A processing method for elevator cable
By establishing a prediction model and incremental learning method based on historical data, the processing parameters of elevator cables are optimized, and the problem of relying on experience in parameter adjustment in traditional methods is solved, and more efficient and scientific cable production is achieved, reducing production costs and scrap products Rate.
Patent Information
- Application Number
- CN202410229878.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-29
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-02-29
AI Technical Summary
In the cable production process, traditional parameter adjustments rely on the operator's experience and trial and error, resulting in delayed identification of quality problems, increased scrap rate and production costs, and it is difficult to achieve consistency in large-scale production.
By collecting historical production parameter data and usage data of elevator cables, a comprehensive prediction model is established, a linear regression model is used to adjust the parameter prediction and revision, and the model is continuously updated through incremental learning methods to optimize the processing plan.
It improves the scientificity and accuracy of decision-making, reduces trial and error costs, optimizes the production plan of elevator cables, improves production efficiency and quality, and reduces production costs.
Smart Images

Figure CN118095081B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of cables, and in particular relates to a processing method for elevator cables. Background Art
[0002] Elevator cables usually consist of two parts: the main core and the control core:
[0003] The main core is mainly used for power transmission in the elevator system, responsible for connecting the power supply in the elevator room or the bottom of the elevator shaft to various parts of the elevator, such as motors, lighting, etc. The main core is usually composed of multiple conductors, usually multiple copper wires, to provide sufficient conductivity and flexibility; the conductors are covered with an insulating layer to prevent short circuits between conductors and provide electrical insulation. The insulating material is usually a wear-resistant, oil-resistant, and corrosion-resistant material to adapt to the elevator environment; the entire outside of the main core is covered with an insulating layer to provide overall insulation and protection, and is produced using a rubber extruder.
[0004] The control core is mainly used to transmit control signals in the elevator system. It is responsible for transmitting various control commands and information to ensure the normal operation, parking and safety performance of the elevator. The conductor of the control core is also composed of multiple strands of copper wire, but compared with the main core, the number of conductors of the control core is usually less. Multiple control cores are twisted around the central steel wire reinforcement core to form a control core unit, which is produced by a tube stranding machine; the outside of the conductor is also covered with an insulating layer to prevent interference and short circuit between conductors and provide electrical insulation; since the control core transmits control signals, in order to prevent the influence of external electromagnetic interference on the signal, the control core is usually equipped with a shielding layer, which can be aluminum foil or copper mesh fabric, and the copper wire is woven by a braiding machine; the entire outside of the control core also has an insulating layer to provide overall insulation and protection, which is produced by a rubber extruder.
[0005] During the cable production process, according to actual use requirements and product upgrade requirements, the processing plan may be appropriately adjusted to obtain better product performance. This adjustment usually involves one or several key parameters, which may include:
[0006] Material selection: Depending on the specific working environment and usage requirements, the materials used may need to be adjusted. For example, the material of the insulation layer can be selected according to the scenario in which the cable will be used to ensure its wear resistance, oil resistance, corrosion resistance and other properties.
[0007] Conductor structure: Adjust the conductor material, diameter or twisting method (reverse or same direction) to optimize conductivity, flexibility and mechanical strength.
[0008] Shielding layer and braided layer: According to the actual situation of electromagnetic interference, it may be necessary to adjust the material and density of the shielding layer, as well as the structure of the copper wire braiding to improve the anti-interference performance of the cable.
[0009] Extrusion process parameters: Adjust the operating parameters of the rubber extruder, such as temperature, pressure, etc., to ensure the quality of the insulation layer and the overall insulation layer.
[0010] Tubular stranding machine parameters: Depending on the requirements of the control core unit, it may be necessary to adjust the parameters of the tubular stranding machine to ensure compact arrangement of the cores and correct stranding of the reinforcing cores.
[0011] In the actual production process, the adjustment of the above key parameters often requires a comprehensive evaluation after the product processing is completed, which may result in the problem existing within a certain time frame in the production process but not being discovered. This time lag may lead to delayed identification of quality problems, increase scrap rate and production costs; traditional parameter adjustment often relies on the operator's experience and trial and error, which may cause the adjustment results to be affected by subjective factors and is not easy to achieve consistency in large-scale production. Summary of the invention
[0012] The present invention provides a processing method for an elevator cable, which can effectively solve the problems in the background technology.
