulated annealing algorithm is affected by the initial temperature and annealing speed. The whale algorithm is combined to quickly search for the initial temperature, which reduces the convergence time and computational cost.

[0026] Deep neural networks can capture the nonlinear characteristics of icing disasters and better adapt to actual conditions by searching for optimal weights and biases, thereby improving the accuracy and stability of prediction results and reducing training time and computing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 It is a flow chart of the method for predicting the probability of ice-covered and broken poles of distribution lines based on deep neural network;

[0029] Figure 2 A computer device diagram of a method for predicting the probability of ice-covered and broken poles of distribution lines based on a deep neural network;

[0030] Figure 3 Schematic diagram of the process of optimizing deep neural network using simulated annealing whale algorithm for the probability prediction method of ice-covered distribution line pole breakage based on deep neural network. DETAILED DESCRIPTION

[0031] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0032] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0033] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0034] Example 1

[0035] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a method for predicting the probability of ice-covered and broken poles of distribution lines using a deep neural network, comprising:

[0036] S100: Obtain historical meteorological data and tower failure information in the distribution line icing disaster area;

[0037] In this embodiment, the historical meteorological data includes wind speed data, wind direction data, temperature data and standard height ice thickness data. The wind speed data is in meters per second, with a measurement range of 0-60m / s and an accuracy of ±0.3m / s; the wind direction data is expressed in degrees (0-360 degrees) with a measurement accuracy of ±2 degrees; the temperature data is in degrees Celsius with a measurement range of -40°C to +40°C and an accuracy of ±0.1°C; the standard height ice thickness is in millimeters with a measurement range of 0-200mm and an accuracy of ±0.5mm.

[0038] In an optional embodiment, historical meteorological data is collected through multiple meteorological monitoring stations along the distribution lines. Each monitoring station is equipped with ultrasonic wind speed sensors, wind vane sensors, PT100 temperature sensors, microwave radar thickness gauges and other equipment. The monitoring stations are arranged at intervals of 5-10 kilometers to achieve comprehensive monitoring of ice disaster areas. The frequency of data collection can be dynamically adjusted according to weather conditions. Under normal weather conditions, data is collected once an hour, and under severe weather conditions, it is encrypted to collect data every 10 minutes.

[0039] In an optional embodiment, the tower failure information includes basic information such as tower number, failure time, failure reason, tower type, tower height, design load level, and meteorological conditions at the time of failure. This information is obtained through the power company's operation and maintenance record system to ensure the authenticity and integrity of the data.

[0040] It should be noted that the collection of historical meteorological data requires the establishment of multiple meteorological monitoring stations along the distribution lines. The layout of the monitoring stations should take into account the topographical characteristics to ensure the representativeness of the data. At the same time, in order to improve the reliability of the data, each monitoring station is equipped with data backup and anomaly detection functions, which will automatically issue an alarm when data anomalies are detected.

[0041] In an optional embodiment, the tower failure information comes from the power company's operation and maintenance record system, including key information such as the specific time of the tower failure, the number and location of the failed tower, the meteorological conditions at the time, and the cause of the failure. The system creates a detailed file for each failure event, recording the complete process and on-site conditions of the failure.

[0042] It should be noted that in order to ensure the accuracy and completeness of the data, the system adopts multiple data verification mechanisms. Meteorological data collection equipment is calibrated regularly to ensure measurement accuracy; monitoring sites are operated in a dual-machine hot backup mode to avoid data loss caused by single point failures; data transmission uses an encrypted channel to ensure data security. At the same time, a complete data quality control process has been established, including data consistency verification, outlier screening, data gap filling and other processing mechanisms.

[0043] S101: Historical meteorological data include: wind speed data, wind direction data, temperature data and ice thickness data at standard height.

[0044] In this embodiment, the historical meteorological data collection cycle is 5 years, including complete observation records under extreme weather and normal weather conditions. Meteorological data collection adopts a layered storage architecture, real-time data is stored in an in-memory database, and historical data is regularly archived to a distributed file system.

