Quasi-Newton optimization elastic network-based optical cable fusion splice box environment detection method
The Newton-optimized elastic network model addresses environmental control issues in optical fiber splicing by predicting and mitigating adverse conditions, enhancing splicing efficiency and success rates.
Patent Information
- Application Number
- CN202510294635.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-13
- Publication Date
- 2025-07-15
AI Technical Summary
In the fiber optic cable head fusion operation, there is a lack of effective environmental guarantee and temperature control optimization solutions, which leads to low welding efficiency, high cost and high risk in outdoor low temperature environments in winter, making it difficult to adapt to temperature fluctuations and severe weather changes in complex environments, resulting in large welding losses and high failure rates.
The quasi-Newtonian optimized elastic network regression model is adopted to detect the temperature, humidity and dust concentration of the optical cable fusion assembly box in real time, and an elastic network regression model is constructed to determine whether the welding conditions are normal, warning for adverse factors in advance, and optimizing the welding environment.
It improves the success rate of fiber optic cable fusion, reduces fusion loss, ensures efficient and stable fusion work, and reduces costs and risks.
Smart Images
Figure CN120316732A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of operation and maintenance of optical fiber communication facilities, and in particular to a method for detecting an environment of an optical cable fusion box of a quasi-Newton optimized elastic network. Background Art
[0002] In the optical cable head welding operation, the lack of effective environmental protection and temperature control optimization solutions is a common problem. At present, the welding work in the low temperature environment outdoors in winter mainly relies on setting up tents and using generators for heating. This method mostly relies on the experience of the operators. It is not only cumbersome and inefficient, but also expands the scope of work and greatly increases costs. Moreover, many outdoor welding scenes basically adopt a fixed way to ensure the welding environment. This single response method is difficult to adapt to various dynamic situations such as large temperature fluctuations and severe weather changes during operation, which often leads to problems such as excessive welding loss and ineffective welding, which greatly reduces the welding efficiency and even increases the difficulty and risk of the operation. Therefore, it is necessary to develop an optical cable head welding guarantee optimization solution that can adapt to complex environments, accurately control welding conditions, and has cost control, so as to improve the quality and efficiency of optical cable head welding work. Summary of the invention
[0003] The present invention provides an optical cable fusion box environment detection method of a quasi-Newton optimized elastic network, which can judge the environmental status of the optical cable fusion component box and whether the fusion conditions are normal, so as to reduce the fusion loss and improve the fusion success rate. It can also warn of adverse factors that affect the quality of optical cable fusion in advance, and ensure the efficient and stable performance of optical cable fusion work.
[0004] In order to achieve the above object, the present invention adopts the following technical solutions:
[0005] A method for detecting an optical cable splicing box environment using a quasi-Newton optimized elastic network comprises the following steps:
[0006] S1, preprocessing the collected component box data;
[0007] S2, calculate the objective function;
[0008]
[0009] Among them, β is the regression coefficient vector, β i is the i-th component of β, λ1 and λ2 are hyperparameters, n is the number of samples, p is the number of features, y is the output variable vector, X is the input feature matrix, is the mean square error term that measures the difference between the predicted value and the true value. is the L1 regularization term, is the L2 regularization term;
[0010] Calculate the gradient of the objective function J with respect to the regression coefficient β In the k-th iteration, calculate the gradient at the current point β (k) Here
[0011] S3. Quasi-Newton iterative update; update the regression coefficient β, hyperparameters λ1, hyperparameter λ2, and the inverse matrix B of the approximate Hessian matrix in the quasi-Newton algorithm until the condition that the variable value ||x k+1 -x k || is less than the pre-set threshold ∈ or reaches the condition of the predetermined maximum number of iterations, and stop the iteration; otherwise, return to step S2 to continue the iteration; by continuously repeating the iteration process, the quasi-Newton optimization algorithm uses the gradient information of the objective function and the information of the previous iteration to update the search direction and the approximate matrix to find the optimal solution;
[0012] S4. Use the optimized hyperparameters λ1 and λ2 and β to construct an elastic net regression model, thereby training the optimal optical cable splicing component box environment judgment model, import the real-time collected component box data into the model, judge the optical cable splicing component box environment status, and give early warnings of adverse factors affecting the optical cable splicing quality.
[0013] Further, the component box data includes three types of data: the internal temperature, humidity, and dust concentration of the optical cable splicing box.