[0013] In order to achieve the above object, the technical solution adopted by the present invention is:
[0014] A method for processing an elevator cable, comprising:
[0015] Collect historical production parameter data and historical usage data of elevator cables and perform preprocessing;
[0016] Establish a comprehensive prediction model through pre-processed historical production parameter data and historical usage data;
[0017] Predicting the usage of the elevator cable processing solution after parameter adjustment according to the comprehensive prediction model;
[0018] Revise the elevator cable processing plan after parameter adjustment according to the prediction results, and execute the processing according to the revised processing plan;
[0019] The comprehensive prediction model is a linear regression model, which is continuously updated through an incremental learning method during use. The incremental learning model uses gradient descent of online learning.
[0020] Furthermore, the preprocessing of the historical production parameter data and the historical usage data includes:
[0021] S1: Determine whether there are missing values in the historical production parameter data set and the historical usage data set. If so, fill in the missing values. Otherwise, execute S2.
[0022] S2: Determine whether there are outliers in the two data sets after the missing value processing. If there are outliers, process them by truncation or tail-shrinking substitution. Otherwise, execute S3.
[0023] S3: Convert all types of data into numerical data and perform feature standardization and normalization;
[0024] S4: Perform feature engineering based on domain knowledge and data characteristics.
[0025] Furthermore, when filling missing values, one of the mean, median, and mode is used.
[0026] Furthermore, the comprehensive prediction model is established, including:
[0027] For the linear regression model, initialize the intercept and weight;
[0028] Divide the data set into training set and test set, and divide the data batches in the training set;
[0029] For each data batch, extract a small batch of data. For each small batch of data, calculate the gradient of the loss function with respect to the model parameters, and use the calculated gradient to update the intercept and weight of the model through the gradient descent algorithm;
[0030] For the trained linear regression model, a performance evaluation is performed on the test set, and the model parameters are tuned according to the evaluation results.
[0031] Furthermore, the calculated gradient is used to update the intercept and weight of the model through the gradient descent algorithm, including:
[0032] B1: Set the initial intercept to b0 and the initial weights to b1, b2, ..., b n ;
[0033] B2: Setting the loss function L (i) for
[0034] B3: Calculate the gradient of the loss function with respect to the intercept and weight using the following formula:
[0035]
[0036] B4: Use the gradient descent algorithm to update the intercept and weight of the model using the following formula:
[0037]
[0038] in,
[0039] is the predicted value of the model for the i-th sample;
[0040] Y (i) is the actual target value;
[0041] is the characteristic value of n data in the i-th sample;
[0042] α is the learning rate, which is used to control the update step size;
[0043] Repeat steps B3 and B4 until convergence.
[0044] Furthermore, the learning rate α is gradually reduced as steps B3 and B4 are repeated.
[0045] Furthermore, the linear regression model is continuously updated during use through an incremental learning method, including:
[0046] Continuously collect new batch production parameters and new batch usage data for new batches of products, and perform preprocessing;
[0047] The linear regression model is updated through an incremental learning method using the preprocessed new batch production parameters and new batch usage data.
[0048] Furthermore, it also includes:
[0049] Setting a number of evaluation indicators, and normalizing each of the evaluation indicators to obtain a normalized indicator value;
[0050] For the processing schemes before and after the revision, score the evaluation indicators respectively, and determine the weight of each score;
[0051] For each solution, the comprehensive score is calculated using the normalized indicator value and the determined weights of each score;
[0052] The comprehensive scores of the processing plans before and after the revision are compared. When the comprehensive score of the revised processing plan is greater than the comprehensive score of the processing plan before the revision, the processing process is executed according to the revised processing plan, otherwise the revision is invalid and the original processing plan is maintained.
[0053] Furthermore, the parameters before the model revision are retained, and when the comprehensive score of the revised processing scheme is less than the comprehensive score of the processing scheme before the revision, the model parameters are rolled back.
[0054] Furthermore, the original parameters of the model are retained, and when the comprehensive score of the revised processing scheme is less than the comprehensive score of the processing scheme before the revision, the revised parameters are weighted averaged with the original parameters.