[0045] In an optional embodiment, the meteorological data collection strategy is dynamically adjusted according to seasonal characteristics: in winter (November to March of the following year), when ice is prone to occur, the sampling frequency is increased to once every 15 minutes; in other seasons, the regular sampling frequency is maintained at once an hour. For sudden severe weather, the system automatically switches to a high-frequency sampling mode of once every 5 minutes.

[0046] In an optional embodiment, the layout of meteorological data collection points follows the principle of "dense in key areas and uniform in general areas". In mountainous areas, river valleys and other areas prone to ice cover, the distance between monitoring points is shortened to 3-5 kilometers; in plain areas, the standard distance is 8-10 kilometers. Each monitoring point is equipped with a UPS uninterruptible power supply and a 4G / 5G dual-mode communication module to ensure the continuity of data collection.

[0047] It should be noted that the quality of meteorological data directly affects the performance of the prediction model. The system adopts a three-level data quality control mechanism: the first level is sensor-level hardware self-checking, including regular calibration and fault diagnosis; the second level is site-level data verification, identifying outliers through spatiotemporal consistency analysis; the third level is system-level data review, using the expert knowledge base to conduct comprehensive evaluation and correction of data.

[0048] S200: Build a deep neural network prediction model and optimize the deep neural network prediction model;

[0049] In this embodiment, the deep neural network prediction model includes an input layer, three hidden layers and an output layer. The input layer is equipped with 4 input neurons, corresponding to wind direction, wind speed, ice thickness and temperature respectively; the number of nodes in the hidden layer is 16, 8 and 4 respectively, forming a pyramid structure; the output layer uses a Sigmoid activation function to output a probability value of a broken pole between 0 and 1.

[0050] In an optional embodiment, the specific configuration of the network structure is as follows: the input layer uses a linear activation function, and the data is input into the network after being standardized; the 16 nodes in the first hidden layer can fully extract the primary representation of the input features; the 8 nodes in the second hidden layer further abstract and combine the features; the 4 nodes in the third hidden layer complete the extraction of high-level semantic features. The hidden layers all use the ReLU activation function, and the Dropout mechanism is introduced to prevent overfitting.

[0051] In an optional embodiment, the model optimization process includes the following aspects: first, the network weights are initialized with Xavier so that the variance of the input and output of each layer remains consistent; then, batch normalization technology is used to speed up the training convergence; finally, the Adam optimizer is used to update the parameters, the initial learning rate is set to 0.001, and the cosine annealing strategy is used to dynamically adjust the learning rate.

[0052] It should be noted that the design of the deep neural network model fully considers the characteristics of the ice-covered broken pole prediction task. The pyramid-shaped network structure can extract features layer by layer, and the number of layers and nodes is set to strike a balance between prediction accuracy and computational complexity. The generalization ability of the model is enhanced by introducing technologies such as regularization and Dropout. At the same time, the model supports online learning and can continuously learn from the newly added historical data to continuously improve the prediction performance.

[0053] S201: The deep neural network prediction model includes: an input layer, which is set with 4 input neurons, corresponding to wind direction, wind speed, ice thickness and temperature respectively; a hidden layer, which is set to 3 layers, with 16 nodes in the first hidden layer, 8 nodes in the second hidden layer, and 4 nodes in the third hidden layer; an output layer, which uses a Sigmoid activation function to output the probability of pole breakage.

[0054] In this embodiment, different activation functions are used in each layer of the network to improve the model's expressiveness. The input layer uses a linear activation function to preserve the original feature information; the three hidden layers use a Leaky ReLU activation function to avoid the problem of neuron death; the Sigmoid function of the output layer maps the prediction results to the [0,1] interval for easy probability interpretation.

[0055] In an optional embodiment, in order to improve the generalization ability of the model, a Dropout layer is added after each hidden layer, and the random inactivation probabilities are 0.3, 0.2 and 0.1 respectively. At the same time, an L2 regularization constraint is introduced, and the regularization coefficient is set to 0.001 to prevent the model from overfitting. The initial weights of each layer of neurons are initialized using the He method, so that the network can maintain an appropriate gradient scale at the beginning of training.

[0056] In an optional embodiment, the network training adopts the mini-batch gradient descent method, and the batch size is set to 64. The learning rate adopts the cosine annealing strategy, with an initial value of 0.001, a minimum value of 0.00001, and a warm-up period of 100 rounds. The early stopping strategy is used during the training process, and the training is stopped when the loss function on the validation set has not improved for 10 consecutive rounds.

