Square pile production parameter dynamic adjustment method, device and equipment and medium

Abstract

The invention relates to a square pile production parameter dynamic adjustment method and device, equipment and a medium. The method comprises the steps that square pile production parameter data are collected in real time, a production parameter set is formed, and a pre-trained neural network model is adopted to analyze the coupling relation between parameters to generate a coupling strength matrix, using an improved K-means clustering algorithm to extract hysteresis effect features in the matrix, combining with a Kalman filtering algorithm to process and generate a parameter prediction adjustment vector, and calculating the deviation between the vector and a preset target production parameter set to obtain prediction deviation data; and a preset parameter adjustment rule base is called based on the deviation to generate a square pile production parameter adjustment instruction, dynamic optimization adjustment of square pile production parameters is achieved, and the stability of the production process and the product quality are improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing of building materials, and in particular relates to a method, device, equipment and medium for dynamic adjustment of production parameters of square piles. Background Technology

[0002] With the development of technology in the field of intelligent manufacturing of building materials, data-driven optimization technology for the production process of square piles has emerged. This technology achieves process adjustment by collecting production parameters in real time and combining them with intelligent algorithms, and features multi-parameter coupling analysis, dynamic response, and predictive control. In traditional technology, the adjustment of square pile production parameters mainly relies on manual experience or fixed rules. By periodically collecting key parameters (such as concrete mix proportion, tension force, and steam curing temperature), manual intervention or local adjustments are made based on empirical formulas or static thresholds. The processing often lags behind actual production changes and it is difficult to fully consider the complex coupling effects and time lag characteristics between parameters.

[0003] Current methods for producing square piles, or traditional approaches, suffer from the following problems: parameter adjustment relies on manual experience and lacks quantitative analysis of the dynamic coupling relationships between multiple parameters, resulting in low adjustment accuracy and delayed response; traditional methods struggle to effectively capture the time-series lag effects during production, often leading to inappropriate timing or over-adjustment, impacting production stability; existing methods lack adaptive optimization mechanisms, failing to dynamically update models and rules based on real-time production data, resulting in declining long-term application effectiveness. Traditional methods typically do not integrate multimodal evaluation of deviations with optimization objectives, making it difficult to achieve precise and stable dynamic parameter adjustment, thus hindering further improvements in product quality and production efficiency. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, equipment, and medium for dynamically adjusting the production parameters of square piles that can solve the above problems.

[0005] Firstly, this application provides a method for dynamically adjusting the production parameters of square piles, including:

[0006] Real-time collection of square pile production parameter data to form a set of square pile production parameters;

[0007] A pre-trained neural network model is used to analyze the coupling relationship between the set of production parameters for square piles and generate a coupling strength matrix.

[0008] An improved K-means clustering algorithm is used to extract the hysteresis effect features in the coupling strength matrix, and a Kalman filter algorithm is used to process the hysteresis effect features to generate a parameter prediction adjustment vector.

[0009] The deviation between the predicted adjustment vector of the calculated parameters and the preset target production parameter set is used to generate predicted deviation data.

[0010] Based on the predicted deviation data, the preset parameter adjustment rule library is invoked to generate instructions for adjusting the production parameters of the square piles.

[0011] In one embodiment, a pre-trained neural network model is used to analyze the coupling relationship between the set of square pile production parameters to generate a coupling strength matrix, including:

[0012] The set of square pile production parameters is time-series aligned and standardized to form a standardized parameter sequence.

[0013] The standardized parameter sequence is input into a pre-trained neural network model to calculate the attention weights of different production parameters in the standardized parameter sequence at a preset time step.

[0014] Based on attention weights, a hybrid metric method based on mutual information and cosine similarity is used to calculate the correlation strength between any two production parameters in the standardized parameter sequence, and generate a correlation strength sequence.

[0015] Perform temporal convolutional fusion operations on the correlation strength sequences to extract multi-scale temporal features;

[0016] Pooling operations are performed on multi-scale temporal features to generate a coupling strength matrix.

[0017] In one embodiment, an improved K-means clustering algorithm is used to extract hysteresis features from the coupling strength matrix, and a Kalman filter algorithm is used to process the hysteresis features to generate a parameter prediction adjustment vector, including:

[0018] An adaptive sliding window is used to dynamically segment the coupling strength matrix over time, generating a sequence of time-series matrix slices.

[0019] An improved K-means clustering algorithm was used to perform cluster analysis on the time series matrix slice sequences, extract the cluster centers of each category, and form the dominant lag pattern sequences;

[0020] The dominant hysteresis mode sequence is input into a Kalman filter for state estimation, resulting in a sequence of predicted parameter adjustments.

[0021] A spatial mapping transformation is performed on the predicted value sequence to generate a parameter prediction adjustment vector.

[0022] In one embodiment, the deviation between the parameter prediction adjustment vector and the preset target production parameter set is calculated to generate prediction deviation data, including:

[0023] Based on the dynamic time warping algorithm, the time dimension of the alignment parameter prediction adjustment vector and the preset target production parameter set is adjusted to generate a vector sequence;

[0024] A multimodal deviation fusion method is adopted to calculate the joint deviation of the time-aligned vector sequence in the time domain and frequency domain, and generate a multimodal deviation matrix.

[0025] Principal component analysis was performed on the multimodal deviation matrix to extract the dominant deviation eigenvectors.

[0026] Based on the dominant bias eigenvector, significance assessment based on Mahalanobis distance is used to generate prediction bias data.

[0027] In one embodiment, based on the dominant bias eigenvector, a significance assessment based on Mahalanobis distance is used to generate predicted bias data, including:

[0028] Based on the dominant deviation feature vector, an empirical probability distribution P is constructed, and a reference probability distribution Q of the preset target production parameter set is obtained;

[0029] The following formula can be used to calculate the relationship between the empirical probability distribution P and the reference probability distribution Q. divergence :

[0030]

[0031] in, For preset adjustment parameters, and Let be the mean vector and covariance matrix of the empirical probability distribution P, respectively. and These are the mean vector and covariance matrix of the reference probability distribution Q, respectively. It is the determinant of a matrix;

[0032] The calculated divergence As prediction bias data.

[0033] In one embodiment, based on the prediction deviation data, a preset parameter adjustment rule base is invoked to generate a square pile production parameter adjustment instruction, including:

[0034] Based on prediction deviation data, generate rule-based query conditions;

[0035] The rule query conditions are matched with the rules in the preset parameter adjustment rule base to generate a matching rule set;

[0036] Parse the adjustment logic and constraints in the matching rule set to generate parameter adjustment constraints;

[0037] Based on parameter adjustment constraints and prediction deviation data, an optimization function is constructed with the objective of minimizing prediction deviation and adjustment magnitude.

[0038] Solve the optimization function to obtain the parameter adjustment solution vector;

[0039] According to the preset vector-instruction mapping rules, the parameter adjustment solution vector is mapped to the square pile production parameter adjustment instruction.

[0040] In one embodiment, after generating the square pile production parameter adjustment instruction, the method further includes:

[0041] Acquire square pile production parameter data within a preset acquisition period after executing the square pile production parameter adjustment command;

[0042] Calculate the actual deviation between the square pile production parameter data within the preset collection period and the preset target production parameter set;

[0043] If the actual deviation data exceeds the preset deviation threshold, then the actual deviation data, the corresponding square pile production parameter adjustment instructions, and the square pile production parameter data within the preset collection period will be used as samples to construct an incremental training sample set.

[0044] Based on the incremental training sample set, the online sequence extreme learning machine algorithm is used to incrementally update the pre-trained neural network model to obtain the updated neural network model.