[0055] The technical solution of the present invention can achieve the following technical effects:
[0056] In the present invention, by collecting and utilizing the historical production parameter data and historical usage data of elevator cables, a data-driven comprehensive prediction model is established, which makes the formulation and adjustment of processing plans more based on actual conditions and experience, and improves the scientificity and accuracy of decision-making; a simple and effective linear regression model is selected, which can better capture the linear relationship between variables, and its simple mathematical form also helps to reduce modeling and calculation costs; the incremental learning method is adopted, which not only improves the real-time performance of the model, but also reduces the demand for computing and storage resources to a certain extent. By revising the processing plan of the elevator cable after parameter adjustment according to the prediction results of the comprehensive prediction model, the system has the ability of fine adjustment and can meet different needs more flexibly, which helps to reduce the trial and error cost in the adjustment process; revision based on the results of the prediction model helps to optimize the production plan of the elevator cable, improve production efficiency and quality, and thus reduce production costs in the long run. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0058] Figure 1 The present invention is a flow chart of a processing method for elevator cables;
[0059] Figure 2 A preprocessing flow chart for historical production parameter data and historical usage data;
[0060] Figure 3 Flowchart for building a comprehensive prediction model;
[0061] Figure 4 The invention provides an optimization step for the processing method of elevator cables. DETAILED DESCRIPTION
[0062] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0064] like Figure 1 As shown, a method for processing an elevator cable comprises:
[0065] Collect historical production parameter data and historical usage data of elevator cables and perform preprocessing;
[0066] Establish a comprehensive prediction model through pre-processed historical production parameter data and historical usage data;
[0067] Predict the usage of elevator cable processing solutions with parameter adjustments based on the comprehensive prediction model;
[0068] Revise the elevator cable processing plan after parameter adjustment according to the prediction results, and execute the processing according to the revised processing plan;
[0069] The comprehensive prediction model is a linear regression model, which is continuously updated through the incremental learning method during use. The incremental learning model adopts the gradient descent of online learning.
[0070] In the embodiment, detailed parameter data of each elevator cable production batch needs to be collected, including production parameters of conductor, insulation layer, control core, copper wire braiding and outer sheath, so as to ensure that the collected data contains sufficient information, such as wire diameter, conductor material, twisting method, insulation material, insulation thickness, number of cores, braiding machine speed, braiding density, outer sheath material, etc. For historical usage data, including installation date, usage environment, known quality problems, service life, maintenance records and possible failure conditions, etc.
[0071] During the implementation process, a data-driven comprehensive prediction model was established by collecting and utilizing the historical production parameter data and historical usage data of elevator cables, which makes the formulation and adjustment of processing plans more based on actual conditions and experience, and improves the scientificity and accuracy of decision-making; a simple and effective linear regression model is selected, which has intuitive interpretability and can better capture the linear relationship between variables. Its simple mathematical form also helps to reduce modeling and computing costs. Since the elevator cable processing plan needs to be adjusted in real time, the training and prediction of the model need to be completed in a short time. Based on the above goals, the incremental learning method is adopted in the present invention, which not only improves the real-time performance of the model, but also reduces the demand for computing and storage resources to a certain extent.
[0072] By revising the elevator cable processing plan after parameter adjustment according to the prediction results of the comprehensive prediction model, the system has the ability to make fine adjustments and can more flexibly meet different needs, which helps to reduce the trial and error costs in the adjustment process; revisions based on the results of the prediction model help optimize the production plan of elevator cables, improve production efficiency and quality, and thus reduce production costs in the long run.
[0073] Take the service life of elevator cables as an example: Assuming that the predicted life of elevator cables is the target variable, based on its production parameters and usage data, data from multiple elevator cable production batches are collected, including production parameters such as conductor diameter, insulation thickness, braiding density, as well as historical usage data such as installation date, usage environment, and maintenance records.
[0074] First, the data is represented. X represents the matrix of historical production parameter data and historical usage data, where each row corresponds to the data of an elevator cable batch, and each column corresponds to a feature, such as conductor diameter, insulation thickness, etc.; Y represents the vector of the corresponding elevator cable life, where each element is the life of an elevator cable batch.
[0075] The basic expression of the linear regression model is: Y = b0 + b1X1 + b2X2 + ... + b n X n +∈, where b0,b1,b2,...,b n are the model parameters, X1,X2,...,X n is a feature, ∈ is an error term; perform model training, use historical data X and Y, and use the linear regression algorithm to find the best parameters b0, b1, b2, ..., b n , to best fit the linear relationship of elevator cable life; by combining the production parameter data and usage data X of the new elevator cable production batch new Substituting into the trained model, the corresponding lifespan can be predicted.