[0057] It should be noted that the structural design of the deep neural network fully considers the characteristics of the ice-covered broken pole prediction task. The pyramid structure with three hidden layers can extract features layer by layer, and the decreasing number of nodes is helpful for information compression and refinement. Specifically, the 16 nodes in the first hidden layer are used to extract local features, the 8 nodes in the second hidden layer are used for feature combination, and the 4 nodes in the third hidden layer complete the extraction of high-level semantic features. This structure achieves a good balance between computational efficiency and model expression ability.

[0058] S300: The simulated annealing whale algorithm is used to train the deep neural network prediction model to obtain the prediction results of the probability of pole breaking under icing conditions.

[0059] In this embodiment, the simulated annealing whale algorithm first initializes the whale group parameters, including the convergence factor, vector coefficient, probability determination coefficient and annealing speed. Then the optimal position of the individual whale is used as the initial temperature of the annealing algorithm, and the temperature after annealing is used as the initial weight and threshold of the deep neural network model.

[0060] In an optional embodiment, the algorithm parameters are configured as follows: the whale group size is set to 30, the maximum number of iterations is 1000, the initial value of the convergence factor is 2 and decreases linearly with the number of iterations, the vector coefficient range is [-1, 1], the probability determination coefficient is 0.5, and the annealing rate is 0.98. These parameters have been verified by a large number of experiments and can achieve a good balance between search efficiency and optimization effect.

[0061] In an optional embodiment, the whale's behavior modes include spiral attack, encircling prey, and random search. When the random number is less than the probability determination coefficient, the spiral attack is executed to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is less than 1, the encircling prey is executed to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is greater than or equal to 1, a random search is executed to update the position. Different behavior modes have their own characteristics. Spiral attack helps local fine search, encircling prey can quickly approach the optimal solution, and random search increases the global exploration capability.

[0062] In the embodiment of the present application, when the spiral attack updates the position, the distance between the current whale position and the optimal position is first calculated, and then the whale position is updated by the spiral equation. The spiral parameter is used to define the spiral shape, and usually takes a value between [0,2π]. Smaller values ​​produce tighter spirals, which are conducive to accurate search, and larger values ​​produce looser spirals, which are conducive to jumping out of the local optimum.

[0063] It should be noted that the simulated annealing whale algorithm combines the swarm intelligence characteristics of the whale optimization algorithm and the ability of the simulated annealing algorithm to escape the local optimum. In each iteration, the algorithm calculates and compares the fitness values ​​of the updated population and the original population whales, and decides whether to accept the new solution based on the Metropolis criterion. This mechanism not only ensures the continuity of the search, but also gives the algorithm a certain probability of accepting suboptimal solutions, thereby enhancing the global search ability of the algorithm. At the same time, the introduction of the temperature parameter provides a good initial weight for the deep neural network and accelerates the convergence process of network training.

[0064] S301: simulated annealing whale algorithm, including: initializing whale group parameters, including convergence factor, vector coefficient, probability determination coefficient and annealing speed; using the optimal position of individual whales as the initial temperature of the annealing algorithm; using the temperature after annealing as the initial weight and threshold of the deep neural network model.

[0065] In this embodiment, the algorithm parameter configuration is optimized through a large number of experiments: the size of the whale group is set to 30, the maximum number of iterations is 1000; the initial value of the convergence factor α is 2, and it decreases linearly to 0 with the number of iterations; the vector coefficient a is uniformly distributed in the range of [-1,1]; the probability determination coefficient p is set to 0.5; the annealing rate λ is taken as 0.98. The initial temperature T0 is set to 100, and the termination temperature Tf is set to 0.01.

[0066] In an optional embodiment, the algorithm adopts an adaptive parameter adjustment strategy. When no better solution is produced after 5 consecutive iterations, the local search is strengthened by increasing the annealing rate λ and the probability determination coefficient p; when the algorithm falls into the local optimum for more than 10 iterations, the convergence factor α and the range of the vector coefficient a are temporarily increased to enhance the global search capability.