[0014] Further, the preprocessing of the collected component box data includes the following steps:
[0015] S1.1. Clean the data, handle missing values and outliers, and then perform standardization processing, and divide the data into a training set, a validation set, and a test set;
[0016] S1.2. Initialize the coefficient β of the elastic net regression model, set it to a zero vector or a random vector following a uniform distribution. For the inverse matrix B of the approximate Hessian matrix in the quasi-Newton algorithm, initialize it to the identity matrix I, set the initial values of the hyperparameters λ1 and λ2, and at the same time set the initial value of the step size α.
[0017] Further, the update of the regression coefficient β is based on the iterative formula of the quasi-Newton method:
[0018]
[0019] where k is the number of iterations, β (k+1) is the regression coefficient of the (k + 1)-th iteration, is the partial derivative of the objective function with respect to β, β (k) is the regression coefficient of the k-th iteration, is the hyperparameter value of the k-th iteration.
[0020] Further, update the hyperparameters λ1 and λ2:
[0021]
[0022] Further, update the inverse matrix B of the approximate Hessian matrix in the quasi-Newton algorithm;
[0023] Calculate the difference vector δ between two adjacent iteration points k and the difference vector y of the objective function gradients at two adjacent iteration points k ,
[0024] δ k = β (k+1) -β (k) (5)
[0025]
[0026] Update B according to the quasi-Newton method update formula Thereby update B;
[0027] where T is the matrix transpose symbol, B (k+1) is the updated inverse matrix of the approximate Hessian matrix, B (k) is the inverse matrix of the approximate Hessian matrix at the k-th iteration, is a positive definite correction term to enhance the matrix's response to gradient changes, is a negative definite correction term to cancel the components in the matrix that are inconsistent with the direction of δ k direction.
[0028] Compared with the prior art, the beneficial effects of the present invention are:
[0029] It solves the problems of fusion loss, efficiency, cost, and safety hazards in the current fusion splicing of optical cable heads due to low outdoor temperature, complex environment, and lack of effective countermeasures in winter. By constructing a quasi-Newton optimized elastic net regression model, whether the temperature is suitable, whether the humidity exceeds the standard, and whether the dust is excessive are used as data to judge the environmental conditions of the optical cable fusion splicing component box, and whether the fusion splicing conditions are normal, so as to reduce the fusion loss, improve the fusion splicing success rate, and be able to early warn of adverse factors affecting the quality of optical cable fusion splicing, ensuring the efficient and stable progress of optical cable fusion splicing work. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 is the flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0031] The following further describes the specific embodiments of the present invention with reference to the drawings:
[0032] A method for detecting the environment of an optical cable splicing box based on quasi - Newton optimized elastic net. First, pre - process the data collected from the component box, including three types of data: temperature, humidity, and dust concentration inside the component box. Then, construct a quasi - Newton optimized elastic net regression model and import the real - time collected data into the model to judge the environmental conditions of the optical cable splicing component box, such as whether the temperature is appropriate, whether the humidity exceeds the standard, whether the dust is excessive, etc., so as to give early warnings of adverse factors affecting the quality of optical cable splicing and ensure the efficient and stable progress of optical cable splicing work.
[0033] The quasi - Newton optimization algorithm BFGS is an algorithm for solving unconstrained optimization problems, especially suitable for large - scale non - linear optimization problems. It gradually approaches the extreme point of the function through iteration. In each iteration, the quasi - Newton method uses the gradient information of the function and the previous iteration information to update the search direction, thus avoiding the complexity of directly calculating the second - derivative Hessian matrix.
[0034] 1. The specific steps of the quasi - Newton optimization algorithm are as follows:
[0035] (1) Initialization
[0036] Select a randomly generated initial point for the variable to be optimized, and initialize the inverse matrix B of the approximate Hessian matrix as the identity matrix to assist in guiding the search direction and determine the initial value of the step size to control the magnitude of variable update in each iteration. (k)
[0037] (2) Gradient calculation
[0038] In the k - th iteration, calculate the gradient at the current point β (k) Preliminarily determine the general direction of variable update according to the opposite direction of the gradient.
[0039] (3) Search direction calculation
[0040] According to the current approximate matrix B (k) and the gradient calculate the search direction Determine the step size α k through line search, such that the objective function J(β (k) +α k d k ) has a sufficient decrease.
