Icing sensor temperature compensation method, device and equipment and storage medium

Abstract

The embodiment of the invention provides an icing sensor temperature compensation method, device and equipment and a storage medium, and the method comprises the steps: collecting capacitance value data under different temperatures and different ice layer thicknesses, obtaining an initial input data set, dividing the initial input data set into a test set and a training set, building a BP neural network structure based on the training set, and carrying out the temperature compensation of an icing sensor. And iterating the weight and threshold of the BP neural network structure according to a preset optimization target and a hybrid optimization algorithm, training the iterated BP neural network structure based on the training set, determining the iterated weight and threshold according to a training result, and obtaining a temperature compensation model. The temperature compensation model is used for carrying out temperature compensation on the ice layer thickness data obtained by the icing sensor, the icing thickness detection error caused by the fact that the ice dielectric constant is influenced by the temperature is effectively overcome, and the accuracy and reliability of the complex impedance type icing monitoring technology under the complex meteorological condition are improved.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent sensing technology, and in particular to a method, apparatus, device and storage medium for temperature compensation of an icing sensor. Background Technology

[0002] Due to the influence of micro-topography and micro-meteorology, wind turbine blades in high-altitude areas are extremely prone to icing in winter. Icing on wind turbine blades not only affects their aerodynamic performance and reduces the power generation efficiency of the turbine, but also endangers the safety of the wind turbine itself, nearby equipment, and personnel, posing a significant hazard. To reduce the harm caused by blade icing, it is necessary to monitor the icing status of the blades.

[0003] Complex impedance icing detection is a novel icing detection technology. Its principle is to infer the icing state of the surface by measuring the complex impedance parameters of the material on the sensor surface. A complex impedance icing detector typically contains a pair of flexible electrodes arranged in the same plane. Because the capacitance between the two electrodes is affected by the dielectric constant and thickness of the dielectric, complex impedance icing detection can not only identify the dry, wet, and icing states of a surface but also measure the icing thickness. However, since the dielectric constant of ice is affected by temperature, the icing thickness measured by the icing monitoring sensor can be affected, leading to inaccurate detection results. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, this disclosure provides a method, apparatus, device and storage medium for temperature compensation of icing sensor.

[0005] This disclosure provides a method for temperature compensation of an icing sensor, the method comprising: Capacitance data at different temperatures and ice thicknesses are collected to obtain an initial input dataset. The initial input dataset is divided into a test set and a training set. A BP neural network structure is built based on the training set. The weights and thresholds of the BP neural network structure are iterated according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set. The iterative BP neural network structure is trained based on the training set, and the weights and thresholds after iteration are determined according to the training results to obtain the temperature compensation model. The temperature compensation model is used to predict the test set to obtain a predicted value sequence. The predicted value sequence is compared with the actual thickness value sequence in the test set. Based on the comparison result, at least one evaluation index of the temperature compensation model is calculated. At least one index is matched with the corresponding preset application threshold. If all evaluation indexes meet the corresponding preset application threshold, the temperature compensation model is marked as qualified. The temperature compensation model is then used to perform temperature compensation on the ice thickness data obtained by the icing sensor.

[0006] The method provided in this disclosure involves collecting capacitance data at different temperatures and ice thicknesses to obtain an initial input dataset, dividing the initial input dataset into a test set and a training set, and constructing a BP neural network structure based on the training set, including: Capacitance data of sensors under different temperatures and ice thicknesses are collected, and the capacitance data is normalized to obtain the initial input dataset. The initial input data is randomly divided to obtain a training set and a test set; The training set is used to perform feature analysis to determine the number of input layer nodes, and the number of output layer nodes is determined based on the prediction target. A BP neural network structure is built based on the number of input layer nodes, the number of output layer nodes, and the preset number of hidden layer nodes.

[0007] The method provided in this disclosure iterates the weights and thresholds of a BP neural network structure according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set, including: The initial weights and thresholds of the BP neural network are set as the initial solution, and the Cubic mapping is introduced to generate the initial population. A hybrid optimization algorithm is determined using the Eagle algorithm and the simulated annealing algorithm, and the initial population is iterated according to the preset optimization objective and the hybrid optimization algorithm; In the first iteration phase, the Eagle algorithm is used to generate a random number for each initial solution in the initialization population. If the random number is less than a preset threshold, the first mode is adopted. If the random number is greater than or equal to the preset threshold, the second mode is adopted to obtain the iterative population and the first iteration phase ends. The first mode involves performing a global search in the initial population using the Eagle algorithm to discover potential advantageous regions, update the initial solution positions, and obtain the first population. The second mode involves performing a spiral fine-grained search at the current position using the Eagle algorithm, updating the initial solution position, and obtaining the second group. The iterative population is the combination of the first population and the second population; In the second iteration stage, the best individual in the iterative population is selected as the initial solution. The optimal solution is determined by the simulated annealing algorithm and the preset optimization objective. Based on the optimal solution, the weights and thresholds after iteration are obtained, and the second iteration stage ends.

[0008] The method provided in this embodiment selects the optimal individual in the iterative population as the initial solution in the second iteration stage, determines the optimal solution using simulated annealing algorithm and a preset optimization objective, obtains the iterative weights and thresholds based on the optimal solution, and ends the second iteration stage, including: In the second iteration stage, the simulated annealing algorithm is used to determine the initial solution from the iterative population and calculate the function value of the initial solution. The initial temperature and perturbation operator are set, and the initial solution is perturbed according to the perturbation operator to obtain a new solution and calculate the function value of the new solution. The function values ​​of the initial solution and the new solution are input into the Metropolis criterion to calculate the acceptance probability of the new solution. If the function value of the new solution is less than the function value of the initial solution, the new solution must be accepted. If the function value of the new solution is greater than or equal to the function value of the initial solution, the new solution is accepted according to the acceptance probability of the new solution. The initial temperature is cooled down, and the process is repeated until the current temperature is lower than the preset termination temperature. The optimal solution is then output according to the preset optimization target, and the weights and thresholds after the iteration are obtained based on the optimal solution.

[0009] The method provided in this disclosure, which trains an iteratively-adjusted BP neural network structure based on a training set, determines the iteratively-adjusted weights and thresholds based on the training results, and obtains a temperature compensation model, includes: The iterated weights and thresholds are assigned to the BP neural network structure, the training set is input into the iterated BP neural network structure for training, and the predicted ice thickness value is output. The predicted ice thickness value is compared with the actual thickness in the training set, and the loss value of the preset loss function is calculated based on the comparison result. Based on the loss value, starting from the output layer of the BP neural network structure, the gradient of the loss with respect to the weights and thresholds after each iteration is calculated in reverse. The gradient and gradient algorithm are used to update and iterate all the weights and thresholds in the BP neural network structure until the preset training rounds are reached, and then the training process is terminated. Based on the training results at the end of training, the corresponding weights and thresholds are fixed, and the temperature compensation model is obtained based on the BP neural network structure after fixing the weights and thresholds.