[0045] Select historical square pile production parameter data within a preset time period before the execution of the square pile production parameter adjustment command, input it into the updated neural network model for re-prediction, and generate a set of model prediction outputs.

[0046] The mean, variance, and deviation fluctuations of the predicted output set of the computational model are used as statistical characteristics.

[0047] Based on statistical characteristics, a threshold iterative optimization method is adopted to adjust the preset parameters and the rule thresholds corresponding to each rule in the rule base.

[0048] Secondly, this application also provides a dynamic adjustment device for square pile production parameters, comprising:

[0049] The data acquisition module is used to collect square pile production parameter data in real time and form a set of square pile production parameters;

[0050] The coupling relationship analysis module is used to analyze the coupling relationship between the set of square pile production parameters using a pre-trained neural network model, and generate a coupling strength matrix.

[0051] The lag feature processing module is used to extract lag effect features from the coupling strength matrix using an improved K-means clustering algorithm, and to process the lag effect features using a Kalman filter algorithm to generate parameter prediction adjustment vectors.

[0052] The deviation calculation module is used to calculate the deviation between the parameter prediction adjustment vector and the preset target production parameter set, and generate prediction deviation data.

[0053] The adjustment instruction generation module is used to generate adjustment instructions for square pile production parameters based on the predicted deviation data and by calling the preset parameter adjustment rule library.

[0054] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method for dynamically adjusting the production parameters of square piles.

[0055] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method for dynamically adjusting the production parameters of square piles.

[0056] The aforementioned method, device, computer equipment, and storage medium for dynamic adjustment of square pile production parameters collect square pile production parameter data in real time and form a production parameter set. A pre-trained neural network model is used to analyze the coupling relationship between parameters to generate a coupling strength matrix. An improved K-means clustering algorithm is used to extract hysteresis features from the matrix, and a Kalman filter algorithm is combined to generate a parameter prediction adjustment vector. The deviation of this vector from the preset target production parameter set is calculated to obtain the predicted deviation data. Based on this deviation, a preset parameter adjustment rule library is invoked to generate square pile production parameter adjustment instructions, achieving dynamic optimization and adjustment of production parameters. This improves the accuracy and real-time performance of parameter adjustment, overcomes the problems of response lag and inaccurate adjustment caused by reliance on manual experience in traditional methods, and enhances the stability of the production process and the consistency of product quality by quantifying the dynamic coupling relationship and hysteresis effect between multiple parameters. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of the method for dynamically adjusting the production parameters of square piles according to the present invention;

[0059] Figure 2 This is a structural diagram of the dynamic adjustment device for square pile production parameters of the present invention;

[0060] Figure 3 This is a structural diagram of a dynamic adjustment device for square pile production parameters in one embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] In one embodiment, such as Figure 1 As shown, a method for dynamically adjusting the production parameters of square piles is provided. The system hardware architecture mainly consists of terminal equipment (such as industrial computers, PLC controllers, and sensor arrays) deployed at the production site, edge servers, and a cloud data center. The terminal equipment is responsible for real-time collection of square pile production parameter data (such as concrete mix proportions, tension force, and steam curing temperature), and transmits the data to the edge server via an industrial network for preliminary processing and coupling relationship analysis. The cloud server undertakes large-scale historical data storage, neural network model training, and rule base updates, and sends optimization instructions to the terminal actuators via a downlink. In practical applications, when multi-parameter coupling fluctuations or lag effects occur in the square pile production line, the terminal and server collaborate to dynamically adjust the parameters through real-time data interaction: the terminal continuously collects data and uploads it to the edge server; the server generates a coupling strength matrix and a predicted adjustment vector based on a pre-trained model; after deviation calculation, it calls the rule base to generate adjustment instructions, which are then sent to the actuators (such as frequency converters and valve controllers) via the terminal, achieving closed-loop optimization adjustment of production parameters. Simultaneously, the model and rule base are updated through a feedback mechanism to improve the system's adaptive capability. This embodiment illustrates the method by applying it to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0063] S01 collects square pile production parameter data in real time to form a set of square pile production parameters.

[0064] Data can be collected in real time using a combination of dedicated data acquisition equipment and sensors deployed at specific workstations on the square pile production line. The data acquisition equipment and parameter correspondence are as follows: the concrete water-cement ratio is collected using an online concrete moisture meter (model: such as HM-2000); the tension control force is collected using a tension sensor (range: ...). Precision: The temperature and humidity of the steam curing chamber are collected by a temperature and humidity sensor (model: SHT30, temperature range: ...). Humidity range: The centrifugal speed is collected by a photoelectric speed sensor (model: such as RS485, measurement range: ...). Acquisition; the preset sampling frequency can be set according to the dynamic response characteristics of the parameters. (Water-cement ratio, tensile force) (Temperature, humidity, centrifuge speed) ensure that instantaneous fluctuations in parameters are captured; data cleaning methods can be adopted. Criteria for removing outliers (i.e., when parameter values ​​exceed...) When a value is outside the specified range, it is identified as an outlier and replaced with the mean of the three adjacent valid data points. Missing data (duration of missing data) is then filled using linear interpolation. (Applicable to time), using the production line synchronization clock signal (accuracy: Timing alignment is performed using Z-score normalization (formula: ,in The parameter is the historical average over the past 3 months. Map the parameters to (the historical 3-month standard deviation). The intervals form a standardized set of square pile production parameters, providing a standardized and highly consistent input source for subsequent coupling relationship analysis.

[0065] S02 uses a pre-trained neural network model to analyze the coupling relationship between the set of production parameters for square piles and generates a coupling strength matrix.

[0066] The set of square pile production parameters collected in the aforementioned steps is input into a neural network model pre-trained based on historical production data (an LSTM model enhanced with an attention mechanism; the model structure is: number of input layer nodes = number of types of square pile production parameters (e.g., for 6 types of parameters, the number of input layer nodes)). ), Input dimension = (time step T, number of parameter types N), where (Corresponding to 30 sampling periods, each period is 5 minutes); the hidden layer is a 2-layer LSTM, the first layer has 128 nodes, and the second layer has [number of nodes missing]. The attention mechanism employs ScaledDot-ProductAttention, with a certain number of attention heads. This applies to the temporal features of the LSTM output; the output layer has a certain number of nodes. 1. Initial values ​​of the correlation strength between N parameters), with Sigmoid as the activation function; 2. Training data: Historical data of square pile production over the past 12 months, totaling 1000 qualified production batches, each batch containing 3000 sets of time-series data (time steps) ),according to The dataset is divided into a training set (700 batches), a validation set (200 batches), and a test set (100 batches); 3. Training parameters: The optimizer used is Adam, and the initial learning rate is... (The iteration count decreases to 0.8 times the previous value every 50 iterations), number of iterations Round, batch size = 32, loss function uses mean squared error (MSE) (calculates the squared difference between the correlation strength of the model output and the correlation strength of the actual process verification); pre-training verification is performed when the test set... Association strength prediction accuracy ( When the model pre-training is complete, training is stopped and the model parameters are saved. The dynamic interaction strength between parameters is quantified by calculating the attention weights of different production parameters at multiple preset time steps. A hybrid measurement method that integrates mutual information and cosine similarity is used to calculate the correlation strength between any two parameters and form a temporal correlation sequence. Multi-scale convolution operations are applied to this sequence in the time dimension to extract coupling features at different time granularities. The sequence is then aggregated by pooling layers to generate a strength matrix that represents the overall coupling relationship between production parameters. This matrix intuitively reflects the interaction strength and time-varying characteristics of each parameter pair in a two-dimensional form, providing a data foundation for subsequent hysteresis effect analysis.