[0041] (4) Iterative point update
[0042] Update the variable β according to the iterative formula of the quasi - Newton method and update the current β (k+1) 、β (k) 、 and αk Substitute into the formula to calculate the next iteration point β (k+1) ;
[0043] (5) Approximate matrix update
[0044] Calculate the variable difference and the corresponding gradient difference between this iteration and the previous iteration. These two differences contain the change information during the iteration process. According to the BFGS update formula, use the calculated difference information to update the inverse matrix B of the approximate Hessian matrix (k) , so that this approximate matrix can better approximate the inverse matrix of the true Hessian matrix during the iteration process, thereby more accurately guiding the iterative optimization;
[0045] The update formula of the BFGS algorithm is:
[0046]
[0047] where δ k = β (k+1) - β (k) ,
[0048] (6) Update iteration information
[0049] Update the approximate matrix according to the new variable values, and use the newly obtained variable values, previous variable values, gradients and other information to further improve the approximate matrix, continuously optimize the accuracy of the search direction, and prepare for the next iteration;
[0050] (7) Iteration termination condition
[0051] Meet the condition that the variable values ||β (k+1) - β (k) || between two adjacent iterations are less than the pre-set threshold ∈ or meet the condition of reaching the predetermined maximum number of iterations.
[0052] 2. Elastic net regression: Elastic net regression is an extended model of linear regression. It combines L1 (Lasso regression) and L2 (ridge regression) regularization. Its goal is to prevent overfitting and perform feature selection while fitting the data;
[0053] The general form of the model is:
[0054] where is the predicted value, β0 is the intercept, β i is the coefficient, x i is the independent variable, and p is the number of independent variables.
[0055] See Figure 1, is the method flow chart of the present invention. The specific steps of optimizing elastic net regression using the quasi-Newton optimization algorithm are as follows:
[0056] S1. Preprocess the data of the component box collected;
[0057] S1.1. Data preparation: Use the three types of data of the internal temperature, humidity and dust of the optical cable fusion box collected as the input vector for training, and use whether the fusion condition is normal as the state label;
[0058] S1.2. Data preprocessing:
[0059] (1) Clean the data and handle missing values and outliers;
[0060] (2) Standardize the temperature, humidity and dust data to meet the requirements of model training. Since the numerical ranges of different environmental factors are different, in order to make different features have the same weight influence in model training, standardize the data. The formula is where x is the original data, μ is the mean of X, and σ is the standard deviation of X;
[0061] (3) Dataset division: Divide the dataset into a training set, a validation set and a test set, and divide it according to the ratio of 70% training set, 15% validation set and 15% test set;
[0062] S1.3. Initialize parameters
[0063] Initialize the coefficient β of the elastic net regression model, set it as a vector of all zeros or a random vector subject to a uniform distribution. For the inverse matrix B of the approximate Hessian matrix in the quasi-Newton algorithm, initialize it as the identity matrix I, set the initial values of the hyperparameters λ1 and λ2, and at the same time set the initial value of the step size α.
[0064] S2. Calculate the objective function and the gradient;
[0065] (1) The objective function of the elastic net regression model is:
[0066]
[0067] where β is the regression coefficient vector, β i is the i-th component of β, λ1 and λ2 are hyperparameters, n is the number of samples, p is the number of features, y is the output variable vector, X is the input feature matrix, is the mean square error term measuring the difference between the predicted value and the true value, is the L1 regularization term, is the L2 regularization term;
[0068] (2) Calculate the gradient
[0069] Calculate the gradient of the objective function J with respect to the regression coefficient β In the k-th iteration, calculate the current point β (k) and the gradient at this point
[0070] S3. Quasi-Newton iteration update
[0071] S3.1 Update the regression coefficient β;
[0072] According to the iteration formula of the quasi-Newton method
[0073] where k is the iteration number, β (k+1) is the regression coefficient for the (k + 1)-th iteration, is the partial derivative of the objective function with respect to β, β (k) is the regression coefficient for the k-th iteration, is the hyperparameter value for the k-th iteration;
[0074] S3.2 Update λ1 and λ2;
[0075]
[0076] S3.3 Update B;
[0077] Calculate the difference vector δ between two adjacent iteration points k and the difference vector y of the objective function gradients at two adjacent iteration points k ;
[0078] δ k = β (k+1) - β (k) (13)
[0079]
[0080] Update B according to the quasi-Newton method update formula thus updating B;
[0081] where T is the matrix transpose symbol, B (k+1) is the updated inverse matrix of the approximate Hessian matrix, B (k) is the inverse matrix of the approximate Hessian matrix for the k-th iteration, is the positive definite correction term, enhancing the matrix's response to gradient changes, is the negative definite correction term, canceling out the components in the matrix that are inconsistent with the direction of δ k ;
[0082] Reach the variable values of two adjacent iterations ||β (k+1) - β (k)If the condition of being less than the pre-set threshold ∈ or the condition of reaching the predetermined maximum number of iterations is met, stop the iteration; otherwise, return to step 4 to continue the iteration. By continuously repeating the iteration process, the quasi-Newton optimization algorithm gradually approaches the optimal solution of the objective function. In each iteration, the gradient information of the objective function and the information of the previous iteration are used to update the search direction and the approximate matrix, enabling the algorithm to more effectively find the optimal solution.