[0010] The method provided in this disclosure includes using the temperature compensation model to predict a test set, obtaining a predicted value sequence, comparing the predicted value sequence with the actual thickness value sequence in the test set, calculating at least one evaluation index of the temperature compensation model based on the comparison result, matching each of the at least one index with a corresponding preset application threshold, and marking the temperature compensation model as qualified if all evaluation indices meet the corresponding preset application thresholds. The method also includes using the temperature compensation model to perform temperature compensation on ice thickness data obtained from an icing sensor. The test set is input into the temperature compensation model, and the temperature compensation model outputs the predicted value corresponding to each sample in the training set. A sequence of predicted values ​​is generated based on all the predicted values. The predicted value sequence is compared one by one with the actual thickness value sequence in the test set; Based on the comparison results, calculate multiple evaluation indicators for the temperature compensation model, and compare each evaluation indicator with its corresponding preset application threshold. If all evaluation indicators meet the preset application thresholds, the temperature compensation model is marked as qualified, and temperature compensation is performed on the ice thickness data measured by the icing sensor. If any evaluation metric fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and the retraining process of the temperature compensation model is triggered.

[0011] The method provided in this disclosure also includes: Obtain actual application requirements and determine preset application thresholds based on those requirements; If all evaluation indicators meet the preset application threshold, the temperature compensation model is marked as qualified and can be used in practical applications. If any evaluation indicator fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and a root cause analysis is performed on the temperature compensation model. The temperature compensation model is then retrained based on the results of the root cause analysis. Once it is determined that the temperature compensation model can be used in practical applications, the temperature compensation model is used to compensate the icing thickness data measured by the sensor to obtain the actual icing thickness data.

[0012] This disclosure also provides an icing sensor temperature compensation device, the device comprising: The acquisition module is used to collect capacitance data at different temperatures and ice thicknesses to obtain an initial input dataset. The initial input dataset is divided into a test set and a training set, and a BP neural network structure is built based on the training set. The iteration module is used to iterate the weights and thresholds of the BP neural network structure according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set. The determination module is used to train the iterative BP neural network structure based on the training set, and determine the weights and thresholds after iteration based on the training results to obtain the temperature compensation model. The comparison module is used to predict the test set using the temperature compensation model to obtain a sequence of predicted values, compare the sequence of predicted values ​​with the sequence of actual thickness values ​​in the test set, calculate the evaluation index of the temperature compensation model based on the comparison results, mark the temperature compensation model as qualified based on the evaluation index, and mark the temperature compensation model as qualified if all evaluation indexes meet the preset application threshold. The temperature compensation model is then used to perform temperature compensation on the ice thickness data obtained by the icing sensor.

[0013] In some implementations, the acquisition module is specifically used for: Capacitance data of sensors under different temperatures and ice thicknesses are collected, and the capacitance data is normalized to obtain the initial input dataset. The initial input data is randomly divided to obtain a training set and a test set; The training set is used to perform feature analysis to determine the number of input layer nodes, and the number of output layer nodes is determined based on the prediction target. A BP neural network structure is built based on the number of input layer nodes, the number of output layer nodes, and the preset number of hidden layer nodes.

[0014] In some implementations, the iterative module is specifically used for: The initial weights and thresholds of the BP neural network are set as the initial solution, and the Cubic mapping is introduced to generate the initial population. A hybrid optimization algorithm is determined using the Eagle algorithm and the simulated annealing algorithm, and the initial population is iterated according to the preset optimization objective and the hybrid optimization algorithm; In the first iteration phase, the Eagle algorithm is used to generate a random number for each initial solution in the initialization population. If the random number is less than a preset threshold, the first mode is adopted. If the random number is greater than or equal to the preset threshold, the second mode is adopted to obtain the iterative population and the first iteration phase ends. The first mode involves performing a global search in the initial population using the Eagle algorithm to discover potential advantageous regions, update the initial solution positions, and obtain the first population. The second mode involves performing a spiral fine-grained search at the current position using the Eagle algorithm, updating the initial solution position, and obtaining the second group. The iterative population is the combination of the first population and the second population; In the second iteration stage, the best individual in the iterative population is selected as the initial solution. The optimal solution is determined by the simulated annealing algorithm and the preset optimization objective. Based on the optimal solution, the weights and thresholds after iteration are obtained, and the second iteration stage ends.

[0015] In some implementations, the iterative module is specifically used for: In the second iteration stage, the simulated annealing algorithm is used to determine the initial solution from the iterative population and calculate the function value of the initial solution. The initial temperature and perturbation operator are set, and the initial solution is perturbed according to the perturbation operator to obtain a new solution and calculate the function value of the new solution. The function values ​​of the initial solution and the new solution are input into the Metropolis criterion to calculate the acceptance probability of the new solution. If the function value of the new solution is less than the function value of the initial solution, the new solution must be accepted. If the function value of the new solution is greater than or equal to the function value of the initial solution, the new solution is accepted according to the acceptance probability of the new solution. The initial temperature is cooled down, and the process is repeated until the current temperature is lower than the preset termination temperature. The optimal solution is then output according to the preset optimization target, and the weights and thresholds after the iteration are obtained based on the optimal solution.

[0016] In some implementations, the determining module is specifically used for: The iterated weights and thresholds are assigned to the BP neural network structure, the training set is input into the iterated BP neural network structure for training, and the predicted ice thickness value is output. The predicted ice thickness value is compared with the actual thickness in the training set, and the loss value of the preset loss function is calculated based on the comparison result. Based on the loss value, starting from the output layer of the BP neural network structure, the gradient of the loss with respect to the weights and thresholds after each iteration is calculated in reverse. The gradient and gradient algorithm are used to update and iterate all the weights and thresholds in the BP neural network structure until the preset training rounds are reached, and then the training process is terminated. Based on the training results at the end of training, the corresponding weights and thresholds are fixed, and the temperature compensation model is obtained based on the BP neural network structure after fixing the weights and thresholds.

[0017] In some implementations, the comparison module is specifically used for: The test set is input into the temperature compensation model, and the temperature compensation model outputs the predicted value corresponding to each sample in the training set. A sequence of predicted values ​​is generated based on all the predicted values. The predicted value sequence is compared one by one with the actual thickness value sequence in the test set; Based on the comparison results, calculate multiple evaluation indicators for the temperature compensation model, and compare each evaluation indicator with its corresponding preset application threshold. If all evaluation indicators meet the preset application thresholds, the temperature compensation model is marked as qualified, and temperature compensation is performed on the ice thickness data measured by the icing sensor. If any evaluation metric fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and the retraining process of the temperature compensation model is triggered.

[0018] In some implementations, the comparison module is specifically used for: Obtain actual application requirements and determine preset application thresholds based on those requirements; If all evaluation indicators meet the preset application threshold, the temperature compensation model is marked as qualified and can be used in practical applications. If any evaluation indicator fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and a root cause analysis is performed on the temperature compensation model. The temperature compensation model is then retrained based on the results of the root cause analysis. Once it is determined that the temperature compensation model can be used in practical applications, the temperature compensation model is used to compensate the icing thickness data measured by the sensor to obtain the actual icing thickness data.

[0019] This disclosure also provides an electronic device, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the icing sensor temperature compensation method as provided in this disclosure.

[0020] This disclosure also provides a computer-readable storage medium storing a computer program for performing the icing sensor temperature compensation method as provided in this disclosure.