[0076] In the above specific implementation, the linear regression model maps the production parameters and usage data in the historical data to the continuous value target of the elevator cable life. The establishment of the model makes it possible to estimate the life of the new cable based on the known production and usage conditions. This method can be used to predict other continuous value targets, such as cable performance indicators, durability, etc.
[0077] As a preferred embodiment of the above, Figure 2 As shown in the figure, the preprocessing of historical production parameter data and historical usage data includes:
[0078] S1: Determine whether there are missing values in the historical production parameter data set and the historical usage data set. If so, fill in the missing values. Otherwise, execute S2.
[0079] In this step, we first check whether each feature contains missing values. For features with missing values, we use one of the mean, median, and mode to fill them, or use a model-based method for predictive filling. The choice of filling method should be based on the distribution of the data and the pattern of missing values.
[0080] S2: Determine whether there are outliers in the two data sets after the missing value processing. If there are outliers, process them by truncation or tail-shrinking substitution. Otherwise, execute S3.
[0081] For outlier detection, statistical methods such as Z-Score or IQR can be used, or outlier detection algorithms based on machine learning can be used.
[0082] In the processing process, truncation is to directly delete or replace some extreme values in the data, usually a certain percentage of extreme values above and below, with a set threshold, so as to cut off the tail of the data set. A threshold is set during the specific execution. For example, the values exceeding 95% or below 5% quantiles in the data are selected to delete or replace, so that the tail of the data set is cut off. It is simple and intuitive, and can remove extreme values, making the data more consistent with the normal distribution without changing the overall structure of the data. Tail reduction is to replace the extreme values in the data with a set threshold, so as to shorten the tail of the data set and reduce the impact of outliers. A threshold is set during the operation. For example, the values exceeding 95% or below 5% quantiles in the data are replaced with the threshold of the corresponding quantile, so that the tail of the data set is shortened, retaining the existence of outliers, while reducing their impact on the overall distribution, which helps to better maintain the original structure of the data. The above two methods can be selected according to specific needs.
[0083] Compared with directly deleting outliers, truncation or tail-shrinking substitution can reduce the impact of outliers on the overall data distribution while retaining them, which helps to retain important information that outliers may contain without completely discarding the observed data brought by outliers; truncation or tail-shrinking substitution methods do not change the structure of the original data set, but only modify the extreme values, which can maintain the overall shape of the data set, so that no additional distortion will be introduced in subsequent analysis due to the processing of outliers; it can balance the existence of outliers and the rationality of data distribution to a certain extent, providing a more reliable basis for subsequent data analysis and modeling;
[0084] S3: Convert all types of data into numerical data and perform feature standardization and normalization;
[0085] The data type conversion process includes at least the following types of data conversion:
[0086] Conversion of categorical data: Take the original data of the usage of elevator cables, such as "residential elevator" and "commercial elevator". One-hot encoding is used during conversion to convert "residential elevator" into the binary feature "Is_Residential" and "commercial elevator" into "Is_Commercial".
[0087] Conversion of time-type data: Taking the production date of elevator cables as an example, the production date is converted into the number of days from a certain base date.
[0088] Conversion of text data: Taking the original data as the technical specification description of elevator cables as an example, text vectorization technology is used to convert the technical specification description into a numerical vector so that the machine learning model can process it.
[0089] Furthermore, through feature standardization and normalization, the dimensional differences between features can be eliminated, making the model more stable. Standardization and normalization can accelerate the convergence of the algorithm and enable the model to reach a stable state faster.
[0090] S4: Perform feature engineering based on domain knowledge and data characteristics to complete feature construction and selection and improve the generalization ability of the model.
[0091] The above preprocessing process helps to improve the performance of the model and enhance the understanding of elevator cable data, thereby improving the reliability of the model in practical applications. During the implementation process, careful processing was carried out taking into account business needs and actual data conditions, which helps to retain useful information and reduce the noise of the model when processing data.
[0092] As a preferred embodiment of the above, a comprehensive prediction model is established, such as Figure 3 As shown, including:
[0093] A1: Initialize the intercept and weight for the linear regression model;
[0094] Based on the above embodiment, the linear regression model expression is: Y = b0 + b1X1 + b2X2 + ... + b n X n +∈, where b0,b1,b2,...,b n are the model parameters, X1,X2,...,X n is the feature, ∈ is the error term.