[0067] In an optional embodiment, in order to improve the parallel efficiency of the algorithm, a population grouping strategy is adopted. The 30 individuals are divided into 3 groups, each group of 10 individuals is searched in parallel on different GPUs, and the groups exchange information every 10 iterations to share the current optimal solution. This strategy not only maintains the diversity of the population, but also makes full use of hardware resources.

[0068] It should be noted that the parameter settings in the initialization phase have an important impact on the performance of the algorithm. The convergence factor controls the speed of contraction of the search range, the vector coefficient affects the change of the search direction, the probability determination coefficient determines the probability of selecting the behavior mode, and the annealing speed affects the speed of temperature drop. The reasonable configuration of these parameters enables the algorithm to search extensively in the global range and conduct detailed exploration when discovering the potential optimal solution.

[0069] S302: The simulated annealing whale algorithm further includes: judging the behavior pattern of the whale according to a random number. When the random number is less than the probability determination coefficient, perform spiral attack to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is less than 1, perform surrounding the prey to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is greater than or equal to 1, perform random search to update the position.

[0070] In this embodiment, the algorithm generates a random number r in the interval [0, 1] in each iteration, and compares it with the probability determination coefficient p (= 0.5) to determine the behavior pattern. This random selection mechanism ensures that different behavior patterns have the opportunity to be executed, increasing the diversity of the search. When r < p, the spiral attack mode is selected; when r ≥ p, the surrounding the prey or random search mode is selected according to the magnitude of the vector coefficient |a|.

[0071] In an optional embodiment, the probability determination coefficient p adopts a dynamic adjustment strategy. In the initial stage of iteration, the value of p is small (about 0.3), tending to select the surrounding the prey and random search modes, which is beneficial to global exploration; as the iteration progresses, the value of p gradually increases (up to 0.7 at most), increasing the execution probability of the spiral attack mode and strengthening the local search ability.

[0072] In an optional embodiment, the change of the vector coefficient a adopts a non-linear attenuation strategy, with an initial value of 2 and updated according to the following formula:

[0073]

[0074] where t is the current iteration number and T is the maximum iteration number. This attenuation method enables the search to maintain a large range in the early stage and gradually concentrate on the local area in the later stage.

[0075] It should be noted that the three behavior patterns each have their own characteristics and play different roles in the optimization process. The spiral attack mode approaches the target through a spiral path and is suitable for precise search; the surrounding the prey mode directly approaches the current optimal solution, which is beneficial to rapid convergence; the random search mode explores a new solution space by randomly selecting reference points to prevent falling into local optima. The organic combination of these three modes enables the algorithm to not only maintain strong global search ability but also have good local convergence characteristics.

[0076] S303: The spiral attack to update the position is carried out in the following way: calculate the distance between the current whale position and the best position; update the whale position through the spiral equation, where the spiral parameter is used to define the spiral shape.

[0077] In this embodiment, the spiral attack process first calculates the Euclidean distance D between the current whale position X(t) and the current best position X*. Then the next position X(t+1) is calculated by the spiral equation, which involves the spiral parameter b(=1) and the angle parameter l(∈[-1,1]). The spiral parameter b determines the density of the spiral, and the angle parameter l affects the rotation direction and amplitude of the spiral.

[0078] In an optional embodiment, the spiral parameter b adopts an adaptive adjustment mechanism. When a better solution is not found for multiple consecutive times, the b value is appropriately increased (up to 1.5) to make the spiral looser and expand the search range; when a better solution is found, the b value is reduced (down to 0.5) to make the spiral tighter, which is conducive to accurate search.

[0079] In an optional embodiment, the angle parameter l is generated using an improved random strategy. Different from simple uniform randomness, a random number generator based on Cauchy distribution is used. This distribution has a longer tail and can produce more large jumps, which helps to jump out of the local optimum.

[0080] It should be noted that the spiral attack is the most distinctive search method in the algorithm. The design of the spiral path not only ensures the continuity of the search, but also provides adaptive search capabilities through dynamic adjustment of parameters. By setting the spiral parameters reasonably, a balance can be achieved between the search range and accuracy, so that the potential optimal solution will not be missed, and the search will not linger in the local area for too long.