[0021] The technical solution provided in this disclosure has the following advantages compared with the prior art: The icing sensor temperature compensation method disclosed in this embodiment collects capacitance data at different temperatures and ice thicknesses, dividing it into training and testing sets. A hybrid optimization algorithm iteratively optimizes the weights and thresholds of a BP neural network to minimize prediction errors, resulting in a temperature compensation model after training. The model's prediction accuracy on the testing set is evaluated; if it passes the evaluation, it is used to compensate for the measured thickness data of the icing sensor. This effectively overcomes the icing thickness detection error caused by the temperature-dependent influence of ice dielectric constant, improves the accuracy and reliability of complex impedance-based icing monitoring technology under complex meteorological conditions, and provides a more accurate data foundation for judging the icing status of wind turbine blades and providing safety warnings. Attached Figure Description

[0022] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0023] Figure 1 A schematic flowchart illustrating the icing sensor temperature compensation method provided in this embodiment of the disclosure; Figure 2 A detailed flowchart of the icing sensor temperature compensation method provided in the embodiments of this disclosure; Figure 3 This is a schematic diagram of the structure of the icing sensor temperature compensation device provided in the embodiments of this disclosure; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0024] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0025] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0026] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0027] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0028] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0029] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0030] To address the aforementioned issues, this disclosure provides a method for temperature compensation of an icing sensor, which will be described below with reference to specific embodiments.

[0031] Figure 1 This is a flowchart illustrating a method for temperature compensation of an icing sensor according to an embodiment of the present disclosure. The method can be executed by an icing sensor temperature compensation device, which can be implemented in software and / or hardware and is generally integrated into an electronic device.

[0032] Example 1: This embodiment of the disclosure provides a method for temperature compensation of an icing sensor, the method comprising: Capacitance data at different temperatures and ice thicknesses are collected to obtain an initial input dataset. The initial input dataset is divided into a test set and a training set. A BP neural network structure is built based on the training set. The weights and thresholds of the BP neural network structure are iterated according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set. The iterative BP neural network structure is trained based on the training set, and the weights and thresholds after iteration are determined according to the training results to obtain the temperature compensation model. The temperature compensation model is used to predict the test set to obtain a predicted value sequence. The predicted value sequence is compared with the actual thickness value sequence in the test set. Based on the comparison result, at least one evaluation index of the temperature compensation model is calculated. At least one index is matched with the corresponding preset application threshold. If all evaluation indexes meet the corresponding preset application threshold, the temperature compensation model is marked as qualified. The temperature compensation model is then used to perform temperature compensation on the ice thickness data obtained by the icing sensor.

[0033] In this embodiment, the capacitance value data is the raw electrical signal directly acquired by the icing sensor. The value of this signal changes with the dielectric constant of the sensor surface, which in turn is affected by the thickness of the ice layer and environmental conditions. For example, under specific experimental conditions, a set of raw data measured by the sensor might be: 155.6 pF at -5°C with a 2 mm ice thickness; and 388.9 pF at 0°C with a 5 mm ice thickness.

[0034] In this embodiment, the initial input dataset is a standard dataset obtained by normalizing all the original capacitance values. Normalization aims to eliminate dimensions and scale the data to a uniform range, thereby improving the training efficiency and stability of the model. For example, suppose the collected original capacitance values ​​range from 100pF to 500pF. After normalization, 155.6pF might become 0.139, and 388.9pF might become 0.722. The initial input dataset is a collection of countless such [temperature, normalized capacitance value] values ​​and their corresponding ice thicknesses.

[0035] In this embodiment, the training set and test set are two parts randomly split from the initial input dataset to evaluate the model's generalization ability. The training set is used to train and optimize the model parameters. The test set is used to finally evaluate the model's performance on unseen data to test its generalization ability. For example, suppose there are 1000 sets of data. 800 sets are randomly selected as the training set to build and train the BP neural network. The remaining 200 sets are used as the test set to simulate real-world applications after the model training is complete, to test the model's performance.

[0036] In this embodiment, sensor capacitance data at different temperatures and ice thicknesses are collected, normalized to obtain an initial dataset, and divided into a training set and a test set. The features of the training set are analyzed to determine the number of nodes in the network input and output layers. Combined with a preset number of hidden layer nodes, the BP neural network structure is constructed.

[0037] In this embodiment, the iterative process uses the initial weights and thresholds of the BP network as the initial solution, and generates a population through Cubic mapping. First, a hybrid iterative process of global exploration and local development is performed using the Eagle algorithm. Then, the individual with the best result is used as the initial solution, and a simulated annealing algorithm is employed for fine-tuning to finally obtain the optimized weights and thresholds.

[0038] In this embodiment, prediction error is a core performance quantification indicator, which specifically refers to the difference between the ice thickness value predicted by the temperature compensation model and the actual ice thickness value measured by standard methods.

[0039] In this embodiment, the comparison process involves comparing the model's predicted values ​​with the known true values ​​in the dataset and calculating the differences in a quantifiable manner. This process is a prerequisite for calculating the loss function. For example, the model predicts a thickness of 2.1 mm, but the actual ice thickness label for that sample is 2.0 mm. The comparison process identifies this 0.1 mm error and feeds it into the loss function for calculation.

[0040] In this embodiment, the predicted value sequence is a list or set of predicted ice thickness values ​​output by the model for each sample when all samples in the test set are sequentially input into a pre-trained temperature compensation model. For example, the test set contains 200 samples collected at different temperatures and with different actual ice thicknesses. Inputting this batch of data into the model will cause the model to output 200 predicted thickness values ​​sequentially; this list containing the 200 predicted values ​​is the predicted value sequence.

[0041] In this embodiment, the preset application threshold is a minimum qualified standard set in advance for each evaluation index based on the accuracy and reliability requirements of actual engineering applications. For example, based on the needs of wind farm safety monitoring, the preset application threshold set for the temperature compensation model may be root mean square error ≤ 0.20 mm, mean absolute error ≤ 0.18 mm, and coefficient of determination ≥ 0.95.

[0042] In this embodiment, the evaluation metrics may include: root mean square error, mean absolute error, and coefficient of determination.

[0043] In this embodiment, the BP neural network structure is the overall architecture definition of the input layer, hidden layer, and output layer described above. The structure of the BP neural network algorithm includes an input layer, a hidden layer, and an output layer. The input layer has 3 neurons, representing temperature, capacitance value, and the measured ice thickness, respectively; the output layer has 1 neuron, representing the ice thickness after temperature compensation; and the hidden layer has 7 neurons, with 4 in the first layer and 3 in the second layer.

[0044] The working principle and beneficial effects of this embodiment are as follows: Capacitance data at different temperatures and ice thicknesses are collected and divided into training and testing sets. A hybrid optimization algorithm is used to iteratively optimize the weights and thresholds of a BP neural network to minimize prediction errors, and a temperature compensation model is obtained after training. The prediction accuracy of the model on the testing set is evaluated, and if it passes the evaluation, it is used to compensate for the measured thickness data of the icing sensor. This effectively overcomes the icing thickness detection error caused by the influence of temperature on the ice dielectric constant, improves the accuracy and reliability of complex impedance icing monitoring technology under complex meteorological conditions, and provides a more accurate data foundation for judging the icing status of wind turbine blades and providing safety warnings.

