Carbon fiber composite core wire dynamic load prediction method based on recurrent neural network

Abstract

The invention discloses a carbon fiber composite core wire dynamic load prediction method based on a recurrent neural network, and the method comprises the steps: collecting the dynamic load data of a carbon fiber composite core wire, and carrying out the preprocessing, and obtaining a dynamic load data set; constructing a hybrid memory loop network model; dividing the dynamic load data set into a training set and a test set, and performing iterative training on a hybrid memory loop network model through the training set to obtain a trained hybrid memory loop network model; evaluating the trained hybrid memory loop network model, and calculating a prediction error of the trained hybrid memory loop network model; and collecting the to-be-tested dynamic load data of the to-be-tested carbon fiber composite core wire through the embedded system to obtain a prediction result. According to the method, the problems of data processing, model construction, training optimization and the like in dynamic load prediction of the carbon fiber composite core wire are solved, and efficient and accurate establishment and deployment of the dynamic load prediction model are realized.

Description

Technical Field

[0001] The invention relates to the technical field of carbon fiber composite core conductor performance, and in particular to a method for predicting dynamic load of a carbon fiber composite core conductor based on a recursive neural network. Background Art

[0002] Carbon fiber composite core conductor is a new type of material used in power transmission lines, which has the advantages of light weight, high strength and corrosion resistance. In practical applications, understanding the dynamic load of carbon fiber composite core conductors under different working conditions is crucial to the safety and stability of the line. Recurrent neural network is a deep learning model suitable for processing sequence data. It has the advantages of capturing timing information and long-term dependencies and is suitable for dynamic load prediction. However, a single long short-term memory network or gated recurrent unit has its limitations. The long short-term memory network model is relatively complex, with a large amount of calculation and a long training time. The prediction efficiency for some low-dimensional dynamic loads is relatively low; the gated recurrent unit model has a relatively simple structure, lacks memory units and output gates, and has a weak ability to cope with complex tasks. At the same time, due to the relatively simplified gating mechanism, it may cause information loss or confusion.

[0003] The Chinese patent with publication number CN113418629A discloses a method for measuring the inflection point temperature of a carbon fiber composite core conductor, which is used to measure the inflection point temperature of a carbon fiber composite core conductor. However, this method only focuses on the temperature change when the carbon fiber composite core conductor is working, and ignores the changes in other loads. Such a test method is relatively limited, and multiple important load data of the carbon fiber composite core conductor should be comprehensively considered when the carbon fiber composite core conductor is working to achieve a comprehensive evaluation and prediction of the carbon fiber composite core conductor. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for predicting the dynamic load of a carbon fiber composite core conductor based on a recursive neural network. The long short-term memory network and the gated recurrent unit in the recursive neural network are used to construct a hybrid memory recurrent network model. Combined with technical means such as data preprocessing, feature extraction, and data standardization, the problems of data processing, model construction, and training optimization in the dynamic load prediction of carbon fiber composite core conductors are solved, and the establishment and deployment of an efficient and accurate dynamic load prediction model are realized.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network, comprising:

[0007] Collecting dynamic load data of the carbon fiber composite core conductor, preprocessing the dynamic load data, and obtaining a dynamic load data set;

[0008] Based on long short-term memory network and gated recurrent unit, a hybrid memory recurrent network model is constructed;

[0009] The dynamic load data set is divided into a training set and a test set, the hybrid memory recurrent network model is iteratively trained through the training set, the optimizer, the loss function and the hyperparameters are configured until the iteration stop condition is reached, and the trained hybrid memory recurrent network model is obtained;

[0010] Evaluating the trained hybrid memory recurrent network model using a test set, and calculating a prediction error of the trained hybrid memory recurrent network model;

[0011] The dynamic load prediction model is deployed in an embedded system, and the dynamic load data of the carbon fiber composite core conductor to be tested is collected through the embedded system to obtain the dynamic load prediction result of the carbon fiber composite core conductor to be tested.