[0067] S03 uses an improved K-means clustering algorithm to extract hysteresis features from the coupling strength matrix, and then uses a Kalman filter algorithm to process the hysteresis features to generate a parameter prediction adjustment vector.

[0068] Specifically, the coupling strength matrix can be dynamically segmented over time using an adaptive sliding window to generate a time-series matrix slice sequence. An improved K-means clustering algorithm (introducing dynamic time warping distance as a similarity metric) is used to perform cluster analysis on this sequence, extracting the cluster centers of each category to form a dominant lag pattern sequence. The dominant lag pattern sequence is then input into a Kalman filter for state estimation and noise filtering to obtain the optimal predicted value sequence of parameter adjustments. This sequence is then transformed into a parameter prediction adjustment vector that conforms to the interface specification of the production control system through spatial mapping transformation, providing predictive input for subsequent deviation calculation and instruction generation.

[0069] S04, calculate the deviation between the parameter prediction adjustment vector and the preset target production parameter set, and generate prediction deviation data.

[0070] Specifically, a dynamic time warping algorithm can be used to align the temporal dimensions of the parameter prediction adjustment vector and the target parameter set, eliminating the time scale difference between them. A multimodal deviation fusion method is employed to simultaneously calculate the difference between the absolute deviation in the time domain and the spectral characteristics in the frequency domain, generating a multimodal deviation matrix that fuses time and frequency characteristics. Principal component analysis is used to reduce the dimensionality of this matrix, extracting the dominant deviation feature vector that represents the most significant deviation direction. A significance assessment model is constructed based on Mahalanobis distance to quantify the statistical distance between the current production state and the target state, generating predictive deviation data with probabilistic significance, providing a quantitative basis for subsequent parameter adjustments.

[0071] S05, based on the predicted deviation data, calls the preset parameter adjustment rule library to generate a square pile production parameter adjustment instruction.

[0072] Specifically, the system can generate rule query conditions based on the characteristics of the predicted deviation data, calculate the similarity between the rules and the rules in the preset parameter adjustment rule library using a fuzzy matching algorithm, and generate a matching rule set; it can parse the adjustment logic and process constraints in the matching rule set, and construct a mathematical expression for a multi-objective optimization function that minimizes the predicted deviation and adjustment range; it can solve the function using a constraint optimization algorithm and obtain the parameter adjustment solution vector, and convert the solution vector into a square pile production parameter adjustment instruction that can be directly issued to the production line actuator through a preset vector-instruction mapping rule, thus completing the closed-loop transformation from deviation analysis to execution instruction.

[0073] The aforementioned method for dynamically adjusting square pile production parameters collects production parameter data in real time to form a set of production parameters. A pre-trained neural network model is used to analyze the coupling relationships between parameters to generate a coupling strength matrix. An improved K-means clustering algorithm is used to extract hysteresis features from the matrix, and a Kalman filter algorithm is combined to generate a parameter prediction adjustment vector. The deviation between this vector and the preset target production parameter set is calculated to obtain the predicted deviation data. Based on this deviation, a preset parameter adjustment rule base is invoked to generate square pile production parameter adjustment instructions. This achieves dynamic optimization and adjustment of square pile production parameters, improving the accuracy and real-time performance of parameter adjustment. It overcomes the problems of response lag and inaccurate adjustment caused by reliance on manual experience in traditional methods. By quantifying the dynamic coupling relationships and hysteresis effects between multiple parameters, the stability of the production process and the consistency of product quality are enhanced, improving the intelligence level and overall efficiency of square pile production.

[0074] In one embodiment, a pre-trained neural network model is used to analyze the coupling relationship between the set of square pile production parameters to generate a coupling strength matrix, including:

[0075] S11, perform time-series alignment and standardization on the set of square pile production parameters to form a standardized parameter sequence;

[0076] S12, Input the standardized parameter sequence into the pre-trained neural network model, and calculate the attention weights of different production parameters in the standardized parameter sequence at a preset time step;

[0077] S13. Based on the attention weight, a hybrid measurement method based on mutual information and cosine similarity is used to calculate the correlation strength between any two production parameters in the standardized parameter sequence, and generate a correlation strength sequence.

[0078] S14 performs a temporal convolutional fusion operation on the correlation strength sequence to extract multi-scale temporal features;

[0079] S15 performs pooling operations on multi-scale temporal features to generate a coupling strength matrix.

[0080] For example, timing alignment uses the production line's synchronous clock signal as a unified time reference to calibrate the acquisition timestamps of different parameters, eliminating time deviations caused by the response delay of the acquisition equipment and ensuring that each parameter corresponds one-to-one in the time dimension; standardization can be achieved using Z-score standardization (subtracting the historical acquisition mean from the value of each parameter and dividing by the standard deviation), mapping the parameter values ​​to... Range, eliminating the difference in units between different parameters (e.g., temperature unit). To mitigate the interference of differences in tensile force (unit: kN) and numerical range on subsequent analysis, a standardized parameter sequence with unified numerical scale was formed.

[0081] The pre-trained neural network model is a deep neural network with an attention mechanism (such as an attention-enhanced Long Short-Term Memory network LSTM or Transformer model) that has been pre-trained based on historical data of square pile production (covering parameter data under normal and abnormal working conditions). Its training objective is to learn the potential correlation between parameters. After the standardized parameter sequence is input into the model, the model calculates the attention weight of different production parameters at a preset time step (set according to the square pile production process cycle, such as the duration of the steam curing stage and the duration of the tensioning operation, such as 5 minutes / step or 10 minutes / step). The weight quantifies the degree of influence of a certain parameter on other parameters at a specific time step. The larger the weight value, the stronger the influence.

[0082] Based on the aforementioned attention weights, a hybrid metric method using mutual information and cosine similarity can be employed to calculate the correlation strength between any two production parameters in a standardized parameter sequence. Mutual information measures the statistical dependency between the two parameters (i.e., the degree to which the information of one parameter explains the information of the other), while cosine similarity measures the similarity of the two parameters in their temporal trends (i.e., the cosine of the angle between the time-series vectors of the two parameters). By weighting and fusing the mutual information and cosine similarity values ​​according to preset weights (determined based on historical data, mutual information weight 0.6, cosine similarity weight 0.4), the correlation strength value of the two parameters at a single time step is obtained. The correlation strength values ​​at all time steps are then arranged chronologically to generate a correlation strength sequence reflecting the change in the correlation strength between the two parameters over time. Finally, a temporal convolutional fusion operation is performed on the correlation strength sequence. Multi-scale 1D convolutional kernels (such as kernels of sizes 3, 5, and 7) are used for convolution operations. Small-scale convolutional kernels extract short-term (1-3 time steps) correlation fluctuation features, while large-scale convolutional kernels extract long-term (5-7 time steps) correlation trend features. The multiple feature maps obtained from convolution are concatenated along the channel dimension to achieve convolutional fusion in the time dimension and extract multi-scale temporal features. Pooling operations are performed on the multi-scale temporal features (maximum pooling to select the maximum value of each local region or average pooling to calculate the average value of each local region) to reduce the feature dimensionality and retain key temporal information. The two-dimensional matrix obtained after pooling is the coupling strength matrix. The rows and columns of this matrix correspond to different square pile production parameters, and the matrix element values ​​represent the overall coupling strength between the corresponding two parameters. This allows for a direct quantification of the dynamic coupling relationship between parameters, providing a data foundation for subsequent lag effect feature extraction.