[0045] Example 2: The method provided in this embodiment of the present disclosure collects capacitance data at different temperatures and ice thicknesses to obtain an initial input dataset. The initial input dataset is divided into a test set and a training set. A BP neural network structure is built based on the training set, including: Capacitance data of sensors under different temperatures and ice thicknesses are collected, and the capacitance data is normalized to obtain the initial input dataset. The initial input data is randomly divided to obtain a training set and a test set; The training set is used to perform feature analysis to determine the number of input layer nodes, and the number of output layer nodes is determined based on the prediction target. A BP neural network structure is built based on the number of input layer nodes, the number of output layer nodes, and the preset number of hidden layer nodes.

[0046] In this embodiment, the network structure design uses an input layer with two nodes, the number of which is the same as the number of neurons in the input layer of a backpropagation (BP) neural network, determined by the feature dimensions of the training data. For example, in a scenario, each training sample has two features: temperature and normalized capacitance value. Therefore, the number of input layer nodes should be set to two, with one node receiving the temperature data and the other receiving the capacitance value.

[0047] In this embodiment, the number of output layer nodes is the number of neurons contained in the output layer of the BP neural network, which is determined by the dimension of the prediction target.

[0048] In this embodiment, the preset number of hidden layer nodes is the number of neurons in the hidden layer. This number is not directly determined by the data, but needs to be preset based on experience or experiments, and determines the network's ability to learn complex patterns.

[0049] The working principle and beneficial effects of this embodiment are as follows: Sensor capacitance data at different temperatures and ice thicknesses are collected, normalized to obtain an initial dataset, and divided into training and testing sets. The features of the training set are analyzed to determine the number of nodes in the network input and output layers. Combined with a preset number of hidden layer nodes, a BP neural network structure is constructed. Through systematic data acquisition and network structure customization, a basic framework for a BP neural network specifically for ice thickness detection is built, laying a reliable model foundation for subsequent high-precision temperature compensation using hybrid algorithms, and effectively improving the adaptability of thickness monitoring.

[0050] Example 3: The method provided in this embodiment iterates the weights and thresholds of a BP neural network structure according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set, including: The initial weights and thresholds of the BP neural network are set as the initial solution, and the Cubic mapping is introduced to generate the initial population. A hybrid optimization algorithm is determined using the Eagle algorithm and the simulated annealing algorithm, and the initial population is iterated according to the preset optimization objective and the hybrid optimization algorithm; In the first iteration phase, the Eagle algorithm is used to generate a random number for each initial solution in the initialization population. If the random number is less than a preset threshold, the first mode is adopted. If the random number is greater than or equal to the preset threshold, the second mode is adopted to obtain the iterative population and the first iteration phase ends. The first mode involves performing a global search in the initial population using the Eagle algorithm to discover potential advantageous regions, update the initial solution positions, and obtain the first population. The second mode involves performing a spiral fine-grained search at the current position using the Eagle algorithm, updating the initial solution position, and obtaining the second group. The iterative population is the combination of the first population and the second population; In the second iteration stage, the best individual in the iterative population is selected as the initial solution. The optimal solution is determined by the simulated annealing algorithm and the preset optimization objective. Based on the optimal solution, the weights and thresholds after iteration are obtained, and the second iteration stage ends.

[0051] In this embodiment, the initial solution corresponds to a complete set of weights and thresholds for the BP neural network; the Cubic mapping is a chaotic mapping formula used to generate more uniform and ergodic random sequences, and the initial population is a set of multiple initial solutions. For example, a set of BP network parameters is randomly generated as the initial solution. The Cubic mapping is used to generate 50 different combinations of initial weights and thresholds, forming the initial population.

[0052] In this embodiment, the Cubic mapping formula is: Where a represents the chaos parameter, and a = 2.595 is taken; This represents the initial solution for the next mapping; This represents the initial solution.

[0053] In this embodiment, the hybrid optimization algorithm is an optimization strategy that combines the advantages of the Eagle algorithm and the simulated annealing algorithm. The algorithm updates and improves the initial solutions in the population generation by generation according to a preset objective, thus continuously improving the overall performance.

[0054] In this embodiment, the preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set. The ultimate goal of the algorithm iteration is usually a function that needs to be minimized or maximized. For example, the hybrid optimization algorithm first executes the Eagle algorithm to search for a large range of excellent parameter regions, and then executes the simulated annealing algorithm for fine adjustment. The iterative process is manifested as follows: the prediction error of the first generation of 50 sets of parameters is large, and a slightly better second generation is generated through algorithm rule updates, and so on for repeated improvement. The preset optimization objective is to minimize the prediction error of the BP neural network on the training set, that is, to minimize the mean square error between the ice thickness predicted by the network and the actual ice thickness.

[0055] In this embodiment, the first iteration stage is the first two-thirds of the entire iteration process. This stage is specifically for executing the search strategy of the Eagle Algorithm, the first model and the second mode. The preset threshold is a critical value used to switch between the two modes of the Eagle Algorithm, usually set to 0.5.

[0056] In this embodiment, the first mode involves moving the initial solution away from its current position and exploring a wide range of parameters throughout the parameter space to discover advantageous parameter regions that can significantly reduce prediction errors. All initial solutions updated through this mode constitute the first group.

[0057] In this embodiment, the formula corresponding to the first mode is: ,in, This represents the new position of the initial solution of the first mode at the (t+1)th iteration, and the new combination of weights and thresholds; This represents the optimal position of the first group in the first mode up to the t-th iteration, i.e., the optimal combination of weights and thresholds; t represents the index of the current iteration number; T represents the maximum number of iterations. This represents the average position of the entire population at the t-th iteration, i.e., the average state of the combination of average weight and threshold; rand1 represents the random number of the first pattern uniformly distributed in the range [0, 1]; Dim represents the total number of weights and thresholds that need to be optimized in the initial solution. The position of the i-th individual is represented by the weight and threshold combination of the i-th initial solution; N represents the number of initial solutions in the first population; and i represents the index of the initial solution in the first population.

[0058] In this embodiment, the second mode involves performing a small-scale, spiral-like fine-tuning search of the initial solution in the vicinity of the current parameter combination, attempting to find better fine-tuning parameters. All initial solutions updated through this mode constitute the second group.

[0059] In this embodiment, the formula corresponding to the second mode is: ,in, This represents the new position of the initial solution of the second mode at the (t+1)th iteration, and the new combination of weights and thresholds; Represents the Lévy flight function; This represents the optimal position of the second population up to the t-th iteration in the second mode, i.e., the optimal combination of weights and thresholds. represents the position of the randomly selected initial solution, i.e., the combination of randomly selected weights and thresholds; rand2 represents a random number uniformly distributed in the range [0, 1] for the second mode; x1 and y1 represent the shape of the spiral flight; the formula corresponding to the Lévy flight function is... , where s represents the scaling factor; u and v represent random numbers that follow a standard normal distribution; The standard deviation is represented by the formula: ; The exponential parameter represents the Lévy distribution; This represents the gamma function.

[0060] In this embodiment, the spiral search formula for the second mode is... Where r represents the current radius of the initial solution position on the spiral trajectory; r1 represents the initial radius of the spiral; U represents the rate constant of the spiral radius as D1 increases; and D1 represents the integer sequence from 1 to the maximum dimension of the search space. The current angle represents the position of the initial solution on the spiral trajectory; This represents the rate constant of change of the helix angle with respect to D1; This indicates the initial angle of the spiral.