[0012] As a preferred solution of the method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network described in the present invention, wherein: the collecting dynamic load data of the carbon fiber composite core conductor includes obtaining the dynamic load data of the carbon fiber composite core conductor through a sensor, and the dynamic load data includes load size, working conditions and timestamp; preprocessing the dynamic load data, and the preprocessing includes data cleaning, feature extraction and data standardization processing.

[0013] As a preferred solution of the method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network described in the present invention, wherein: the preprocessing includes performing abnormal value processing and missing value processing on the dynamic load data to obtain first dynamic load data;

[0014] The outlier processing adopts the IQR method, and the missing value processing adopts the mean interpolation method;

[0015] Extracting time series features from the first dynamic load data to obtain second dynamic load data;

[0016] The calculation formula for extracting the time series features is:

[0017]

[0018] Among them, X t is the value of the first dynamic load data at time t, X t+k is the value of the first dynamic load data at time t+k, is the mean value of the first dynamic load data, N is the total length of the first dynamic load data, and k is the step length of the first dynamic load data;

[0019] performing data standardization processing on the second dynamic load data to obtain a dynamic load data set;

[0020] The data standardization process adopts the Z-score standardization method, and the calculation formula is:

[0021]

[0022] Among them, X norm is the standardized data, X is the original data, μ is the mean of the original data, and ω is the standard deviation of the original data.

[0023] As a preferred solution of the method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network described in the present invention, wherein: the hybrid memory recurrent network model includes an input layer, a hidden layer and an output layer; the hidden layer is formed by alternating connection of long short-term memory network and gated recurrent unit;

[0024] The input layer receives dynamic load data as the input of the model, and the number of neurons n in the input layer is:

[0025] n=(pe) 2 -m

[0026] Among them, p is the number of neurons in the hidden layer, e is the input constant of the hybrid memory recurrent network model, and m is the number of neurons in the output layer.

[0027] As a preferred solution of the method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network described in the present invention, wherein: the long short-term memory network includes:

[0028] Input Gate:

[0029] Forget gate: f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f )

[0030] Cell state update: c t =f t ⊙c t-1 +i t ⊙tanh(W xc x t +W hc h t-1 +b c )

[0031] Output gate: ot =σ(W xo x t +W ho h t-1 +W co c t +b o )

[0032] Hide status update:h t =o t ⊙tanh(c t )

[0033] Among them, i t is the output of the input gate of the LSTM network, f t is the output of the forget gate of the long short-term memory network, c t is the cell state update at the current time step, o t is the output of the output gate, σ is the sigmoid function, x t is the input at the current moment, h t is the hidden state of the current time step, h t-1 is the hidden state of the previous time step, c t-1 is the cell state update at the previous time step, W xi , W hi , W ci is the weight of the input gate, W xf , W hf , W cf is the weight of the forget gate, W xc , W hc is the weight of the cell state update, W xo , W ho , W co is the weight of the output gate, b i , b f , b c , b o They are the bias of input gate, forget gate, cell state update, and output gate, respectively. ⊙ is the Hadamard product.

[0034] The gated recurrent unit comprises:

[0035] Update gate: z t =σ(W xz x t +W hz h t-1 +b z )

[0036] Reset gate: r t =σ(W xr x t +W hr ht-1 +b r )

[0037] Update implicit state: h t =(1-z t )⊙h t-1 +z t ⊙tanh(W xh x t +W hh (r t ⊙h t-1 )+b h )

[0038] Among them, z t is the output of the update gate, r t is the output of the reset gate, W xz and W hz is the weight of the update gate, W xr and W hr is the weight of the reset gate, W xh , W hh is the weight for updating the hidden state, b z , b r and b h They are the update gate, reset gate and bias for updating the hidden state respectively.

[0039] As a preferred solution of the method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network described in the present invention, wherein: the long short-term memory network also includes that the activation function of the output layer is:

[0040]

[0041] Among them, s is the input of the output layer, and m is the number of neurons in the output layer.