[0095] In practice, the intercept b0 is usually initialized to zero or a small random value, and the weights b0,b1,b2,...,b n Initialization can be done by choosing zero, random values, or using some heuristic method. During the iteration of gradient descent, these weights will be gradually adjusted to minimize the loss function.
[0096] A2: Divide the dataset into training set and test set, and divide the data batches in the training set to better handle large-scale data and achieve online learning;
[0097] A3: For each data batch, extract small batches of data. For each small batch of data, calculate the gradient of the loss function with respect to the model parameters, and use the calculated gradient to update the intercept and weight of the model through the gradient descent algorithm.
[0098] A4: For the trained linear regression model, perform performance evaluation on the test set and tune the model parameters based on the evaluation results.
[0099] The above process is the modeling process. The gradient descent algorithm iteratively updates the model parameters on the training set. This process is used to build the model. The intercept and weight of the model are adjusted through repeated iterations so that the model can better fit the training data. In the real-time process, performance evaluation can be achieved by calculating the difference between the predicted results and the actual results of the model on the test set. Performance evaluation indicators include mean square error, mean absolute error, etc.; according to the evaluation results, the model parameters can be adjusted to improve the generalization performance of the model.
[0100] Specifically, the calculated gradient is used to update the intercept and weight of the model through the gradient descent algorithm, including:
[0101] B1: Set the initial intercept to b0 and the initial weights to b1, b2, ..., b n ;
[0102] B2: Setting the loss function L (i) for
[0103] B3: Calculate the gradient of the loss function with respect to the intercept and weight using the following formula:
[0104]
[0105] B4: Use the gradient descent algorithm to update the intercept and weight of the model using the following formula:
[0106]
[0107] in,
[0108] is the predicted value of the model for the i-th sample;
[0109] Y (i) is the actual target value;
[0110] is the characteristic value of n data in the i-th sample;
[0111] α is the learning rate, which is used to control the update step size;
[0112] Repeat steps B3 and B4 until convergence, that is, the update of model parameters tends to be stable after multiple iterations and no longer changes significantly.
[0113] In the implementation process, it is assumed that the relationship between the life Y of the elevator cable and its conductor diameter X1 and insulation thickness X2 needs to be predicted. The model expression is: Y = b0 + b1X1 + b2X2 + ∈. The specific initialization parameters are b0 = 0, b1 = 0, b2 = 0, and the gradient of the loss function to the model parameters is calculated as follows:
[0114]
[0115]
[0116]
[0117] Use the gradient descent algorithm to update the model parameters, specifically:
[0118]
[0119]
[0120]
[0121] In the above embodiment, the learning rate α gradually decreases with the repetition of steps B3 and B4, thereby accelerating convergence and improving the stability of training, while preventing the model from overfitting the training data in the later stages of training. As the learning rate decreases, the model adjusts the parameters more finely to make it closer to the optimal solution, which helps the model better adapt to the characteristics of the data in the later stages of training; the gradually decreasing learning rate helps to improve the generalization ability of the model, allowing it to perform better when faced with new data.
[0122] Common learning rate decay methods include exponential decay, step decay, or cosine annealing. Exponential decay means that after each training cycle (epoch), the learning rate gradually decreases according to the exponential function, which makes the learning rate decrease at an exponential rate, which is suitable for fast convergence. Step decay means reducing the learning rate at a specific period of training or a specific number of training steps. Cosine annealing means that the learning rate is periodically adjusted on the shape of the cosine function. The cosine annealing method adjusts the learning rate by the shape of the cosine function within a certain period. This helps to smoothly adjust the learning rate during training and avoid oscillation.
[0123] During the use of the model, as a preferred embodiment of the above embodiment, the linear regression model is continuously updated through an incremental learning method during use, including: continuously collecting new batch production parameters and new batch usage data for new batches of products, and preprocessing them. This step requires ensuring the adequacy of the new batch data; and updating the linear regression model through an incremental learning method using the preprocessed new batch production parameters and new batch usage data.
[0124] In this preferred solution, the new batch data is similar to the small batch data extracted from the training set, so the update of the model parameters is as described in the above embodiment and will not be repeated here. Through the above update method, the model can be updated immediately when new data is available without retraining the entire model, which provides a more real-time and flexible opportunity to adapt to changes and new trends. The immediate processing of new batch data helps the model to continuously improve performance in a dynamic environment.