[0061] In this embodiment, the iterative population is a new population formed by merging the first and second populations after the first iteration phase. For example, in the first iteration phase, random numbers are generated for each set of parameters. If the random numbers are less than a preset threshold of 0.5, a global search is used. For instance, a set of parameters that originally focused on predicting small thicknesses may jump to a new parameter space that is more suitable for predicting medium thicknesses. This potential advantageous region is found to have lower errors, and the initial solution position is updated before being incorporated into the first population. Otherwise, a spiral fine-tuning search is used to fine-tune a set of parameters that has performed well, reducing the prediction error from 0.12mm to 0.11mm. The initial solution position is then updated before being incorporated into the second population. Merging the two populations yields a new set of 50 parameters, i.e., the iterative population, whose overall performance is better than the initial population.

[0062] In this embodiment, the second iteration stage, based on the results of the Skyhawk algorithm, constitutes the last third of the entire iteration and executes the simulated annealing algorithm. The optimal solution and the weights and thresholds after iteration represent the final optimal parameter combination found by the entire hybrid optimization algorithm. For example, in the second iteration stage, the set of parameters with the smallest prediction error is selected from the iterative population as the starting point, and the simulated annealing algorithm is used for fine-tuning iterations. The final output is the optimal solution, i.e., the weights and thresholds after iteration. This set of parameters enables the BP network to most accurately compensate for the sensor's temperature and capacitance values, outputting the ice layer thickness closest to the true value.

[0063] The working principle and beneficial effects of this embodiment are as follows: The initial weights and thresholds of the BP network are used as initial solutions, and a population is generated through Cubic mapping. First, a hybrid iteration of global exploration and local development is performed using the Eagle algorithm. Then, the individual with the best result is used as the initial solution, and simulated annealing is employed for fine-tuning. Finally, optimized weights and thresholds are obtained. Combining the extensive search capability of the Eagle algorithm with the local refinement advantage of simulated annealing, the sensitivity of the BP neural network to initial values ​​is effectively overcome, significantly improving the optimization quality of weights and thresholds, and laying a solid foundation for constructing a high-precision temperature compensation model.

[0064] Example 4: The method provided in this embodiment selects the optimal individual in the iterative population as the initial solution in the second iteration stage, determines the optimal solution using simulated annealing algorithm and a preset optimization objective, obtains the weights and thresholds after iteration based on the optimal solution, and the second iteration stage ends, including: In the second iteration stage, the simulated annealing algorithm is used to determine the initial solution from the iterative population and calculate the function value of the initial solution. The initial temperature and perturbation operator are set, and the initial solution is perturbed according to the perturbation operator to obtain a new solution and calculate the function value of the new solution. The function values ​​of the initial solution and the new solution are input into the Metropolis criterion to calculate the acceptance probability of the new solution. If the function value of the new solution is less than the function value of the initial solution, the new solution must be accepted. If the function value of the new solution is greater than or equal to the function value of the initial solution, the new solution is accepted according to the acceptance probability of the new solution. The initial temperature is cooled down, and the process is repeated until the current temperature is lower than the preset termination temperature. The optimal solution is then output according to the preset optimization target, and the weights and thresholds after the iteration are obtained based on the optimal solution.

[0065] In this embodiment, the initial temperature is a high value at the start of the algorithm, determining the probability of accepting bad solutions initially. The higher the temperature, the greater the likelihood of accepting inferior solutions and the wider the search range. The perturbation operator is a rule that makes small, random modifications to the initial weights and thresholds of the current solution, used to explore new solutions near the current position. For example, the initial temperature is set to 100℃. The rule of the perturbation operator is to randomly add a value in the range [-0.1, 0.1] to each element of the current optimal weight vector.

[0066] In this embodiment, the new solution is a new set of candidate weights and thresholds obtained by perturbing the current solution using a perturbation operator. The function value of the new solution is the numerical value calculated by substituting the new solution into the objective function, which directly reflects the quality of the set of parameters. For example, after applying the perturbation operator, a new set of solutions, new weights, and thresholds are obtained. This new set of parameters is assigned to the BP neural network and run once on the entire training set, calculating the root mean square error between the predicted thickness and the actual thickness as 0.13 mm. This 0.13 mm is the function value of the new solution.

[0067] In this embodiment, the Metropolis criterion is the core rule of the simulated annealing algorithm, used to decide whether to accept a worse new solution in order to avoid the algorithm getting trapped in a local optimum too early. The acceptance probability of a new solution is a probability value calculated according to the Metropolis criterion. If the acceptance probability of a new solution is 0.999, that is, the new solution has an error deterioration of 0.001, but there is still a 99.9% probability of accepting this worse solution, thus giving it a chance to escape the current local optimum.

[0068] In this embodiment, the preset termination temperature is a threshold at which the algorithm stops. The iteration ends when the system temperature cools below this threshold. For example, the preset termination temperature is 0.1℃. When the current temperature drops from 100℃ to 0.09℃ through a cooling strategy, the algorithm stops searching.

[0069] In this embodiment, the optimal solution is the set of parameters with the minimum function value, i.e. the minimum prediction error, found throughout the entire simulated annealing iteration process.

[0070] In this embodiment, the iterated weights and thresholds, i.e., the specific content of the optimal solution, are the optimized final parameters to be assigned to the BP neural network. For example, when the algorithm terminates, the recorded optimal solution has a function value of 0.08 mm. This set of parameters that optimizes the network performance, namely the iterated weights and thresholds, will be used to construct the final temperature compensation model.

[0071] In this embodiment, the initial temperature K for simulated annealing is set, the current solution is x, and the target value corresponding to the current solution is calculated. . This represents the difference between the temperature-compensated icing thickness and the actual icing thickness. A new solution is generated within its neighborhood by randomly perturbing the current solution x. That is, to generate a new solution using Gaussian noise. Calculate the corresponding new solution function value. The acceptance probability of the new solution is calculated according to the Metropolis criterion, and the corresponding formula is: , where P represents the probability of accepting the new solution; This represents the function value of the new solution; Indicates the new solution; x represents the current solution; This represents the function value of the current solution; when P=1, the new solution is accepted directly; otherwise, it is... Accept the new solution, then cool the temperature K by the cooling factor α, where α = 0.98, i.e., K = αK. Repeat this process until the current temperature K is lower than the preset termination temperature, and then output the best solution.

[0072] The working principle and beneficial effects of this embodiment are as follows: In the second iteration stage, the optimal individual output by the Skyhawk algorithm is used as the initial solution. A new solution is generated by the perturbation of the simulated annealing mechanism. Based on the Metropolis criterion, inferior solutions are accepted with probability. After continuous cooling iteration until the termination temperature, the optimal weights and thresholds are finally output. The simulated annealing algorithm effectively escapes local optima by accepting inferior solutions with controllable probability, ensuring the global search capability of the weight and threshold optimization process, and enabling the BP neural network to obtain more stable and reliable initial parameters.