[0042] As a preferred solution of the method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network described in the present invention, wherein: the iterative training includes dividing the dynamic load data set into a training set and a test set in a ratio of 7:3, iteratively training the hybrid memory recurrent network model through the training set, forward propagating and calculating the loss according to the input data, and updating the model parameters through back propagation;

[0043] The iteration stop condition is set to reach the preset maximum number of iterations or the loss function converges, and the optimizer, loss function and hyperparameters are configured. The optimizer is used to adjust the model parameters, the loss function is used to measure the difference between the model predicted value and the actual value, and the hyperparameters are used to affect the effect of model training.

[0044] As a preferred solution of the method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network described in the present invention, wherein: the iterative training also includes the calculation formula of the loss function is:

[0045]

[0046] Among them, MSE is the loss function value, g is the number of samples in the training set, and y is l is the actual value, is the model prediction value.

[0047] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, steps of a method for predicting dynamic load of a carbon fiber composite core conductor based on a recursive neural network are implemented.

[0048] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of a method for predicting dynamic load of a carbon fiber composite core conductor based on a recursive neural network are implemented.

[0049] The beneficial effects of the present invention are as follows: by adopting the long short-term memory network and the gated recurrent unit in the recurrent neural network to construct a hybrid memory recurrent network model, combined with technical means such as data preprocessing, feature extraction, and data standardization, the problems of data processing, model construction, and training optimization in the dynamic load prediction of carbon fiber composite core conductors are solved, and the establishment and deployment of an efficient and accurate dynamic load prediction model are realized;

[0050] The hidden layer of the hybrid memory recurrent network model is composed of alternating connections between long short-term memory networks and gated recurrent units, which combines two different memory units, thereby improving the model's learning and expression capabilities for time series data, which is beneficial to improving the accuracy and robustness of the dynamic load prediction model;

[0051] Feature extraction is performed through the time series feature extraction method, and the scattered first dynamic load data is aggregated and converted into second dynamic load data with time series characteristics. The time series relationship and characteristics between the data are extracted, which helps to capture the time evolution law and characteristic information of the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0053] Figure 1A schematic flow chart of a method for predicting dynamic load of a carbon fiber composite core conductor based on a recursive neural network is provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0057] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0058] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0059] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0060] Example 1

[0061] Reference Figure 1 , which is the first embodiment of the present invention, provides a method for predicting dynamic load of a carbon fiber composite core conductor based on a recursive neural network, comprising:

[0062] S1, collecting dynamic load data of the carbon fiber composite core conductor, preprocessing the dynamic load data, and obtaining a dynamic load data set;

[0063] S2, based on long short-term memory network and gated recurrent unit, build a hybrid memory recurrent network model;

[0064] S3, dividing the dynamic load data set into a training set and a test set, iteratively training the hybrid memory recurrent network model through the training set, configuring the optimizer, loss function and hyperparameters until the iteration stop condition is reached, and obtaining the trained hybrid memory recurrent network model;

[0065] S4, evaluating the trained hybrid memory recurrent network model through the test set, calculating the prediction error of the trained hybrid memory recurrent network model, and when the prediction error is not greater than the preset error threshold, completing the training and obtaining the dynamic load prediction model, otherwise returning to and executing step S3;

[0066] S5, deploying the dynamic load prediction model to an embedded system, collecting the dynamic load data of the carbon fiber composite core conductor to be tested through the embedded system, and obtaining a dynamic load prediction result of the carbon fiber composite core conductor to be tested.

[0067] The method for predicting the dynamic load of a carbon fiber composite core conductor based on a recursive neural network in this embodiment combines multiple links such as data processing, model construction, training evaluation, and deployment application. By effectively processing data and building an efficient neural network model, it achieves accurate prediction of the dynamic load of a carbon fiber composite core conductor, while improving the interpretability of the model prediction and having high practicality.

[0068] Step S1 includes the following steps:

[0069] S11, acquiring dynamic load data of the carbon fiber composite core conductor through a sensor, wherein the dynamic load data includes load size, working conditions and timestamp;

[0070] S12, preprocessing the dynamic load data, wherein the preprocessing includes data cleaning, feature extraction and data standardization.