[0081] S304: Also includes: calculating and comparing the fitness values ​​of individual whales in the updated population and the original population; determining whether to accept the new solution according to the MetroPolis criterion; and adjusting the weights and biases of the deep neural network using the temperature value updated by the simulated annealing whale algorithm.

[0082] In this embodiment, for each individual whale, the fitness value corresponding to its new position and original position is calculated. The fitness function uses the prediction error of the deep neural network on the validation set, including the weighted sum of the mean square error (MSE) and the cross entropy loss. According to the Metropolis criterion, when the new solution is better than the original solution, it is directly accepted; when the new solution is worse than the original solution, the new solution is accepted with a certain probability, and this probability is related to the temperature T and the fitness difference ΔE.

[0083] In an optional embodiment, the temperature update adopts a segmented annealing strategy. In the early stage of the search (the first 30% of iterations), a slower annealing speed (λ=0.98) is used; in the middle stage (30% to 70% of iterations), the annealing process is accelerated (λ=0.95); in the later stage (the remaining iterations), the annealing speed is slowed down again (λ=0.99) to achieve a refined search.

[0084] In an optional embodiment, to prevent precision loss, the algorithm maintains an elite solution archive, recording the best K solutions (K=5) found during the search process. When the temperature drops to a certain threshold, a solution is randomly selected from the elite solution archive for local search. This strategy ensures the diversity of solutions and improves the stability of the algorithm.

[0085] It should be noted that the MetroPolis criterion is the key mechanism for the algorithm to jump out of the local optimum. By giving the algorithm the ability to accept suboptimal solutions, the randomness and jumpiness of the search are enhanced. At the same time, the introduction of the temperature parameter T makes this acceptance mechanism adaptive: in the high temperature stage, the algorithm is more inclined to accept suboptimal solutions, which is conducive to global search; as the temperature decreases, the algorithm becomes more "picky" and gradually focuses on local fine search. This feature is highly consistent with the training characteristics of deep neural networks and shows good results in network weight optimization.

[0086] Furthermore, this embodiment also provides a distribution line ice-covered pole-breaking probability prediction system based on a deep neural network, comprising:

[0087] Data acquisition module, which obtains historical meteorological data and tower failure information in the distribution line icing disaster area;

[0088] A construction module is used to construct a deep neural network prediction model and optimize the deep neural network prediction model;

[0089] The result prediction module uses the simulated annealing whale algorithm to train the deep neural network prediction model to obtain the prediction result of the probability of pole breaking under ice conditions.

[0090] In summary, by adopting a pyramidal design with a multi-layer deep neural network structure (4 nodes in the input layer, 16 / 8 / 4 nodes in the three hidden layers, and an output layer), the layer-by-layer extraction and information compression of features are achieved, effectively capturing the nonlinear characteristics of the distribution line under ice-covered conditions. This structural design overcomes the problem of insufficient feature extraction of ice-covered lines by traditional models and improves prediction accuracy.

[0091] By introducing the simulated annealing whale hybrid optimization algorithm, the whale group intelligent search is combined with the simulated annealing local optimization:

[0092] The three behavioral modes of the whale algorithm, spiral attack, encirclement of prey, and random search, provide global search capabilities; the temperature control mechanism of the simulated annealing algorithm provides local fine search capabilities; the synergy of the two algorithms avoids the problem of a single algorithm easily falling into local optimality, greatly improving the efficiency of model training.

[0093] By designing an adaptive parameter adjustment mechanism:

[0094] Dynamically adjust the probability determination coefficient, focus on global search in the early stage of training, and strengthen local search in the later stage; adopt a segmented annealing strategy, using different annealing speeds in different stages; maintain elite solution archives and save the optimal solution for local search; enable the model to better adapt to different icing conditions.

[0095] By using the optimal position of individual whales as the initial annealing temperature and then the temperature after annealing as the initial weight of the neural network, the problem of difficulty in initializing the weights of deep neural networks was solved and the network convergence speed was accelerated.

[0096] By designing a solution acceptance mechanism based on the Metropolis criterion, the algorithm is given the ability to accept suboptimal solutions with a certain probability while maintaining its convergence, thereby improving the algorithm's ability to escape from local optimality and enhancing the generalization performance of the model.