[0083] In one embodiment, an improved K-means clustering algorithm is used to extract hysteresis features from the coupling strength matrix, and a Kalman filter algorithm is used to process the hysteresis features to generate a parameter prediction adjustment vector, including:

[0084] S21, an adaptive sliding window is used to dynamically segment the coupling strength matrix over time to generate a time-series matrix slice sequence;

[0085] S22, an improved K-means clustering algorithm is used to perform cluster analysis on the time series matrix slice sequence, extract the cluster centers of each category, and form the dominant lag pattern sequence;

[0086] S23, input the dominant hysteresis mode sequence into the Kalman filter for state estimation to obtain the predicted value sequence of parameter adjustment;

[0087] S24, perform spatial mapping transformation on the predicted value sequence to generate parameter prediction adjustment vector.

[0088] Specifically, the adaptive sliding window is a segmentation tool that dynamically adjusts the window length based on the temporal fluctuation characteristics of the coupling strength matrix (such as the frequency of abrupt changes in parameter correlation strength and the duration of stable periods). The window length adjustment range is preset to 5-15 time steps according to the production process cycle of square piles (such as the duration of the steam curing stage and tensioning stage). By calculating the variance change rate of the data within the window in real time, when the variance change rate exceeds a preset threshold (0.3), the window is reduced to capture abrupt changes; when the variance change rate is below the threshold, the window is expanded to cover stable periods. In this way, the coupling strength matrix is ​​dynamically segmented over time. The sub-matrices corresponding to each window segment are arranged in chronological order to form a time-series matrix slice sequence (initial window length = 10 time steps, corresponding to 50 minutes, adjustment range...). Each time step; the variance change rate is calculated as follows: let the current window variance be... The variance of the previous window is rate of change of variance (Dimensionless); when this value When the window length is 2, the window length is reduced by 2; when this value is... If the window length is increased by 2, otherwise the window length remains unchanged; - the coupling strength matrix is ​​segmented according to the adjusted window to generate a sequence of time-series matrix slices. (N is the number of parameter categories); the improved K-means clustering algorithm is an optimization of the traditional K-means algorithm. Its improvement lies in using Dynamic Time Warping (DTW) distance instead of Euclidean distance as the similarity measure to solve the problem of time-dimension offset between different slices in the time-series matrix slice sequence. Before clustering, the optimal number of clusters is determined by the elbow rule (usually 3-5 classes, corresponding to the common lag pattern types in square pile production). Clustering iteration is performed on the time-series matrix slice sequence: after initializing the cluster centers, the DTW distance between each slice and each center is calculated and assigned to the nearest cluster. The cluster centers are updated until the center positions converge. The converged cluster centers of each category are extracted (each center is a sub-matrix representing a certain type of lag feature). These centers are arranged according to the time order of the slices corresponding to the clusters to form the dominant lag pattern sequence. Among them, the TW distance calculation adopts the Sakoe-Chiba window constraint, and the window size = slice length. (For example, if the slice length is 10, then the window size is 1), to avoid excessive computational complexity; the elbow rule steps are: set the candidate range for the number of clusters k. For each k, calculate the sum of the DTW distances (denoted as SSE) from all slices to their cluster centers; plot the results. The curve shows the decrease in SSE as k increases from 3 to 4. (i.e., the elbow point) to determine the optimal number of clusters. The clustering iteration is as follows: initialize 3 cluster centers (randomly select 3 slices), iteratively calculate the DTW distance from each slice to the center and assign it to a cluster, update the cluster centers (taking the mean of all elements in that cluster), until the change in cluster centers is reached. (At this point, the process stops, and three cluster centers are extracted as the dominant hysteresis pattern sequence).

[0089] The Kalman filter is an algorithm module used for state estimation and noise suppression. Its state vector is set as the parameter adjustment value corresponding to the dominant hysteresis pattern sequence, and the observation vector is the current dominant hysteresis pattern value. First, the Kalman prediction equation estimates the predicted parameter adjustment value and error covariance at the current time based on the state at the previous time step. Then, the Kalman update equation, combined with the currently observed dominant hysteresis pattern value, corrects the prediction error to obtain the optimal parameter adjustment value estimate. The optimal estimates at each time step are arranged in chronological order to generate a sequence of predicted parameter adjustment values. The spatial mapping transformation is the operation of converting the predicted parameter adjustment value sequence from the feature space of cluster analysis to the actual control space of square pile production parameters. During the transformation, the physical constraints of each production parameter (such as the water-cement ratio control range of concrete) are considered. Tension control range The system calculates the interface parameter format of the production control system, performs scale normalization (mapping the predicted values ​​to the actual control range of the corresponding parameters) and dimension matching (ensuring that the dimension of the mapped vector is consistent with the number of production parameters), and generates a parameter prediction adjustment vector. This vector can be directly used for subsequent deviation calculation from the preset target production parameter set, providing a quantitative basis for the generation of parameter adjustment instructions.

[0090] In one embodiment, the deviation between the parameter prediction adjustment vector and the preset target production parameter set is calculated to generate prediction deviation data, including:

[0091] S31, Based on the dynamic time warping algorithm, align the parameter prediction adjustment vector with the temporal dimension of the preset target production parameter set to generate a vector sequence;

[0092] S32 employs a multimodal deviation fusion method to calculate the joint deviation of the time-aligned vector sequence in the time and frequency domains, generating a multimodal deviation matrix;

[0093] S33, perform principal component analysis on the multimodal deviation matrix to extract the dominant deviation eigenvectors;

[0094] S34. Based on the dominant bias eigenvector, significance assessment based on Mahalanobis distance is used to generate prediction bias data.

[0095] For example, the preset target production parameter set is a time series dataset containing the target values ​​of each production parameter, which is pre-set based on the quality standards of square pile products (such as concrete strength grade and pile size accuracy requirements) and production process optimization goals. The target parameter values ​​are recorded according to the time step of the production process, and have a correspondence with the time dimension of the prediction vector.

[0096] Dynamic Time Warping (DTW) is an algorithm used to align time series data with time offsets or length differences. It constructs a cost matrix between two sequences (with the Euclidean distance between the predicted adjustment value and the target value at each time step as the cost element), and uses dynamic programming to solve for the minimum cost matching path. This achieves alignment of the parameter prediction adjustment vector with the preset target production parameter set in the time dimension. After alignment, the predicted adjustment value and the corresponding target value at each time step are combined into a vector pair, which are then arranged in chronological order to form a time-aligned vector sequence.

[0097] The multimodal bias fusion method is a method that simultaneously calculates and fuses biases from both the time and frequency domains. In the time-domain bias calculation, the predicted adjustment value and the target value at each time step in the time-series alignment vector sequence are calculated using the absolute bias. or relative deviation The single-mode time-domain deviation is calculated to form a time-domain deviation sequence. For frequency-domain deviation calculation, a Fast Fourier Transform (FFT) is first performed on the time-series aligned vector sequence to convert the time-domain signal into a frequency-domain signal. The fundamental frequency, second harmonic, and other major frequency components are extracted, and the amplitude difference between the predicted signal and the target signal at each major frequency component is calculated. and phase difference A frequency domain deviation sequence is formed; a two-dimensional matrix is ​​constructed with time steps as rows and time domain deviation and each frequency domain deviation type as columns to generate a multimodal deviation matrix (the matrix element value is the deviation amount of the corresponding time step and the corresponding deviation type).