[0071] It should be noted that:

[0072] Step S1 obtains the dynamic load data of the carbon fiber composite core conductor from the sensor, including the load size, working conditions and timestamp, and effectively pre-processes the data, including data cleaning, feature extraction and data standardization. These steps help improve data quality, accuracy and consistency, and provide a reliable data foundation for the subsequent construction of the dynamic load prediction model, thereby improving the accuracy and reliability of the prediction model.

[0073] Step S12 includes:

[0074] Performing outlier processing and missing value processing on the dynamic load data to obtain first dynamic load data;

[0075] The outlier processing adopts the IQR method, and the missing value processing adopts the mean interpolation method;

[0076] Extracting time series features from the first dynamic load data to obtain second dynamic load data;

[0077] The calculation formula for extracting the time series features is:

[0078]

[0079] Among them, X t is the value of the first dynamic load data at time t, X t+k is the value of the first dynamic load data at time t+k, is the mean value of the first dynamic load data, N is the total length of the first dynamic load data, and k is the step length of the first dynamic load data;

[0080] performing data standardization processing on the second dynamic load data to obtain a dynamic load data set;

[0081] The data standardization process adopts the Z-score standardization method, and the calculation formula is:

[0082]

[0083] Among them, X norm is the standardized data, X is the original data, μ is the mean of the original data, and ω is the standard deviation of the original data.

[0084] It should be noted that:

[0085] This embodiment uses the IQR method to process outliers and the mean interpolation method to process missing values, which can effectively clean and improve the dynamic load data and improve the accuracy and completeness of the data.

[0086] Feature extraction is performed through the time series feature extraction method, and the scattered first dynamic load data is aggregated and converted into second dynamic load data with time series characteristics. The time series relationship and characteristics between the data are extracted, which helps to capture the time evolution law and characteristic information of the data.

[0087] The Z-score standardization method is used to process the second dynamic load data so that the data has zero mean and unit variance, which is conducive to eliminating the dimensional differences between different features and improving the convergence speed and stability of model training.

[0088] The technical means such as outlier processing, missing value processing, time series feature extraction and data standardization processing in step S12 help to optimize the quality and characteristics of dynamic load data, provide a more reliable and effective data basis for subsequent model training and prediction, and thus improve the accuracy and reliability of the dynamic load prediction method.

[0089] Step S2 includes:

[0090] The hybrid memory recurrent network model includes an input layer, a hidden layer and an output layer;

[0091] The hidden layer is formed by alternatingly connecting a long short-term memory network and a gated recurrent unit.

[0092] The input layer receives dynamic load data as the input of the model, and the number of neurons n in the input layer is:

[0093] n=(pe) 2 -m

[0094] Among them, p is the number of neurons in the hidden layer, e is the input constant of the hybrid memory recurrent network model, and m is the number of neurons in the output layer.

[0095] It should be noted that:

[0096] The hybrid memory recurrent network model includes an input layer, a hidden layer and an output layer, wherein the beneficial effects of the input layer are:

[0097] Receiving dynamic load data: The input layer receives dynamic load data as the input of the model and passes the original data to the hidden layer for processing and learning, thus playing the role of data transmission and information input.

[0098] Determine the number of neurons: The number of neurons n in the input layer is determined through a given calculation formula. This number directly affects the input dimension of the model and the complexity of information transmission. The appropriate number of neurons helps to improve the expressiveness and learning effect of the model.

[0099] Control model input constant: The input constant e in the input layer is an important parameter of the hybrid memory recurrent network model. By adjusting this constant, the input data of the model can be adjusted and controlled, affecting the learning speed and convergence of the model.

[0100] The input layer in the hybrid memory recurrent network model plays a key role in receiving dynamic load data, determining the number of neurons and controlling input constants. It provides effective input information and data basis for the model, facilitates the model to learn and predict dynamic load data, and improves the accuracy and generalization ability of the model.