[0073] Example 5: The method provided in this embodiment of the present disclosure trains the iteratively trained BP neural network structure based on the training set, determines the iteratively trained weights and thresholds based on the training results, and obtains a temperature compensation model, including: The iterated weights and thresholds are assigned to the BP neural network structure, the training set is input into the iterated BP neural network structure for training, and the predicted ice thickness value is output. The predicted ice thickness value is compared with the actual thickness in the training set, and the loss value of the preset loss function is calculated based on the comparison result. Based on the loss value, starting from the output layer of the BP neural network structure, the gradient of the loss with respect to the weights and thresholds after each iteration is calculated in reverse. The gradient and gradient algorithm are used to update and iterate all the weights and thresholds in the BP neural network structure until the preset training rounds are reached, and then the training process is terminated. Based on the training results at the end of training, the corresponding weights and thresholds are fixed, and the temperature compensation model is obtained based on the BP neural network structure after fixing the weights and thresholds.

[0074] In this embodiment, the predicted ice thickness value is the result calculated and output by the backpropagation (BP) neural network after the training data is input. It is an estimate of the true thickness based on the input temperature and capacitance values. For example, if a sample with a temperature of -5°C and a normalized capacitance value of 0.65 is input into the network, the network will output a value, such as 2.1 mm. This 2.1 mm is the ice thickness predicted by the model.

[0075] In this embodiment, the output layer of the BP neural network structure is the last layer of the neural network, responsible for generating the final prediction result. In this scenario, there is typically only one neuron, using a linear activation function to directly output a continuous value representing the ice thickness.

[0076] In this embodiment, the preset training epochs are a key hyperparameter, referring to the number of times the entire training set data is completely trained through the neural network. This determines the model's learning cycle. For example, assuming the training set has 800 samples, the preset training epochs are set to 1000. This means that each sample will be used to train the model 1000 times, or overall, the algorithm will perform 1000 complete learning and parameter updates on the entire dataset of 800 samples. After 1000 training iterations, the training process will terminate regardless of other conditions.

[0077] The working principle and beneficial effects of this embodiment are as follows: the optimized weights and thresholds are assigned to the BP network, the training set is input for forward propagation and the prediction error is calculated, the gradient is calculated through the backpropagation algorithm, the network parameters are iteratively updated using the gradient descent method, and finally the trained weights and thresholds are fixed to obtain the temperature compensation model. By combining the high-quality initial parameters provided by the intelligent optimization algorithm with the powerful local search capability of the BP algorithm, the model convergence is effectively accelerated and its generalization performance is improved, and finally a high-precision and high-stability ice layer thickness temperature compensation model is constructed.

[0078] Example 6: The method provided in this embodiment of the present disclosure uses the temperature compensation model to predict the test set, obtains a predicted value sequence, compares the predicted value sequence with the actual thickness value sequence in the test set, calculates at least one evaluation index of the temperature compensation model based on the comparison result, and matches each of the at least one index with a corresponding preset application threshold. If all evaluation indicators meet the corresponding preset application threshold, the temperature compensation model is marked as qualified. The method uses the temperature compensation model to perform temperature compensation on the ice thickness data obtained from the icing sensor, including: The test set is input into the temperature compensation model, and the temperature compensation model outputs the predicted value corresponding to each sample in the training set. A sequence of predicted values ​​is generated based on all the predicted values. The predicted value sequence is compared one by one with the actual thickness value sequence in the test set; Based on the comparison results, calculate multiple evaluation indicators for the temperature compensation model, and compare each evaluation indicator with its corresponding preset application threshold. If all evaluation indicators meet the preset application thresholds, the temperature compensation model is marked as qualified, and temperature compensation is performed on the ice thickness data measured by the icing sensor. If any evaluation metric fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and the retraining process of the temperature compensation model is triggered.

[0079] In this embodiment, the comparison results are evaluated using multiple metrics calculated by systematically comparing the predicted value sequence with the corresponding true thickness value sequence in the test set. These metrics quantify model performance from different perspectives. In this embodiment, the retraining process is a series of improvement measures triggered when the temperature compensation model is marked as unqualified. The aim is to improve model performance so that it eventually passes the qualification test. For example, if a model obtained from a training iteration has a root mean square error ≤ 0.25mm but greater than the threshold of 0.20mm, it is marked as unqualified. The system then triggers the retraining process, including: adjusting the network structure, such as increasing the number of neurons in the hidden layer; adjusting the parameters of the Eagle-simulated annealing hybrid optimization algorithm; or using data augmentation methods to expand the training dataset, and then restarting the entire data preparation-model optimization-training process with the new configuration.

[0080] The working principle and beneficial effects of this embodiment are as follows: A test set is input into the temperature compensation model to obtain a predicted value sequence, which is then compared with the actual thickness sequence to calculate multiple evaluation indicators. Each indicator is compared with a preset threshold; if all indicators meet the threshold, the model is marked as qualified and put into application; otherwise, it is marked as unqualified and retraining is triggered. Through rigorous test set verification and multi-indicator threshold determination, the temperature compensation model is ensured to meet the preset accuracy requirements, effectively guaranteeing the accuracy and reliability of icing thickness monitoring data and providing reliable data support for the safe operation of wind turbines.

[0081] Example 7: The method provided in this embodiment of the disclosure further includes: Obtain actual application requirements and determine preset application thresholds based on those requirements; If all evaluation indicators meet the preset application threshold, the temperature compensation model is marked as qualified and can be used in practical applications. If any evaluation indicator fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and a root cause analysis is performed on the temperature compensation model. The temperature compensation model is then retrained based on the results of the root cause analysis. Once it is determined that the temperature compensation model can be used in practical applications, the temperature compensation model is used to compensate the icing thickness data measured by the sensor to obtain the actual icing thickness data.

[0082] In this embodiment, practical application requirements refer to the specific performance and requirements that the model must meet when deployed in real industrial scenarios to ensure safety, improve efficiency, and other objectives. These are not theoretical metrics, but rather hard constraints driven by business goals.

[0083] In this embodiment, determining the preset application threshold is a process of transforming qualitative practical application requirements into quantitative, measurable model performance standards. For example, to meet safety requirements, the root mean square error of the model is required to be ≤0.5mm. Thus, when the actual ice thickness reaches 10mm, the model's predicted value is between 9.5mm and 10.5mm, ensuring that the alarm is triggered within a safe range.

[0084] In this embodiment, the actual icing thickness data refers to the raw measurement data collected in real-time by the icing sensor on-site, which includes temperature interference. This raw data is then input into a qualified temperature compensation model, resulting in a final output value that eliminates temperature effects and more closely approximates the physical reality of the ice thickness. For example, in an environment of -10°C, the raw reading directly measured by the icing sensor is 3.5 mm. This reading is inflated due to the change in the dielectric constant of ice at low temperatures. Inputting this raw reading along with the ambient temperature (-10°C) into a qualified temperature compensation model, the model outputs a corrected actual icing thickness of 2.9 mm. The wind farm control system will use this more accurate 2.9 mm data to assess risks and make decisions, rather than the distorted raw 3.5 mm data.

[0085] The working principle and beneficial effects of this embodiment are as follows: Application thresholds are set according to actual needs. The qualification of the temperature compensation model is determined by comparing the evaluation indicators with the thresholds. Qualified models are put into application and compensate for the measured sensor data. Unqualified models undergo root cause analysis and retraining. A complete quality control process is established from requirement definition and model verification to application deployment, ensuring that the finally deployed compensation model truly meets engineering accuracy requirements and effectively improves the accuracy and practicality of icing monitoring data.

[0086] To achieve the above embodiments, this disclosure also proposes an icing sensor temperature compensation device.