[0097] Compared with the traditional tower-line system finite element model mentioned in the background technology, the deep learning method adopted by the present invention avoids the complex mechanical modeling process, automatically learns the characteristics of ice cover state in a data-driven way, and has stronger adaptability and generalization ability. At the same time, the whale algorithm is innovatively combined with the simulated annealing algorithm to overcome the limitations of a single optimization algorithm and achieve unexpected optimization effects.

[0098] Example 2

[0099] Reference Figure 1-Figure 3 , which is the second embodiment of the present invention, and provides a method for predicting the probability of ice-covered and broken poles in distribution lines based on a deep neural network.

[0100] Step 1: Collect historical meteorological conditions including wind speed, wind direction, temperature, ice thickness at standard height, and tower failure information.

[0101] Step 2: Preprocess the above information including data cleaning, default value processing, data standardization and sample balancing to obtain the data set.

[0102] Step 2.1, preprocessing includes data cleaning, default value processing, data standardization, and sample balancing, which is implemented as follows:

[0103] Step 2.2, data cleaning: the original data may contain obvious outliers, which will affect the reliability of the model. The processing method is to delete them directly;

[0104] Step 2.3, default value processing, for samples with partially missing data, interpolation filling method is used to process continuous numerical variables, and mode filling method is used to process categorical variables;

[0105] Step 2.4, data standardization, convert the wind speed and ice thickness at standard height according to the tower altitude; standardize the original data with dimensions;

[0106] Step 2.5, sample balancing, uses a combination of SMOTE and Tomek Links algorithm to perform sample balancing.

[0107] Preprocess the case set in step 1 and delete outliers. For samples with missing data, interpolation filling is used to process continuous numerical variables, and mode filling is used to process categorical variables. The Min-Max normalization method is used to process numerical features. The process is as follows: n To transform,

[0108]

[0109] Then the new sequence y1,y2,...,y n ∈[0,1] and dimensionless; for the problem of sample imbalance, the SMOTE and Tomek Links combined algorithm is used to process it, and then the data set for model training is obtained.

[0110] Step 3: construct a deep neural network (DNNs) prediction model and optimize it based on the simulated annealing whale algorithm to obtain a prediction model for the probability of pole breakage under icing conditions.

[0111] Based on the deep learning algorithm, a deep neural network (DNNs) model is constructed to predict the probability of distribution line pole collapse under ice disasters; the deep neural network DNNs consists of multiple layers, including input layer, hidden layer and output layer. Each layer consists of several neurons, which are connected by weights and biases;

[0112] Input layer, the input dimension is set to 4, including wind direction, converted wind speed, ice thickness and temperature;

[0113] The hidden layer is used to perform nonlinear transformations and uses the NoisyReLU activation function, which is a variant activation function of the rectified linear unit that is improved based on Gaussian noise. For the input value x of the neuron, the noisy linear rectification adds a certain degree of uncertainty in the normal distribution, that is,

[0114] f(x)=max(0,x+Y)

[0115] The random variable

[0116] The hidden layer is set to 3 layers; the first hidden layer has 16 nodes; the second hidden layer has 8 nodes; the third hidden layer has 4 nodes; for the ice-covered broken pole data set, the design of three hidden layers with a gradually decreasing number of neurons in each layer can provide good performance, and the design of gradually reducing neurons can capture the complex patterns of the data;

[0117] The output layer is used to generate prediction results, using the Sigmoid activation function, and the output is probability;

[0118]

[0119] The Sigmoid function is used for binary classification problems and compresses the input value to the range of 0 to 1 to indicate the probability of belonging to a certain category.