[0101] The long short-term memory network comprises:

[0102] Input Gate:

[0103] Forget gate: f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f )

[0104] Cell state update: c t =f t ⊙c t-1 +i t ⊙tanh(W xc x t +W hc h t-1 +b c )

[0105] Output gate: o t =σ(W xo x t +W ho h t-1 +W co c t +b o )

[0106] Hide status update:h t =o t ⊙tanh(c t )

[0107] Among them, i t is the output of the input gate of the LSTM network, f t is the output of the forget gate of the long short-term memory network, c t is the cell state update at the current time step, o t is the output of the output gate, σ is the sigmoid function, xt is the input at the current moment, h t is the hidden state of the current time step, h t-1 is the hidden state of the previous time step, c t-1 is the cell state update at the previous time step, W xi , W hi , W ci is the weight of the input gate, W xf , W hf , W cf is the weight of the forget gate, W xc , W hc is the weight of the cell state update, W xo , W ho , W co is the weight of the output gate, b i , b f , b c , b o They are the bias of input gate, forget gate, cell state update, and output gate, respectively. ⊙ is the Hadamard product.

[0108] The gated recurrent unit comprises:

[0109] Update gate: z t =σ(W xz x t +W hz h t-1 +b z )

[0110] Reset gate: r t =σ(W xr x t +W hr h t-1 +b r )

[0111] Update implicit state: h t =(1-z t )⊙h t-1 +z t ⊙tanh(W xh x t +W hh (r t ⊙h t-1 )+b h )

[0112] Among them, z t is the output of the update gate, r t is the output of the reset gate, W xz and W hz is the weight of the update gate, W xr and W hris the weight of the reset gate, W xh , W hh is the weight for updating the hidden state, b z , b r and b h They are the update gate, reset gate and bias for updating the hidden state respectively.

[0113] It should be noted that:

[0114] The hidden layer of the hybrid memory recurrent network model is formed by alternatingly connecting a long short-term memory network and a gated recurrent unit.

[0115] LSTM and GRU both work independently and take advantage of alternating connections, with the following beneficial effects:

[0116] The role of long short-term memory network:

[0117] Input gate and forget gate: control the input and forgetting of information, helping the model decide when to remember new information and discard old information.

[0118] Cell state update: By updating the cell state, LSTM networks can retain and update important information, which helps to capture long-term dependencies in time series data.

[0119] Output gate: controls the output of the cell state and determines the update and output of the hidden state.

[0120] The role of the gated recurrent unit:

[0121] Update gate and reset gate: The gated recurrent unit controls the updating and forgetting of information through the update gate and reset gate, which helps the model learn effective time series features.

[0122] Update hidden state: Based on the output of the update gate and reset gate, the hidden state is updated to help the model better capture important information and patterns in the data.

[0123] Advantages of alternate connection:

[0124] Through the alternating connection of long short-term memory networks and gated recurrent units, the hidden layer can better capture the long-term dependencies and short-term changes in time series data, improving the model's memory and generalization capabilities.

[0125] This alternating connection structure helps balance the advantages of long short-term memory networks and gated recurrent units, making the model more effective and accurate in processing dynamic load data.

[0126] In summary, the hidden layer is composed of alternating connections between long short-term memory networks and gated recurrent units, combining two different memory units, thereby improving the model's learning and expression capabilities for time series data, which is beneficial to improving the accuracy and robustness of the dynamic load prediction model.

[0127] Step S2 also includes:

[0128] The activation function of the output layer is:

[0129]

[0130] Among them, s is the input of the output layer, and m is the number of neurons in the output layer.

[0131] It should be noted that:

[0132] Non-linear mapping: The activation function σ(s) is a sigmoid function that can map the input s of the output layer to a range between 0 and m. It has the characteristics of non-linear mapping and helps the model learn complex non-linear relationships and patterns.

[0133] Output range control: The output range of the activation function σ(s) is limited to [0,m], which helps control the range of output values ​​and is suitable for tasks that require the output to be limited to a specific range.

[0134] Output smoothness: The activation function σ(s) is defined in the entire real number domain and has a smooth property, which helps reduce gradient mutations and facilitates stable training and convergence of the model.

[0135] Output interpretability: The output of the activation function σ(s) can be interpreted as a value between 0 and m, which helps to explain and understand the output results, especially for tasks that require the output value to have a specific meaning.

[0136] Applicability: The activation function σ(s) is suitable for tasks that require output values ​​to be within a specific range, such as counting tasks or situations where the output values ​​need to be restricted. It can effectively map the output of the model to the specified range.