[0098] Principal Component Analysis (PCA) is an algorithm used for dimensionality reduction and key feature extraction. It calculates the covariance matrix of the standardized matrix, solves for the eigenvalues ​​and eigenvectors of the covariance matrix, and selects the cumulative contribution rate of the eigenvalues. The first k eigenvectors (k is the dimension after dimensionality reduction) are used to project the multimodal bias matrix into a new space formed by these k eigenvectors. The resulting dimension-reduced eigenvectors are the dominant bias eigenvectors (which centrally reflect the core information of multimodal bias and eliminate redundant noise).

[0099] In significance assessment based on Mahalanobis distance, Mahalanobis distance is a distance metric that considers the data distribution characteristics (mean and covariance). First, an empirical probability distribution is constructed based on the dominant deviation feature vector (assuming it follows a multivariate normal distribution, and its mean vector and covariance matrix are calculated). Then, a reference probability distribution corresponding to the preset target production parameter set is obtained (its mean vector and covariance matrix are obtained through statistical analysis of historical target production data). The Mahalanobis distance between the two distributions is calculated using the Mahalanobis distance formula. This distance value is the predicted deviation data (the larger the distance, the more significant the deviation between the parameter prediction adjustment vector and the target parameter, providing a quantitative basis for the subsequent generation of adjustment instructions).

[0100] In one embodiment, based on the dominant bias eigenvector, a significance assessment based on Mahalanobis distance is used to generate predicted bias data, including:

[0101] S41, Based on the dominant deviation feature vector, construct the empirical probability distribution P, and obtain the reference probability distribution Q of the preset target production parameter set;

[0102] S42, calculate the relationship between the empirical probability distribution P and the reference probability distribution Q using the following formula. divergence :

[0103]

[0104] in, For preset adjustment parameters, and Let be the mean vector and covariance matrix of the empirical probability distribution P, respectively. and These are the mean vector and covariance matrix of the reference probability distribution Q, respectively. It is the determinant of a matrix;

[0105] S43, the calculated divergence As prediction bias data.

[0106] Specifically, the empirical probability distribution P is a probability distribution constructed based on the dominant deviation feature vector, used to characterize the statistical distribution characteristics of the predicted deviation data of the current production parameters, while the reference probability distribution Q is a probability distribution constructed based on the historical qualified production data (selected production batch data with a product qualification rate ≥99% within the past 3 months) corresponding to the preset target production parameter set, which characterizes the deviation distribution characteristics under ideal production conditions.

[0107] When constructing the empirical probability distribution P, the dominant deviation eigenvector can first undergo a data distribution test (using the KS test to verify whether it conforms to a multivariate normal distribution). Under the assumption of conforming to a multivariate normal distribution, the mean of each dimension of the dominant deviation eigenvector can be calculated using the sample mean formula, forming the mean vector of the empirical probability distribution P. Then, the covariance between data in each dimension is calculated using the sample covariance formula, and the covariance matrix of the empirical probability distribution P is constructed. When obtaining the reference probability distribution Q, extract the deviation feature data that has the same dimension as the dominant deviation feature vector from the historical production data corresponding to the preset target production parameter set. Use the same mean and covariance calculation method to obtain the mean vector of the reference probability distribution Q. Covariance Matrix Calculate the relationship between the empirical probability distribution P and the reference probability distribution Q. When divergence, The preset adjustment parameter (α) is determined based on historical deviation data verification: the verification data is selected from the past 6 months of square pile production deviation data, totaling 500 groups (each group contains samples of empirical probability distribution P and reference distribution Q); the verification index is defined as deviation identification accuracy = ( Divergence can correctly distinguish between acceptable deviations ( ) and non-conforming deviation ( The verification steps are: (Number of samples) / Total number of samples; Calculate the results for each of the 500 samples. Divergence; Statistical analysis of each The corresponding deviation recognition accuracy; results show Highest accuracy (time) ),and In the formula, the weights of the mean difference term and the covariance ratio term are equal (both are expressed as follows: ...). The coefficients (reflected in the equation) can balance the influence of the distribution mean and covariance on the divergence, therefore, they are determined. (Optimal adjustment parameters) This represents matrix determinant operations, and the calculation process is as follows: calculate the mean vector difference. transpose Calculate the weighted covariance matrix inverse matrix Multiply the transposed mean difference by the inverse matrix and the original mean difference in sequence, then multiply by... The first part of the formula is obtained; the weighted covariance matrix is ​​calculated. determinant Calculate the covariance matrix of the empirical probability distribution P. determinant And seek its Power of 1 Calculate the Q covariance matrix of the reference probability distribution. determinant And find its α power. Substitute the three into the logarithmic term Multiplying by 1 / 2 gives the second part of the formula. Adding the two parts gives the relationship between the empirical probability distribution P and the reference probability distribution Q. Divergence.

[0108] Since α-divergence can quantify the degree of difference between two probability distributions (the larger the divergence value, the more significant the difference between the distribution of the current production parameter prediction deviation data and the distribution of the deviation data under the ideal target condition, i.e., the more serious the current prediction deviation), the calculated... Divergence is directly used as prediction deviation data, providing a precise basis for deviation quantification for subsequent generation of adjustment instructions by calling the preset parameter adjustment rule library.

[0109] In one embodiment, based on the prediction deviation data, a preset parameter adjustment rule base is invoked to generate a square pile production parameter adjustment instruction, including:

[0110] S51, Generate rule query conditions based on prediction deviation data;

[0111] S52, perform fuzzy matching between the rule query conditions and the rules in the preset parameter adjustment rule base to generate a matching rule set;

[0112] S53, parse the adjustment logic and constraints in the matching rule set, and generate parameter adjustment constraints;

[0113] S54, Based on parameter adjustment constraints and prediction deviation data, construct an optimization function with the objective of minimizing prediction deviation and adjustment magnitude;

[0114] S55, Solve the optimization function to obtain the parameter adjustment solution vector;

[0115] S56, according to the preset vector-instruction mapping rules, maps the parameter adjustment solution vector to the square pile production parameter adjustment instruction.

[0116] For example, the preset parameter adjustment rule library has four types of parameters (concrete mix proportion, tension parameters, steam curing parameters, and centrifugal parameters), with a total of 20 core rules. Each rule is stored in a structured manner with trigger conditions, adjustment logic, and constraint conditions. An example is as follows: 1. Steam curing temperature adjustment rule (Rule ID: T001): The trigger condition is the predicted deviation data (…). divergence The corresponding production parameters are the steam curing chamber temperature and the duration of the deviation. One sampling period; the adjustment logic is as follows: temperature adjustment amount. (Where D is the prediction bias data, unit: dimensionless quantity), that is, for every increase of 0.1 in D, Increase The constraint is that it is a single... Adjusted temperature (Complies with the steam curing temperature requirements of GB / T13476-2017 "Pre-tensioned Prestressed Concrete Pipe Piles"); Tension Adjustment Rules (Rule ID: F001): Triggering condition is the predicted deviation data. The corresponding production parameters are tension control force and deviation duration. One sampling period; the adjustment logic is as follows: tension adjustment amount. (D represents the prediction bias data), meaning that for every 0.1 increase in D, Increase The constraint is that it is a single... Adjusted tension (Matching the rated range of the tensioning equipment on the production line); The initial size of the rule base is 20 rules, covering all core production parameters, and each rule has been verified through at least 50 sets of historical optimization data (adjusted deviation reduction rate). (Consider it a valid rule); the rule query conditions are query items based on the predicted deviation data, used to match the corresponding rules in the preset parameter adjustment rule base. When generating the rule, key features of the predicted deviation data need to be extracted, including the specific deviation value, the corresponding production parameter type (such as steam curing temperature deviation, tension force deviation), and the time series period associated with the deviation (the duration of the deviation determined based on the aforementioned time series data), and the query should be performed according to the preset query format of the rule base (such as deviation value: ( (Divergence unit), parameter category: steam curing temperature, duration: 2 sampling periods) are organized into rule query conditions. Fuzzy matching is used to avoid rule omissions due to slight deviations in prediction data caused by minor equipment fluctuations in actual production. It is a matching method adopted to avoid rule omissions caused by strict matching. The similarity between the rule query conditions and the triggering conditions of each rule in the preset parameter adjustment rule base can be calculated using a triangular membership function (e.g., if the deviation value in the query condition is 2.7, the deviation range of a certain rule triggering condition is...). Then the similarity is calculated as follows: (Where 2.6 is the midpoint of the interval and 0.4 is the half-width of the interval), set a similarity threshold (e.g., 0.7), and filter all similarities. The threshold rules form a matching rule set.