[0120] Step 3.1, construct a deep neural network DNNs distribution line pole collapse prediction model under ice disaster. The deep neural network DNNs consists of multiple layers, including input layer, hidden layer and output layer. Each layer consists of several neurons, which are connected by weights and biases;

[0121] Input layer, the input neurons are set to 4, including wind direction, converted wind speed, ice thickness and temperature;

[0122] The hidden layer is used to perform nonlinear transformations and uses the NoisyReLU activation function, which is a variant activation function of the rectified linear unit that is improved based on Gaussian noise. For the input value x of the neuron, the noisy linear rectification adds a certain degree of uncertainty in the normal distribution, that is,

[0123] f(x)=max(0,x+Y)

[0124] The random variable

[0125] The hidden layer is set to 3 layers; the first hidden layer has 16 nodes; the second hidden layer has 8 nodes; the third hidden layer has 4 nodes; this design can capture complex patterns of data;

[0126] The output layer is used to generate prediction results, using the Sigmoid activation function, and the output is probability;

[0127]

[0128] The Sigmoid function is used for binary classification problems and compresses the input value to the range of 0 to 1 to indicate the probability of belonging to a certain category.

[0129] Step 3.2, simulated annealing whale optimization deep neural network is initialized for the distribution line pole breaking probability prediction scenario under ice disaster, including initializing whale group, maximum number of iterations G, convergence coefficient a, current iteration coefficient t, individual update mode values ​​P and A, and annealing speed δ;

[0130] Preferably, the updated individual whale positions are used as the initial temperature of the annealing algorithm;

[0131] Preferably, the temperature T after annealing is used as the initial weight and threshold of the deep neural network model, and the training error value of the deep neural network model is traversed as the optimized fitness function; the weights and thresholds between the layers of the deep neural network are updated according to the annealing whale algorithm, and the formula is as follows:

[0132] T new =δT old

[0133]

[0134]

[0135] Among them, T new is the new temperature, T old is the current temperature, δ is the annealing rate, usually 0<δ<1, α is the learning rate, and They are the gradients of the loss function of the deep neural network with respect to the weights and biases of layer l;

[0136] Use binary cross entropy as the loss function of the deep neural network and use it as the fitness function,

[0137]

[0138] Where N is the number of samples, y i is the true label of the i-th sample, is the probability that the i-th sample predicted by the model is a positive class;

[0139] Step 3.3, simulated annealing whale algorithm, initialize the whale group, including convergence factor a, vector coefficient A, probability determination coefficient P and annealing speed δ;

[0140] Calculate the value (fitness) of the whale's position based on the neural network calculation error and find the optimal position;

[0141] A random number P between 0 and 1 is randomly generated to determine whether the whale launches a spiral attack or surrounds the prey. If P>0.5, execute step 3.3.1, otherwise, execute step 3.3.2;

[0142] Step 3.3.1, the whale launches a spiral attack to update the position, and the update formula is:

[0143] X(t+1)=D′·e bl ·cos(2πl)+X * (t)

[0144] Where D′ represents the distance between the current whale position and the optimal position, b is a constant used to define the spiral shape, l is a random number ranging from [-1,1], X(t+1) is the next position of the whale, and X * (t) is the whale’s current optimal position;

[0145] Step 3.3.2, if |A| < 1, the whale surrounds the prey and updates the position, otherwise, execute step 3.3.3, the whale surrounds the prey and updates the formula:

[0146] X(t+1)=X * (t)-A·D

[0147] D=|C·X * (t)-X(t)|

[0148] Where D represents the distance between the prey and the current whale position, parameters A and C are two vectors, A is between [-a, a], C is between [0, 2], and a is a parameter that decreases with the number of iterations and is between [0, 2];

[0149]

[0150] Step 3.3.3, random search update, the whale randomly selects a prey and updates the formula:

[0151] X(t+1)=X rand (t)-A·D rand

[0152] Among them, D rand represents the distance of the randomly selected prey from the current whale position;

[0153] Calculate and compare the fitness values ​​of individual whales in the updated population and the original population to obtain the optimal individual whale position, and use it as the initial temperature T of the simulated annealing algorithm;

[0154] Randomly generate a new solution in the neighborhood of the current solution T;

[0155] According to the Metropolis criterion, decide whether to accept the new solution. If the fitness of the new solution is better, accept the new solution. Otherwise, accept the new solution with probability P:

[0156]

[0157] in, is the fitness value difference between the new solution and the current solution;

[0158] Annealing, renewal temperature,

[0159] T new =δT old

[0160] Iterative optimization, repeatedly calculating the objective function value, generating a new solution, calculating the objective function value of the new solution, accepting the new solution and updating until the termination condition is met;

[0161] The temperature values ​​updated using the simulated annealing whale algorithm are used to adjust the weights and biases of a deep neural network.