[0137] The output layer uses an activation function It has technical effects such as nonlinear mapping, output range control, smoothness and output interpretability, which helps to improve the expressiveness of the model and the controllability of the output results, and is suitable for specific types of tasks and data.

[0138] Step S3 includes:

[0139] S31, dividing the dynamic load data set into a training set and a test set in a ratio of 7:3, iteratively training the hybrid memory recurrent network model through the training set, performing forward propagation calculation loss according to the input data, and updating the model parameters through back propagation;

[0140] S32, setting the iteration stop condition to reach a preset maximum number of iterations or the loss function converges, configuring the optimizer, loss function and hyperparameters, the optimizer is used to adjust the model parameters, the loss function is used to measure the difference between the model prediction value and the actual value, and the hyperparameters are used to affect the effect of model training.

[0141] It should be noted that:

[0142] Step S3 of this embodiment is used to complete the training of the hybrid memory recurrent network model, which has the following

[0143] Beneficial effects:

[0144] Dividing the dynamic load dataset into a training set and a test set in a ratio of 7:3 helps to evaluate the generalization ability of the model on unseen data.

[0145] By iteratively training the hybrid memory recurrent network model, performing forward propagation on the training set to calculate the loss, and updating the model parameters through backpropagation, it helps the model learn the characteristics and patterns of the data.

[0146] Setting a preset maximum number of iterations or loss function convergence as the iteration stop condition helps control the timing of model training and avoid overfitting or undertraining.

[0147] Selecting a suitable optimizer to adjust model parameters, such as Adam, SGD, etc., helps accelerate model convergence and optimize model performance.

[0148] Selecting an appropriate loss function to measure the difference between the model's predicted value and the actual value, such as mean square error, cross entropy, etc., helps guide the training direction of the model.

[0149] Setting hyperparameters, such as learning rate and batch size, is used to influence the effect of model training and helps adjust the learning speed and stability of the model.

[0150] Through iterative training and optimizer adjustment, the model can gradually optimize parameters and improve its ability to fit dynamic load data.

[0151] The data set partitioning and iteration stopping conditions help evaluate the generalization ability of the model on unknown data and improve the practicality and reliability of the model.

[0152] By configuring optimizers and hyperparameters, you can improve the efficiency and speed of model training and accelerate the model convergence process.

[0153] Step S3 improves the performance, generalization ability and training efficiency of the model by data set partitioning, model training, iteration stop conditions and configuration optimizer, loss function, hyperparameters, etc., so that it can be better applied to the prediction and analysis of dynamic load data.

[0154] Step S32 includes:

[0155] The calculation formula of the loss function is:

[0156]

[0157] Among them, MSE is the loss function value, g is the number of samples in the training set, yl is the actual value, is the model prediction value.

[0158] It should be noted that:

[0159] MSE is a commonly used evaluation metric that can quantify the difference between the model prediction results and the true values, helping to evaluate the accuracy and precision of the model.

[0160] Through MSE, we can understand the average error size of the model when predicting dynamic loads, which helps to identify the situations in which the model performs poorly and make improvements and optimizations.

[0161] Based on the evaluation results of MSE, the performance of the model can be adjusted and optimized, such as adjusting hyperparameters, improving the model structure, or increasing training data to improve the prediction accuracy of the model.

[0162] As an evaluation indicator, MSE can provide a basis for decision-making and help determine whether the model has achieved the expected prediction accuracy requirements, thereby deciding whether to continue using the model or make further improvements.

[0163] Evaluating the trained hybrid memory recurrent network model through MSE can effectively measure the prediction performance of the model and provide guidance and basis for further optimization and improvement of the model.