[0117] Parameter adjustment constraints are a set of constraints formed by integrating, deduplicating, and standardizing the constraints of each rule in the matching rule set. The parsing process needs to extract the restriction information on parameter adjustment from each rule, including the upper / lower limit of a single adjustment for each production parameter (e.g., the single adjustment of steam curing temperature should not exceed...). The tension should not be adjusted in a single instance. ), and the process compliance range of the adjusted parameters (e.g., the water-cement ratio of concrete needs to be within the range after adjustment). The relationships between parameters and cross-parameter adjustments (such as confirming that the tension is stable before adjusting the centrifugal speed) are established, and this information is transformed into mathematical constraint expressions (such as...). ,in (This is for adjusting the steam curing temperature), forming parameter adjustment constraints.

[0118] The optimization function aimed at minimizing prediction bias and adjustment magnitude needs to be constructed by combining prediction bias data and parameter adjustment constraints. The objective function is set as follows: Where D is the prediction bias data ( (divergence value) The parameter adjustment vector (containing the adjustment amount for each production parameter) ), for of Norm (characterizing the overall adjustment range of all parameters) , Weighting coefficients (set according to production priorities, such as prioritizing product quality). Prioritizing production stability );

[0119] The constraint is a parameter adjustment constraint (such as...) , , These represent the upper and lower limits of the adjustment amount for the i-th parameter, respectively, thus forming a complete constrained optimization function. When solving the optimization function, an appropriate algorithm can be selected based on the linear / nonlinear characteristics of the function. For linear optimization functions, gradient descent is used; for nonlinear optimization functions, a genetic algorithm is used. Through iterative calculation, the parameter adjustment vector that minimizes the objective function and satisfies all constraints is found (its dimension is consistent with the number of production parameters for the square pile, and each element corresponds to the optimal adjustment amount of a production parameter, such as...). (where F is the tension force).

[0120] The preset vector-instruction mapping rule is a pre-established correspondence table between parameter adjustment solution vectors and executable production instructions. The table clearly defines the device ID corresponding to each parameter adjustment (e.g., temperature controller with device ID T001 for steam curing temperature adjustment, tension controller with device ID F003 for tension force adjustment) and the instruction format (e.g., device ID: T001, parameter type: steam curing temperature, adjustment amount: ...). The execution timing (immediately) and data transmission protocol (such as Modbus protocol) are used. The mapping process requires matching each element in the parameter adjustment solution vector with the corresponding device and instruction information according to the rule table, and integrating them into a square pile production parameter adjustment instruction that conforms to the interface specification of the square pile production control system. This instruction can be directly sent to the corresponding production equipment for execution through the industrial bus to achieve precise and real-time adjustment of production parameters.

[0121] In one embodiment, after generating the square pile production parameter adjustment instruction, the method further includes:

[0122] S61, acquire the square pile production parameter data within the preset acquisition period after executing the square pile production parameter adjustment command;

[0123] S62, calculate the actual deviation data between the square pile production parameter data within the preset collection period and the preset target production parameter set;

[0124] S63, if the actual deviation data exceeds the preset deviation threshold, then the actual deviation data, the corresponding square pile production parameter adjustment instructions, and the square pile production parameter data within the preset collection period are used as samples to construct an incremental training sample set.

[0125] S64, based on the incremental training sample set, uses the online sequence extreme learning machine algorithm to incrementally update the pre-trained neural network model, and obtains the updated neural network model;

[0126] S65: Select historical square pile production parameter data within a preset time period before the square pile production parameter adjustment command is executed, input it into the updated neural network model for re-prediction, and generate a set of model prediction outputs.

[0127] S66, the mean, variance, and deviation fluctuations of the predicted output set of the calculation model are used as statistical characteristics;

[0128] S67. Based on statistical characteristics, a threshold iterative optimization method is adopted to adjust the preset parameters and the rule thresholds corresponding to each rule in the rule base.

[0129] Specifically, the preset data acquisition cycle is based on the dynamic response characteristics of the square pile production parameters (such as temperature parameters). It takes minutes to reach a stable state, and the tension parameters need to be... (Minutes stable) A preset time interval for collecting adjusted parameter data, usually set to... Each sampling period (a single sampling period is consistent with the time step of real-time data acquisition mentioned above, such as...) This ensures that the stable state data of the adjusted parameters can be fully captured;

[0130] The actual deviation data is the quantified value of the deviation between the square pile production parameter data within the preset collection period and the preset target production parameter set after the adjustment command is executed. Its calculation method is consistent with the predicted deviation data mentioned earlier (based on...). Divergence calculation); The incremental training sample set is a set of samples used to update the model. Each sample consists of a triplet consisting of (actual deviation data after executing the adjustment instruction - corresponding square pile production parameter adjustment instruction - square pile production parameter data within the preset collection period), containing complete correlation information of adjustment input, process data and effect feedback; The online sequence extreme learning machine algorithm is a highly efficient incremental neural network training algorithm. Its feature is that the weight parameters of the input layer and hidden layer of the neural network are fixed, and the output layer weights are updated iteratively only by adding new samples. It does not require retraining all historical data, which can significantly improve the update efficiency while ensuring model accuracy; The preset duration is the time length used to select historical data before adjustment, set according to the historical influence range of square pile production parameter adjustment. For example, selecting the data within 30 minutes before the execution of the adjustment instruction (corresponding to 6... Historical production parameter data for square piles (with a sampling period of 5 minutes per period) fully reflects the production status before adjustment; statistical features are key statistical indicators of the model prediction output set, including mean (arithmetic mean of all predicted values, reflecting the overall level of the prediction results), variance (square mean of the deviations of predicted values ​​from the mean, reflecting the dispersion of the prediction results), and deviation fluctuation amplitude (difference between the maximum and minimum values ​​among the predicted values, reflecting the stability of the prediction results); the threshold iteration optimization method is an optimization method that adjusts the rule threshold through multiple rounds of iteration to match the triggering frequency of rules in the preset parameter adjustment rule library with the actual deviation improvement effect; the rule threshold is the critical value of the triggering condition of each rule in the preset parameter adjustment rule library (such as the deviation value trigger threshold and the deviation duration threshold), which directly determines whether the rule is activated.