[0162] Step 4: Use the trained fault probability prediction model to predict the probability of pole failure under icing conditions.

[0163] Example 3

[0164] This embodiment also provides a computer device, which is applicable to a method for predicting the probability of ice-covered and broken poles in distribution lines using a deep neural network, and includes a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a forced oscillation detection and positioning method for distribution networks as proposed in the above embodiment.

[0165] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, a forced oscillation detection and positioning method for a distribution network is implemented as proposed in the above embodiment.

[0166] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0167] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0168] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0169] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0170] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0171] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the probability of ice-covered and broken poles in distribution lines based on a deep neural network, characterized in that: This includes obtaining historical meteorological data and tower failure information in distribution line icing disaster areas; Constructing a deep neural network prediction model and optimizing the deep neural network prediction model; The simulated annealing whale algorithm is used to train the deep neural network prediction model to obtain the prediction results of the probability of pole breaking under icing conditions.

2. The method for predicting the probability of ice-covered distribution lines and broken poles based on a deep neural network according to claim 1, characterized in that: The historical meteorological data include: wind speed data, wind direction data, temperature data and ice thickness data at standard height.

3. The method for predicting the probability of ice-covered distribution lines and broken poles based on a deep neural network as claimed in claim 2, characterized in that: The deep neural network prediction model includes: an input layer, which is set with 4 input neurons, corresponding to wind direction, wind speed, ice thickness and temperature respectively; a hidden layer, which is set to 3 layers, with 16 nodes in the first hidden layer, 8 nodes in the second hidden layer, and 4 nodes in the third hidden layer; an output layer, which uses a Sigmoid activation function to output the probability of pole breaking.

4. The method for predicting the probability of ice-covered distribution lines and broken poles using a deep neural network as claimed in claim 3 is characterized by: The simulated annealing whale algorithm includes: initializing whale group parameters, including convergence factor, vector coefficient, probability determination coefficient and annealing speed; using the optimal position of individual whales as the initial temperature of the annealing algorithm; and using the temperature after annealing as the initial weight and threshold of the deep neural network model.

5. The method for predicting the probability of ice-covered distribution lines and broken poles based on a deep neural network as claimed in claim 4, characterized in that: The simulated annealing whale algorithm also includes: judging the behavior pattern of the whale according to the random number, and when the random number is less than the probability determination coefficient, executing the spiral attack to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is less than 1, executing the encirclement of the prey to update the position; when the random number is greater than the probability determination coefficient and the absolute value of the vector coefficient is greater than or equal to 1, executing the random search to update the position.

6. The method for predicting the probability of ice-covered distribution lines and broken poles using a deep neural network as claimed in claim 5, characterized in that: The spiral attack position update method is as follows: calculate the distance between the current whale position and the optimal position; update the whale position through the spiral equation, wherein the spiral parameters are used to define the spiral shape.

7. The method for predicting the probability of ice-covered distribution lines and broken poles based on a deep neural network as claimed in claim 6, characterized in that: Also includes: Calculate and compare the individual fitness values ​​of whales in the updated population and the original population; Determine whether to accept the new solution according to MetroPolis guidelines; The temperature values ​​updated using the simulated annealing whale algorithm are used to adjust the weights and biases of a deep neural network.

8. A distribution line ice-covered pole-breakage probability prediction system based on a deep neural network, based on a distribution line ice-covered pole-breakage probability prediction method based on a deep neural network according to any one of claims 1 to 7, characterized in that: It also includes a data acquisition module to obtain historical meteorological data and tower failure information in the distribution line icing disaster area; A construction module is used to construct a deep neural network prediction model and optimize the deep neural network prediction model; The result prediction module uses the simulated annealing whale algorithm to train the deep neural network prediction model to obtain the prediction result of the probability of pole breaking under ice conditions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting the probability of ice-covered distribution lines and broken poles using a deep neural network as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the probability of ice-covered distribution lines and broken poles using a deep neural network according to any one of claims 1 to 7 are implemented.

Images