[0125] Different batches of products may have different characteristics and performance requirements. By updating by batch, the model can be customized to better adapt to the characteristics of the current production batch. By updating by batch, finer-grained control can be achieved to ensure that the model pays more attention to the data of the most recent batch, avoiding frequent updates of the entire model. It can be updated on demand only when necessary, which can reduce computing and storage costs to a certain extent.
[0126] like Figure 4 As shown, the processing method of the elevator cable also includes:
[0127] C1: Set several evaluation indicators and normalize each evaluation indicator to obtain the normalized indicator value;
[0128] In this step, the evaluation indicators may include cost and production efficiency, etc. Unlike usage which focuses more on customer experience, the evaluation indicators may focus more on business operations issues; each indicator is normalized to ensure that they are on the same scale, which can be achieved through standardization or scaling, so that the weights of each indicator are relatively balanced when calculating the comprehensive score.
[0129] C2: For the processing schemes before and after revision, score each evaluation indicator and determine the weight of each score. During the implementation process, the weight can be determined based on professional knowledge, team consensus or mathematical methods such as hierarchical analysis method.
[0130] C3: For each solution, use the normalized indicator value and the determined weights of each score to calculate the overall score; this can be achieved by a simple weighted sum, that is, each indicator value is multiplied by the corresponding weight and then added together;
[0131] C4: Compare the comprehensive scores of the processing plans before and after revision. When the comprehensive score of the revised processing plan is greater than the comprehensive score of the processing plan before revision, the processing process is executed according to the revised processing plan, otherwise the revision is invalid and the original processing plan is maintained.
[0132] Through the set evaluation indicators, the plan can not only take into account factors directly related to the product, such as product quality and customer experience, but also pay attention to all aspects of the company's operations, such as cost, production efficiency, etc. This overall business perspective helps the company to more comprehensively understand the impact of decisions; when determining the scores and weights for the processing plans before and after revision, the weights of each indicator can be adjusted according to business needs and actual conditions, so that each indicator can be adjusted according to its own priorities to adapt to different business scenarios.
[0133] As a preferred embodiment of the above, when the comprehensive score of the revised processing scheme is less than the comprehensive score of the processing scheme before the revision, there are the following two optimization methods:
[0134] Method 1:
[0135] The parameters before the model revision are retained. When the comprehensive score of the revised processing plan is less than the comprehensive score of the processing plan before the revision, the model parameters are rolled back directly to the previous parameter state. The operation is simple, easy to understand and implement. It is more suitable when more attention is paid to the stability of the model and it is more sensitive to mutations.
[0136] Method 2:
[0137] The original parameters of the model are retained. When the comprehensive score of the revised processing scheme is less than the comprehensive score of the processing scheme before the revision, the revised parameters are weighted averaged with the original parameters. Through weighted average, smooth adjustment is achieved, and the revised parameters are gradually introduced to slow down the changes in the model and avoid drastic fluctuations in the model parameters. This helps to maintain the relative stability of the model during adjustment. It is more suitable if you want a smooth transition during adjustment and avoid excessive drastic changes.
[0138] When the comprehensive score of the revised processing scheme is less than the comprehensive score of the processing scheme before the revision, this may imply that the update of the model parameters has not achieved overall improvement, and may even introduce some problems or instability. Therefore, retaining the parameter state of the model before the revision or gradually introducing the revised parameters through smooth adjustment can be regarded as a way to limit and control the model update to ensure the stability and reliability of the model. This processing method helps to avoid the performance degradation or instability of the model due to inappropriate parameter updates. The common role of the above two methods is that when adjusting the model parameters, according to the comparison results of the comprehensive score of the revised processing scheme and the comprehensive score of the processing scheme before the revision, different model parameter corrections are performed to achieve the goal of better adapting to the actual situation while maintaining the stability of the model and avoiding frequent revisions; the specific method to be selected depends on the sensitivity to model changes and the requirements for model stability.