[0131] Sensors (such as temperature sensors and pressure sensors) deployed at key workstations on the square pile production line can collect all production parameter data in real time within a preset collection period after the execution of the square pile production parameter adjustment command; the data used in calculating and predicting deviations above can be reused. The divergence method uses the production parameter data collected within a preset collection period as the basis data for the empirical probability distribution P, and the distribution corresponding to the preset target production parameter set as the reference probability distribution Q, calculating the difference between the two. Divergence, the divergence value is the actual deviation data, ensuring the consistency of the deviation calculation logic; preset deviation threshold (determined based on statistical analysis of actual deviation data from the past 3 months of standard-compliant square pile production batches, such as...). If the divergence value is 0.7, exceeding this value indicates that the adjustment effect has not met the expected quality control requirements. If the calculated actual deviation data exceeds the preset deviation threshold, the actual deviation data corresponding to this adjustment, the square pile production parameter adjustment command that triggered this adjustment, and the complete production parameter data within the preset collection period are integrated into a sample and added to the incremental training sample set in chronological order. If it does not exceed the threshold, the sample is not constructed.

[0132] The samples in the incremental training sample set are input into the online sequence extreme learning machine algorithm in time series. This algorithm is based on the pre-trained neural network model structure (keeping the number of input layer nodes, hidden layer nodes, and activation function unchanged). It solves the output layer weight increment corresponding to the new sample by the least squares method. This increment is superimposed with the output layer weight of the original model to complete the incremental update of the model, resulting in an updated neural network model that can adapt to the changes in the current production status, avoiding the efficiency loss caused by retraining all historical data. According to the backtracking requirements of the impact of the square pile production parameter adjustment, the historical production parameter data within a preset time period before the square pile production parameter adjustment command is executed is selected. After preprocessing it according to the standardization method mentioned above, it is input into the updated neural network model for re-prediction, resulting in a model prediction output set containing the predicted parameter values ​​of each time step.

[0133] The model's predicted output set is statistically calculated to obtain the mean (sum of all predicted values ​​divided by the amount of predicted data), variance (sum of squared differences between each predicted value and the mean divided by the amount of predicted data), and deviation fluctuation (maximum value minus minimum value in the predicted output set). This forms a comprehensive statistical characteristic reflecting the model's predictive performance. A threshold iterative optimization method is employed, aiming to minimize the deviation fluctuation in the statistical characteristics (smaller fluctuation indicates more stable model predictions and more accurate rule triggering). The initial iteration step size for the rule threshold is set (e.g., adjusted by 0.05 each time). In the first iteration, the preset parameters are adjusted according to this step size. Adjust the trigger threshold of each rule in the rule base (e.g., adjust the deviation trigger threshold of a rule from 0.8 to 0.85), and then verify the deviation improvement effect when the adjusted rule is triggered through the updated neural network model. If the improvement effect (actual deviation reduction ratio) increases, continue iterating in this direction; if the improvement effect decreases, adjust the step size in the opposite direction. Repeat this iterative process until the deviation fluctuation is less than the preset stable threshold (e.g., 0.1). Determine the optimized rule threshold for each rule, and complete the dynamic optimization of the preset parameter adjustment rule base to further improve the stability of the production process and the consistency of product quality.

[0134] The aforementioned dynamic adjustment method for square pile production parameters involves real-time acquisition of square pile production parameter data to form a production parameter set. A pre-trained neural network model is used to analyze the coupling relationships between parameters to generate a coupling strength matrix. An improved K-means clustering algorithm is used to extract hysteresis features from the matrix, and a Kalman filter algorithm is combined to generate a parameter prediction adjustment vector. The deviation of this vector from the preset target production parameter set is calculated to obtain the prediction deviation data. Based on this deviation, a preset parameter adjustment rule base is invoked to generate square pile production parameter adjustment instructions, achieving dynamic optimization and adjustment of square pile production parameters. By quantifying the dynamic coupling relationships and hysteresis effects between multiple parameters, a hybrid metric method based on mutual information and cosine similarity is used to calculate the correlation strength, combined with multi-scale time-series feature extraction and... Divergence deviation assessment improves the accuracy and real-time performance of parameter adjustment, overcoming the problems of response lag and inaccurate adjustment caused by reliance on human experience in traditional methods. The neural network model is incrementally updated through an online sequence extreme learning machine algorithm, and the rule thresholds in the rule base are dynamically adjusted based on statistical features using a threshold iterative optimization method, which enhances the adaptive capability and long-term stability, effectively improving the stability of the square pile production process, the consistency of product quality, and the overall level of intelligence.

[0135] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0136] Based on the same inventive concept, this application also provides a device for dynamically adjusting the production parameters of square piles to implement the aforementioned method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for dynamically adjusting the production parameters of square piles provided below can be found in the limitations of the method for dynamically adjusting the production parameters of square piles described above, and will not be repeated here.

[0137] In one exemplary embodiment, such as Figure 2 As shown, a dynamic adjustment device for square pile production parameters is provided, comprising:

[0138] Data acquisition module 101 is used to collect square pile production parameter data in real time and form a set of square pile production parameters;

[0139] The coupling relationship analysis module 102 is used to analyze the coupling relationship between the set of square pile production parameters using a pre-trained neural network model, and generate a coupling strength matrix.

[0140] The lag feature processing module 103 is used to extract lag effect features in the coupling strength matrix using an improved K-means clustering algorithm, and to process the lag effect features using a Kalman filter algorithm to generate a parameter prediction adjustment vector.

[0141] Deviation calculation module 104 is used to calculate the deviation between the parameter prediction adjustment vector and the preset target production parameter set, and generate prediction deviation data;

[0142] The adjustment instruction generation module 105 is used to generate adjustment instructions for square pile production parameters based on the predicted deviation data and by calling the preset parameter adjustment rule library.

[0143] In one embodiment, the coupling analysis module 102 is further configured to:

[0144] The set of square pile production parameters is time-series aligned and standardized to form a standardized parameter sequence.

[0145] The standardized parameter sequence is input into a pre-trained neural network model to calculate the attention weights of different production parameters in the standardized parameter sequence at a preset time step.

[0146] Based on attention weights, a hybrid metric method based on mutual information and cosine similarity is used to calculate the correlation strength between any two production parameters in the standardized parameter sequence, and generate a correlation strength sequence.

[0147] Perform temporal convolutional fusion operations on the correlation strength sequences to extract multi-scale temporal features;

[0148] Pooling operations are performed on multi-scale temporal features to generate a coupling strength matrix.

[0149] In one embodiment, the hysteresis feature processing module 103 is further configured to:

[0150] An adaptive sliding window is used to dynamically segment the coupling strength matrix over time, generating a sequence of time-series matrix slices.

[0151] An improved K-means clustering algorithm was used to perform cluster analysis on the time series matrix slice sequences, extract the cluster centers of each category, and form the dominant lag pattern sequences;

[0152] The dominant hysteresis mode sequence is input into a Kalman filter for state estimation, resulting in a sequence of predicted parameter adjustments.

[0153] A spatial mapping transformation is performed on the predicted value sequence to generate a parameter prediction adjustment vector.

[0154] In one embodiment, the deviation calculation module 104 is further configured to:

[0155] Based on the dynamic time warping algorithm, the time dimension of the alignment parameter prediction adjustment vector and the preset target production parameter set is adjusted to generate a vector sequence;

[0156] A multimodal deviation fusion method is adopted to calculate the joint deviation of the time-aligned vector sequence in the time domain and frequency domain, and generate a multimodal deviation matrix.

[0157] Principal component analysis was performed on the multimodal deviation matrix to extract the dominant deviation eigenvectors.

[0158] Based on the dominant bias eigenvector, significance assessment based on Mahalanobis distance is used to generate prediction bias data.