[0164] Step S4 includes:

[0165] The trained hybrid memory recurrent network model is evaluated by the test set, and the prediction error of the trained hybrid memory recurrent network model is calculated. When the prediction error is not greater than a preset error threshold, the training is completed to obtain a dynamic load prediction model, otherwise, the method returns to step S3;

[0166] It should be noted that:

[0167] Step S4 of this embodiment also includes:

[0168] Apply the trained hybrid memory recurrent network model to the test set to evaluate the performance of the model on unseen data;

[0169] Use the trained model to predict the data in the test set and get the prediction results;

[0170] Use the selected evaluation metric (such as mean square error MSE) to calculate the prediction error of the model on the test set;

[0171] comparing the calculated prediction error with a preset error threshold;

[0172] If the prediction error is not greater than the preset error threshold, it means that the model has achieved the expected prediction accuracy requirements, the training can be completed, and the dynamic load prediction model can be obtained.

[0173] If the prediction error exceeds the preset error threshold, it is necessary to return to step S3 to continue optimizing the model parameters or adjusting the model structure to improve the prediction accuracy of the model.

[0174] Based on the prediction error results, it is decided whether the training is complete or the model needs to be further optimized.

[0175] If the model needs to be further optimized, return to step S3 and repeatedly perform training, evaluation, and adjustment until the preset error requirements are met.

[0176] Through the above sub-steps of step S4, the performance of the trained hybrid memory recurrent network model on the test set can be effectively evaluated, and decisions can be made based on the results of the prediction error to achieve the expected dynamic load prediction accuracy.

[0177] Step S5 includes:

[0178] The dynamic load prediction model is deployed in an embedded system, and the dynamic load data of the carbon fiber composite core conductor to be tested is collected through the embedded system to obtain the dynamic load prediction result of the carbon fiber composite core conductor to be tested.

[0179] It should be noted that:

[0180] Step S5 of this embodiment also includes:

[0181] Deploy the trained dynamic load prediction model to the embedded system to ensure that the model can run in the embedded environment;

[0182] Use an embedded system to collect dynamic load data of the carbon fiber composite core conductor to be tested. This data can be sensor data collected in real time, which is used to input into the model deployed in the embedded system for prediction;

[0183] Preprocess the collected dynamic load data, including data cleaning, feature extraction and other steps, to meet the requirements of model input;

[0184] The dynamic load prediction model deployed in the embedded system is used to predict the dynamic load data of the carbon fiber composite core conductor to obtain the prediction result.

[0185] Analyze the model prediction results and evaluate the model's performance on actual data, including indicators such as prediction accuracy and precision;

[0186] Output the prediction results to the display device or storage device of the embedded system for user viewing or subsequent analysis;

[0187] In practical applications, a real-time monitoring and feedback mechanism can be set up to continuously collect dynamic load data and make predictions, and provide timely feedback on the prediction results to guide decision-making or adjustments.

[0188] Through the above sub-steps, the dynamic load prediction model can be successfully deployed into the embedded system, the dynamic load data of the carbon fiber composite core conductor to be tested can be collected in real time, and the corresponding prediction results can be obtained, providing effective prediction support for practical applications.

[0189] Example 2

[0190] The second embodiment of the present invention is different from the previous embodiment in that:

[0191] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0192] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0193] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0194] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0195] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0196] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network, characterized in that: include, Collecting dynamic load data of the carbon fiber composite core conductor, preprocessing the dynamic load data, and obtaining a dynamic load data set; Based on long short-term memory network and gated recurrent unit, a hybrid memory recurrent network model is constructed; The dynamic load data set is divided into a training set and a test set, the hybrid memory recurrent network model is iteratively trained through the training set, the optimizer, the loss function and the hyperparameters are configured until the iteration stop condition is reached, and the trained hybrid memory recurrent network model is obtained; Evaluating the trained hybrid memory recurrent network model using a test set, and calculating a prediction error of the trained hybrid memory recurrent network model; The dynamic load prediction model is deployed in an embedded system, and the dynamic load data of the carbon fiber composite core conductor to be tested is collected through the embedded system to obtain the dynamic load prediction result of the carbon fiber composite core conductor to be tested.

2. The method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network according to claim 1, characterized in that: The collecting of the dynamic load data of the carbon fiber composite core conductor comprises obtaining the dynamic load data of the carbon fiber composite core conductor through a sensor, wherein the dynamic load data comprises load size, working conditions and timestamp; The dynamic load data is preprocessed, and the preprocessing includes data cleaning, feature extraction and data standardization.