[0083] S4. Use the optimized hyperparameters λ1 and λ2 and β to construct an elastic net regression model, thereby training the optimal optical cable splicing component box environment judgment model. Input the data of the internal temperature, humidity, and dust concentration of the optical cable splicing box into the model, and judge the environment status of the optical cable splicing component box and whether the splicing conditions are normal according to the model output, so as to reduce the splicing loss and improve the splicing success rate.
[0084] The above embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation manners and specific operation processes are given, but the protection scope of the present invention is not limited to the above embodiments. The methods used in the above embodiments are all conventional methods unless otherwise specified.
Claims
1. A method for detecting the environment of an optical cable splicing box by quasi - Newton optimized elastic net, characterized in that, It includes the following steps: S1. Preprocess the collected data of the component box; S2. Calculate the objective function; where β is the regression coefficient vector, and β i is the i-th component of β, λ1 and λ2 are hyperparameters, n is the number of samples, p is the number of features, y is the output variable vector, X is the input feature matrix, is the mean squared error term that measures the difference between the predicted value and the true value, is the L1 regularization term, is the L2 regularization term; Calculate the gradient of the objective function J with respect to the regression coefficient β In the k-th iteration, calculate the gradient at the current point β( k ) S3. Quasi - Newton iteration update: Update the regression coefficient β, hyperparameters λ1, hyperparameter λ2 and the inverse matrix B of the approximate Hessian matrix in the quasi - Newton algorithm until the condition that the variable value ∥x k+1 -x k ∥ is less than a pre - set threshold ∈ or the condition of reaching a predetermined maximum number of iterations is met, and stop the iteration; otherwise, return to step S2 to continue the iteration; By continuously repeating the iteration process, the quasi - Newton optimization algorithm uses the gradient information of the objective function and the information of the previous iteration to update the search direction and the approximate matrix to find the optimal solution; S4. Use the optimized hyperparameters λ1 and λ2 and β to construct an elastic net regression model, thereby training an optimal optical cable splicing component box environment judgment model, import the real-time collected component box data into the model, judge the environment condition of the optical cable splicing component box, and give early warnings of adverse factors affecting the quality of optical cable splicing.
2. The environmental detection method for an optical cable fusion splicing box using quasi-Newton optimized elastic network according to claim 1, characterized in that The component box data includes three types of data: the internal temperature, humidity, and dust concentration of the optical cable splicing box.
3. A method for detecting the environment of an optical cable splicing box with quasi - Newton optimized elastic network according to claim 1, characterized in that The preprocessing of the collected component box data includes the following steps: S1.
1. Clean the data, handle missing values and outliers, and then perform standardization processing. Divide the data into a training set, a validation set, and a test set; S1.
2. Initialize the coefficient β of the elastic net regression model, set it as a zero vector or a random vector obeying a uniform distribution. For the inverse matrix B of the approximate Hessian matrix in the quasi-Newton algorithm, initialize it as the identity matrix I, set the initial values of the hyperparameters λ1 and λ2, and at the same time set the initial value of the step size α.
4. A method for detecting the environment of an optical cable splicing box with quasi-Newton optimized elastic network according to claim 3, characterized in that, The update of the regression coefficient β is based on the iteration formula of the quasi-Newton method: where k is the number of iterations, and β (k+1) is the regression coefficient for the (k + 1)-th iteration, is the partial derivative of the objective function with respect to β, and β (k) is the regression coefficient for the k-th iteration, is the hyperparameter value for the k-th iteration.
5. The environmental detection method for an optical cable fusion splicing box with quasi-Newton optimized elastic network according to claim 3, characterized in that, The update of the hyperparameter λ1 and the hyperparameter λ2:
6. A method for detecting the environment of an optical cable splicing box with quasi-Newton optimized elastic network according to claim 3, characterized in that, The update of the inverse matrix B of the approximate Hessian matrix in the quasi-Newton algorithm; Calculate the difference vector δ between two adjacent iteration points k and the difference vector y of the objective function gradients at two adjacent iteration points k , δ k = β (k+1) - β (k) (5) Update the formula according to the quasi-Newton method Thereby update B; where T is the matrix transpose symbol, B (k+1) is the updated approximate inverse Hessian matrix, and B (k) is the approximate inverse Hessian matrix at the k-th iteration, is the positive definite correction term to enhance the matrix's response to gradient changes, is the negative definite correction term to cancel out the components in the matrix that are inconsistent with the k δ direction.