[0139] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A method for processing an elevator cable, characterized in that: include: Collect historical production parameter data and historical usage data of elevator cables and perform preprocessing; Establish a comprehensive prediction model through pre-processed historical production parameter data and historical usage data; Predicting the usage of the elevator cable processing solution after parameter adjustment according to the comprehensive prediction model; Revise the elevator cable processing plan after parameter adjustment according to the prediction results, and execute the processing according to the revised processing plan; The comprehensive prediction model is a linear regression model, which is continuously updated by an incremental learning method during use, and the incremental learning model adopts gradient descent of online learning; The prediction result is the service life of the elevator cable, and data of multiple elevator cable production batches are collected, including historical production parameter data such as conductor diameter, insulation thickness, and braiding density, as well as historical usage data such as installation date, usage environment, and maintenance records; First, the data is represented. X represents the matrix of historical production parameter data and historical usage data, where each row corresponds to the data of an elevator cable batch and each column corresponds to a feature. Y represents the vector of the corresponding elevator cable life, where each element is the life of a batch of elevator cables; the basic expression of the linear regression model is: Y = b0 + b1X1 + b2X2 + ... + b n X n +∈, where b0,b1,b2,...,b n are the model parameters, X1,X2,...,X n is a feature, ∈ is an error term; perform model training, use historical data X and Y, and use the linear regression algorithm to find the best parameters b0, b1, b2, ..., b n ; The linear regression model is continuously updated during use through an incremental learning method, including: Continuously collect new batch production parameters and new batch usage data for new batches of products, and perform preprocessing; The linear regression model is updated by an incremental learning method using the preprocessed new batch production parameters and new batch usage data; The processing methods of elevator cables also include: Setting a number of evaluation indicators, and normalizing each of the evaluation indicators to obtain a normalized indicator value; For the processing schemes before and after the revision, score the evaluation indicators respectively, and determine the weight of each score; For each solution, the comprehensive score is calculated using the normalized indicator value and the determined weights of each score; The comprehensive scores of the processing plans before and after the revision are compared. When the comprehensive score of the revised processing plan is greater than the comprehensive score of the processing plan before the revision, the processing process is executed according to the revised processing plan, otherwise the revision is invalid and the original processing plan is maintained.
2. The elevator cable processing method according to claim 1, characterized in that: The preprocessing of the historical production parameter data and the historical usage data includes: S1: Determine whether there are missing values in the historical production parameter data set and the historical usage data set. If so, fill in the missing values. Otherwise, execute S2. S2: Determine whether there are outliers in the two data sets after the missing value processing. If there are outliers, process them by truncation or tail-shrinking substitution. Otherwise, execute S3. S3: Convert all types of data into numerical data and perform feature standardization and normalization; S4: Perform feature engineering based on domain knowledge and data characteristics.
3. The elevator cable processing method according to claim 2, characterized in that: When filling missing values, use one of the mean, median, and mode.
4. The elevator cable processing method according to claim 1, characterized in that: Establishing the comprehensive prediction model comprises: For the linear regression model, initialize the intercept and weight; Divide the data set into training set and test set, and divide the data batches in the training set; For each data batch, extract a small batch of data. For each small batch of data, calculate the gradient of the loss function with respect to the model parameters, and use the calculated gradient to update the intercept and weight of the model through the gradient descent algorithm; For the trained linear regression model, a performance evaluation is performed on the test set, and the model parameters are tuned according to the evaluation results.
5. The elevator cable processing method according to claim 4, characterized in that: Use the calculated gradients to update the intercept and weights of the model through the gradient descent algorithm, including: B1: Set the initial intercept to b0 and the initial weights to b1, b2, ..., b n ; B2: Setting the loss function L (i) for ; B3: Calculate the gradient of the loss function with respect to the intercept and weight using the following formula: B4: Use the gradient descent algorithm to update the intercept and weight of the model using the following formula: in, is the predicted value of the model for the i-th sample; Y (i) is the actual target value; is the characteristic value of n data in the i-th sample; α is the learning rate, which is used to control the update step size; Repeat steps B3 and B4 until convergence.
6. The elevator cable processing method according to claim 5, characterized in that: The learning rate α is gradually reduced as steps B3 and B4 are repeated.
7. The elevator cable processing method according to claim 1, characterized in that: The parameters before the model revision are retained. When the comprehensive score of the revised processing plan is less than the comprehensive score of the processing plan before the revision, the model parameters are rolled back.
8. The elevator cable processing method according to claim 1, characterized in that: The original parameters of the model are retained, and when the comprehensive score of the revised processing scheme is less than the comprehensive score of the processing scheme before the revision, the revised parameters are weighted averaged with the original parameters.