[0087] Figure 3 This is a schematic diagram of the structure of the icing sensor temperature compensation device provided in an embodiment of this disclosure. This device can be implemented by software and / or hardware, and is generally integrated into an electronic device. Figure 3 As shown, the device includes: a data acquisition module, an iteration module, a determination module, and a comparison module, wherein... The acquisition module is used to collect capacitance data at different temperatures and ice thicknesses to obtain an initial input dataset. The initial input dataset is divided into a test set and a training set, and a BP neural network structure is built based on the training set. The iteration module is used to iterate the weights and thresholds of the BP neural network structure according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set. The determination module is used to train the iterative BP neural network structure based on the training set, and determine the weights and thresholds after iteration based on the training results to obtain the temperature compensation model. The comparison module is used to predict the test set using the temperature compensation model to obtain a sequence of predicted values, compare the sequence of predicted values ​​with the sequence of actual thickness values ​​in the test set, calculate the evaluation index of the temperature compensation model based on the comparison results, mark the temperature compensation model as qualified based on the evaluation index, and mark the temperature compensation model as qualified if all evaluation indexes meet the preset application threshold. The temperature compensation model is then used to perform temperature compensation on the ice thickness data obtained by the icing sensor.

[0088] In some implementations, the acquisition module is specifically used for: Capacitance data of sensors under different temperatures and ice thicknesses are collected, and the capacitance data is normalized to obtain the initial input dataset. The initial input data is randomly divided to obtain a training set and a test set; The training set is used to perform feature analysis to determine the number of input layer nodes, and the number of output layer nodes is determined based on the prediction target. A BP neural network structure is built based on the number of input layer nodes, the number of output layer nodes, and the preset number of hidden layer nodes.

[0089] In some implementations, the iterative module is specifically used for: The initial weights and thresholds of the BP neural network are set as the initial solution, and the Cubic mapping is introduced to generate the initial population. A hybrid optimization algorithm is determined using the Eagle algorithm and the simulated annealing algorithm, and the initial population is iterated according to the preset optimization objective and the hybrid optimization algorithm; In the first iteration phase, the Eagle algorithm is used to generate a random number for each initial solution in the initialization population. If the random number is less than a preset threshold, the first mode is adopted. If the random number is greater than or equal to the preset threshold, the second mode is adopted to obtain the iterative population and the first iteration phase ends. The first mode involves performing a global search in the initial population using the Eagle algorithm to discover potential advantageous regions, update the initial solution positions, and obtain the first population. The second mode involves performing a spiral fine-grained search at the current position using the Eagle algorithm, updating the initial solution position, and obtaining the second group. The iterative population is the combination of the first population and the second population; In the second iteration stage, the best individual in the iterative population is selected as the initial solution. The optimal solution is determined by the simulated annealing algorithm and the preset optimization objective. Based on the optimal solution, the weights and thresholds after iteration are obtained, and the second iteration stage ends.

[0090] In some implementations, the iterative module is specifically used for: In the second iteration stage, the simulated annealing algorithm is used to determine the initial solution from the iterative population and calculate the function value of the initial solution. The initial temperature and perturbation operator are set, and the initial solution is perturbed according to the perturbation operator to obtain a new solution and calculate the function value of the new solution. The function values ​​of the initial solution and the new solution are input into the Metropolis criterion to calculate the acceptance probability of the new solution. If the function value of the new solution is less than the function value of the initial solution, the new solution must be accepted. If the function value of the new solution is greater than or equal to the function value of the initial solution, the new solution is accepted according to the acceptance probability of the new solution. The initial temperature is cooled down, and the process is repeated until the current temperature is lower than the preset termination temperature. The optimal solution is then output according to the preset optimization target, and the weights and thresholds after the iteration are obtained based on the optimal solution.

[0091] In some implementations, the determining module is specifically used for: The iterated weights and thresholds are assigned to the BP neural network structure, the training set is input into the iterated BP neural network structure for training, and the predicted ice thickness value is output. The predicted ice thickness value is compared with the actual thickness in the training set, and the loss value of the preset loss function is calculated based on the comparison result. Based on the loss value, starting from the output layer of the BP neural network structure, the gradient of the loss with respect to the weights and thresholds after each iteration is calculated in reverse. The gradient and gradient algorithm are used to update and iterate all the weights and thresholds in the BP neural network structure until the preset training rounds are reached, and then the training process is terminated. Based on the training results at the end of training, the corresponding weights and thresholds are fixed, and the temperature compensation model is obtained based on the BP neural network structure after fixing the weights and thresholds.

[0092] In some implementations, the comparison module is specifically used for: The test set is input into the temperature compensation model, and the temperature compensation model outputs the predicted value corresponding to each sample in the training set. A sequence of predicted values ​​is generated based on all the predicted values. The predicted value sequence is compared one by one with the actual thickness value sequence in the test set; Based on the comparison results, calculate multiple evaluation indicators for the temperature compensation model, and compare each evaluation indicator with its corresponding preset application threshold. If all evaluation indicators meet the preset application thresholds, the temperature compensation model is marked as qualified, and temperature compensation is performed on the ice thickness data measured by the icing sensor. If any evaluation metric fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and the retraining process of the temperature compensation model is triggered.

[0093] In some implementations, the comparison module is specifically used for: Obtain actual application requirements and determine preset application thresholds based on those requirements; If all evaluation indicators meet the preset application threshold, the temperature compensation model is marked as qualified and can be used in practical applications. If any evaluation indicator fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and a root cause analysis is performed on the temperature compensation model. The temperature compensation model is then retrained based on the results of the root cause analysis. Once it is determined that the temperature compensation model can be used in practical applications, the temperature compensation model is used to compensate the icing thickness data measured by the sensor to obtain the actual icing thickness data.

[0094] The icing sensor temperature compensation device provided in this disclosure can execute the icing sensor temperature compensation method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of executing the method.

[0095] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program / instructions, which, when executed by a processor, implements the icing sensor temperature compensation method in the above embodiments.

[0096] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure.

[0097] The following is a detailed reference. Figure 4 The diagram illustrates a structural schematic suitable for implementing the electronic device 300 in the embodiments of this disclosure. The electronic device 300 in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0098] like Figure 4 As shown, the electronic device 300 may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a memory 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0099] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0100] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from memory 308, or installed from ROM 302. When the computer program is executed by processor 301, it performs the functions defined in the icing sensor temperature compensation method of embodiments of this disclosure.

[0101] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0102] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0103] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0104] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the aforementioned icing sensor temperature compensation method.

[0105] Electronic devices can be programmed with computer program code in one or more programming languages ​​or combinations thereof to perform the operations of this disclosure. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0106] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, or sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0107] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0108] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0109] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0110] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover technical solutions formed by any combination of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0111] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0112] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for temperature compensation of an icing sensor, characterized in that, include: Capacitance data at different temperatures and ice thicknesses are collected to obtain an initial input dataset. The initial input dataset is divided into a test set and a training set. A BP neural network structure is built based on the training set. The weights and thresholds of the BP neural network structure are iterated according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set. The iterative BP neural network structure is trained based on the training set, and the weights and thresholds after iteration are determined according to the training results to obtain the temperature compensation model. The temperature compensation model is used to predict the test set to obtain a predicted value sequence. The predicted value sequence is compared with the actual thickness value sequence in the test set. Based on the comparison result, at least one evaluation index of the temperature compensation model is calculated. At least one index is matched with the corresponding preset application threshold. If all evaluation indexes meet the corresponding preset application threshold, the temperature compensation model is marked as qualified. The temperature compensation model is then used to perform temperature compensation on the ice thickness data obtained by the icing sensor.