3. The method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network according to claim 2, characterized in that: The preprocessing includes performing outlier processing and missing value processing on the dynamic load data to obtain first dynamic load data; The outlier processing adopts the IQR method, and the missing value processing adopts the mean interpolation method; Extracting time series features from the first dynamic load data to obtain second dynamic load data; The calculation formula for extracting the time series features is: Among them, X t is the value of the first dynamic load data at time t, X t+k is the value of the first dynamic load data at time t+k, is the mean value of the first dynamic load data, N is the total length of the first dynamic load data, and k is the step length of the first dynamic load data; performing data standardization processing on the second dynamic load data to obtain a dynamic load data set; The data standardization process adopts the Z-score standardization method, and the calculation formula is: Among them, X norm is the standardized data, X is the original data, μ is the mean of the original data, and ω is the standard deviation of the original data.

4. The method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network according to claim 3, characterized in that: The hybrid memory recurrent network model includes an input layer, a hidden layer and an output layer; the hidden layer is formed by alternatingly connecting a long short-term memory network and a gated recurrent unit; The input layer receives dynamic load data as the input of the model, and the number of neurons n in the input layer is: n=(p-e) 2 -m Among them, p is the number of neurons in the hidden layer, e is the input constant of the hybrid memory recurrent network model, and m is the number of neurons in the output layer.

5. The method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network according to claim 4, characterized in that: The long short-term memory network includes: Input Gate: Forget gate: f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f ) Cell state update: c t =f t ⊙c t-1 +i t ⊙tanh(W xc x t +W hc h t-1 +b c ) Output gate: o t =σ(W xo x t +W ho h t-1 +W co c t +b o ) Hide status update:h t =o t ⊙tanh(c t ) Among them, i t is the output of the input gate of the LSTM network, f t is the output of the forget gate of the long short-term memory network, c t is the cell state update at the current time step, o t is the output of the output gate, σ is the sigmoid function, x t is the input at the current moment, h t is the hidden state of the current time step, h t-1 is the hidden state of the previous time step, c t-1 is the cell state update at the previous time step, W xi , W hi , W ci is the weight of the input gate, W xf , W hf , W cf is the weight of the forget gate, W xc , W hc is the weight of the cell state update, W xo , W ho , W co is the weight of the output gate, b i , b f , b c , b o They are the bias of input gate, forget gate, cell state update, and output gate, respectively. ⊙ is the Hadamard product. The gated recurrent unit comprises: Update gate: z t =σ(W xz x t +W hz h t-1 +b z ) Reset gate: r t =σ(W xr x t +W hr h t-1 +b r ) Update implicit state: h t =(1-z t )⊙h t-1 +z t ⊙tanh(W xh x t +W hh (r t ⊙h t-1 )+b h ) Among them, z t is the output of the update gate, r t is the output of the reset gate, W xz and W hz is the weight of the update gate, W xr and W hr is the weight of the reset gate, W xh , W hh is the weight for updating the hidden state, b z , b r and b h They are the update gate, reset gate and bias for updating the hidden state respectively.

6. The method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network according to claim 5, characterized in that: The long short-term memory network also includes an activation function of the output layer: Among them, s is the input of the output layer.

7. The method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network according to claim 6, characterized in that: The iterative training includes dividing the dynamic load data set into a training set and a test set in a ratio of 7:3, iteratively training the hybrid memory recurrent network model through the training set, performing forward propagation calculation of losses according to input data, and updating model parameters through back propagation; The iteration stop condition is set to reach the preset maximum number of iterations or the loss function converges, and the optimizer, loss function and hyperparameters are configured. The optimizer is used to adjust the model parameters, the loss function is used to measure the difference between the model predicted value and the actual value, and the hyperparameters are used to affect the effect of model training.

8. The method for predicting dynamic load of carbon fiber composite core conductor based on recursive neural network according to claim 7, characterized in that: The iterative training also includes that the loss function is calculated as follows: Among them, MSE is the loss function value, g is the number of samples in the training set, and y l is the actual value, is the model prediction value.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images