2. The method according to claim 1, characterized in that, Capacitance data at different temperatures and ice thicknesses are collected to obtain an initial input dataset. This initial input dataset is then divided into a test set and a training set. A backpropagation (BP) neural network structure is built based on the training set, including: Capacitance data of sensors under different temperatures and ice thicknesses are collected, and the capacitance data is normalized to obtain the initial input dataset. The initial input data is randomly divided to obtain a training set and a test set; The training set is used to perform feature analysis to determine the number of input layer nodes, and the number of output layer nodes is determined based on the prediction target. A BP neural network structure is built based on the number of input layer nodes, the number of output layer nodes, and the preset number of hidden layer nodes.

3. The method according to claim 1, characterized in that, The weights and thresholds of the BP neural network structure are iteratively evaluated based on a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set, including: The initial weights and thresholds of the BP neural network are set as the initial solution, and the Cubic mapping is introduced to generate the initial population. A hybrid optimization algorithm is determined using the Eagle algorithm and the simulated annealing algorithm, and the initial population is iterated according to the preset optimization objective and the hybrid optimization algorithm; In the first iteration phase, the Skyhawk algorithm is used to generate a random number for each initial solution in the initialization population. If the random number is less than a preset threshold, the first mode is adopted; if the random number is greater than or equal to the preset threshold, the second mode is adopted, and the iterative population is obtained, and the first iteration phase ends. The first mode involves performing a global search in the initial population using the Eagle algorithm to discover potential advantageous regions, update the initial solution positions, and obtain the first population. The second mode involves performing a spiral fine-grained search at the current position using the Eagle algorithm, updating the initial solution position, and obtaining the second group. The iterative population is the combination of the first population and the second population; In the second iteration stage, the best individual in the iterative population is selected as the initial solution. The optimal solution is determined by the simulated annealing algorithm and the preset optimization objective. Based on the optimal solution, the weights and thresholds after iteration are obtained, and the second iteration stage ends.

4. The method according to claim 3, characterized in that, In the second iteration phase, the optimal individual in the iterative population is selected as the initial solution. The simulated annealing algorithm and a preset optimization objective are used to determine the optimal solution. Based on the optimal solution, the weights and thresholds after iteration are obtained. The second iteration phase then ends, including: In the second iteration stage, the simulated annealing algorithm is used to determine the initial solution from the iterative population and calculate the function value of the initial solution. The initial temperature and perturbation operator are set, and the initial solution is perturbed according to the perturbation operator to obtain a new solution and calculate the function value of the new solution. The function values ​​of the initial solution and the new solution are input into the Metropolis criterion to calculate the acceptance probability of the new solution. If the function value of the new solution is less than the function value of the initial solution, the new solution must be accepted. If the function value of the new solution is greater than or equal to the function value of the initial solution, the new solution is accepted according to the acceptance probability of the new solution. The initial temperature is cooled down, and the process is repeated until the current temperature is lower than the preset termination temperature. The optimal solution is then output according to the preset optimization target, and the weights and thresholds after the iteration are obtained based on the optimal solution.

5. The method according to claim 2, characterized in that, The iteratively trained BP neural network structure is trained based on the training set. The weights and thresholds after the iteration are determined according to the training results to obtain the temperature compensation model, including: The iterated weights and thresholds are assigned to the BP neural network structure, the training set is input into the iterated BP neural network structure for training, and the predicted ice thickness value is output. The predicted ice thickness value is compared with the actual thickness in the training set, and the loss value of the preset loss function is calculated based on the comparison result. Based on the loss value, starting from the output layer of the BP neural network structure, the gradient of the loss with respect to the weights and thresholds after each iteration is calculated in reverse. The gradient and gradient algorithm are used to update and iterate all the weights and thresholds in the BP neural network structure until the preset training rounds are reached, and then the training process is terminated. Based on the training results at the end of training, the corresponding weights and thresholds are fixed, and the temperature compensation model is obtained based on the BP neural network structure after fixing the weights and thresholds.

6. The method according to claim 1, characterized in that, The temperature compensation model is used to predict the test set, resulting in a predicted value sequence. This predicted value sequence is compared with the actual thickness value sequence in the test set. Based on the comparison results, at least one evaluation index of the temperature compensation model is calculated. Each of these at least one index is matched with a corresponding preset application threshold. If all evaluation indices meet the corresponding preset application thresholds, the temperature compensation model is marked as qualified. Temperature compensation is then performed on the ice thickness data obtained from the icing sensor using the temperature compensation model, including: The test set is input into the temperature compensation model, and the temperature compensation model outputs the predicted value corresponding to each sample in the training set. A sequence of predicted values ​​is generated based on all the predicted values. The predicted value sequence is compared one by one with the actual thickness value sequence in the test set; Based on the comparison results, calculate multiple evaluation indicators for the temperature compensation model, and compare each evaluation indicator with its corresponding preset application threshold. If all evaluation indicators meet the preset application thresholds, the temperature compensation model is marked as qualified, and temperature compensation is performed on the ice thickness data measured by the icing sensor. If any evaluation metric fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and the retraining process of the temperature compensation model is triggered.

7. The method according to claim 6, characterized in that, Also includes: Obtain actual application requirements and determine preset application thresholds based on those requirements; If all evaluation indicators meet the preset application threshold, the temperature compensation model is marked as qualified and can be used in practical applications. If any evaluation indicator fails to meet the preset application threshold, the temperature compensation model is marked as unqualified, and a root cause analysis is performed on the temperature compensation model. The temperature compensation model is then retrained based on the results of the root cause analysis. Once it is determined that the temperature compensation model can be used in practical applications, the temperature compensation model is used to compensate the icing thickness data measured by the sensor to obtain the actual icing thickness data.

8. A temperature compensation device for an icing sensor, the device comprising: The acquisition module is used to collect capacitance data at different temperatures and ice thicknesses to obtain an initial input dataset. The initial input dataset is divided into a test set and a training set, and a BP neural network structure is built based on the training set. The iteration module is used to iterate the weights and thresholds of the BP neural network structure according to a preset optimization objective and a hybrid optimization algorithm. The preset optimization objective is to minimize the prediction error of the BP neural network structure on the training set. The determination module is used to train the iterative BP neural network structure based on the training set, and determine the weights and thresholds after iteration based on the training results to obtain the temperature compensation model. The comparison module is used to predict the test set using the temperature compensation model to obtain a sequence of predicted values, compare the sequence of predicted values ​​with the sequence of actual thickness values ​​in the test set, calculate the evaluation index of the temperature compensation model based on the comparison results, mark the temperature compensation model as qualified based on the evaluation index, and mark the temperature compensation model as qualified if all evaluation indexes meet the preset application threshold. The temperature compensation model is then used to perform temperature compensation on the ice thickness data obtained by the icing sensor.

9. An electronic device, characterized in that, include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and configured to be executed by the processor to implement the icing sensor temperature compensation method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, causes the processor to implement the icing sensor temperature compensation method provided in any one of claims 1-7.

Images