[0159] In one embodiment, the adjustment instruction generation module 105 is further configured to:

[0160] Based on the dominant deviation feature vector, an empirical probability distribution P is constructed, and a reference probability distribution Q of the preset target production parameter set is obtained;

[0161] The following formula can be used to calculate the relationship between the empirical probability distribution P and the reference probability distribution Q. divergence :

[0162]

[0163] in, For preset adjustment parameters, and Let be the mean vector and covariance matrix of the empirical probability distribution P, respectively. and These are the mean vector and covariance matrix of the reference probability distribution Q, respectively. It is the determinant of a matrix;

[0164] The calculated divergence As prediction bias data.

[0165] In one embodiment, the adjustment instruction generation module 105 is further configured to:

[0166] Based on prediction deviation data, generate rule-based query conditions;

[0167] The rule query conditions are matched with the rules in the preset parameter adjustment rule base to generate a matching rule set;

[0168] Parse the adjustment logic and constraints in the matching rule set to generate parameter adjustment constraints;

[0169] Based on parameter adjustment constraints and prediction deviation data, an optimization function is constructed with the objective of minimizing prediction deviation and adjustment magnitude.

[0170] Solve the optimization function to obtain the parameter adjustment solution vector;

[0171] According to the preset vector-instruction mapping rules, the parameter adjustment solution vector is mapped to the square pile production parameter adjustment instruction.

[0172] In one embodiment, such as Figure 3 As shown, it also includes a feedback optimization module 106, used for:

[0173] Acquire square pile production parameter data within a preset acquisition period after executing the square pile production parameter adjustment command;

[0174] Calculate the actual deviation between the square pile production parameter data within the preset collection period and the preset target production parameter set;

[0175] If the actual deviation data exceeds the preset deviation threshold, then the actual deviation data, the corresponding square pile production parameter adjustment instructions, and the square pile production parameter data within the preset collection period will be used as samples to construct an incremental training sample set.

[0176] Based on the incremental training sample set, the online sequence extreme learning machine algorithm is used to incrementally update the pre-trained neural network model to obtain the updated neural network model.

[0177] Select historical square pile production parameter data within a preset time period before the execution of the square pile production parameter adjustment command, input it into the updated neural network model for re-prediction, and generate a set of model prediction outputs.

[0178] The mean, variance, and deviation fluctuations of the predicted output set of the computational model are used as statistical characteristics.

[0179] Based on statistical characteristics, a threshold iterative optimization method is adopted to adjust the preset parameters and the rule thresholds corresponding to each rule in the rule base.

[0180] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the dynamic adjustment method for square pile production parameters as described above.

[0181] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0182] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0183] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for dynamically adjusting square pile production parameters, characterized in that, The method comprises: Real-time acquisition of square pile production parameter data forms a square pile production parameter set; Using a pre-trained neural network model to analyze the coupling relationship between the square pile production parameter set, a coupling strength matrix is generated; Using an improved K-means clustering algorithm to extract the hysteresis effect features in the coupling strength matrix, and using a Kalman filtering algorithm to process the hysteresis effect features, a parameter prediction adjustment vector is generated; Calculate the deviation between the parameter prediction adjustment vector and the preset target production parameter set, and generate prediction deviation data; Based on the prediction deviation data, call the preset parameter adjustment rule library to generate square pile production parameter adjustment instructions.

2. The method of claim 1, wherein, The method comprises: The square pile production parameter set is time-aligned and standardized to form a standardized parameter sequence; The standardized parameter sequence is input into the pre-trained neural network model, and the attention weight of different production parameters in the standardized parameter sequence at a preset time step is calculated; According to the attention weight, using a hybrid measurement method based on mutual information and cosine similarity, the correlation strength between any two production parameters in the standardized parameter sequence is calculated, and a correlation strength sequence is generated; Convolution fusion operation is performed on the correlation strength sequence in the time dimension to extract multi-scale time sequence features; The multi-scale time sequence features are pooled to generate the coupling strength matrix.

3. The method of claim 1, wherein, The method comprises: Using an adaptive sliding window to dynamically time segment the coupling strength matrix to generate a time matrix slice sequence; Using the improved K-means clustering algorithm to cluster analyze the time matrix slice sequence, extract the cluster centers of each category, and form a dominant lag mode sequence; The dominant lag mode sequence is input into the Kalman filter for state estimation to obtain a predicted value sequence of the parameter adjustment amount; The predicted value sequence is spatially mapped and transformed to generate the parameter prediction adjustment vector.

4. The method of claim 1, wherein, The method comprises: Aligning the time sequence dimensions of the parameter prediction adjustment vector and the preset target production parameter set based on the dynamic time warping algorithm to generate a vector sequence; Using a multi-modal deviation fusion method, the joint deviation of the time-aligned vector sequence in the time and frequency domains is calculated to generate a multi-modal deviation matrix; Principal component analysis is performed on the multi-modal deviation matrix to extract a dominant deviation feature vector; According to the dominant deviation feature vector, using a significance evaluation based on Mahalanobis distance, the prediction deviation data is generated.

5. The method of claim 4, wherein, The method comprises: construct an empirical probability distribution P based on the dominant deviation feature vector, and obtain a reference probability distribution Q of the preset target production parameter set; The empirical probability distribution P is calculated by the following equation: divergence : wherein is a preset adjustment parameter, and are a mean vector and a covariance matrix of the empirical probability distribution P, respectively, and are a mean vector and a covariance matrix of the reference probability distribution Q, respectively, is a matrix determinant; The calculated divergence as the predicted bias data.

6. The method of claim 1, wherein, The method comprises the following steps of: Based on the predicted deviation data, a preset parameter adjustment rule library is called to generate a square pile production parameter adjustment instruction, which comprises: Based on the predicted deviation data, a rule query condition is generated; The rule query condition is matched with the rules in the preset parameter adjustment rule library to generate a matching rule set; The adjustment logic and constraint conditions in the matching rule set are analyzed to generate a parameter adjustment constraint; Based on the parameter adjustment constraint and the predicted deviation data, an optimization function is constructed to minimize the predicted deviation and the adjustment amplitude; The optimization function is solved to obtain a parameter adjustment solution vector; 7. The method of claim 1, wherein, According to a preset vector-instruction mapping rule, the parameter adjustment solution vector is mapped to the square pile production parameter adjustment instruction. After the square pile production parameter adjustment instruction is generated, the method further comprises the following steps of: Obtaining square pile production parameter data in a preset collection period after the square pile production parameter adjustment instruction is executed; Calculating the actual deviation data between the square pile production parameter data in the preset collection period and the preset target production parameter set; If the actual deviation data exceeds a preset deviation threshold, an incremental training sample set is constructed based on the actual deviation data, the corresponding square pile production parameter adjustment instruction and the square pile production parameter data in the preset collection period; Based on the incremental training sample set, an online sequence extreme learning machine algorithm is used to incrementally update the pre-trained neural network model to obtain an updated neural network model; The historical square pile production parameter data in a preset time period before the execution of the square pile production parameter adjustment instruction is selected and input into the updated neural network model for re-prediction to generate a model prediction output set; The mean, variance and deviation fluctuation amplitude of the model prediction output set are calculated as statistical characteristics; 8. A square pile production parameter dynamic adjustment device, characterized in that, According to the statistical characteristics, a threshold iterative optimization method is used to adjust the rule threshold values corresponding to each rule in the preset parameter adjustment rule library. The device comprises: A data acquisition module is configured to acquire square pile production parameter data in real time to form a square pile production parameter set; A coupling relationship analysis module is configured to analyze the coupling relationship between the square pile production parameter set by using a pre-trained neural network model to generate a coupling strength matrix; A lag feature processing module is configured to extract lag effect features from the coupling strength matrix by using an improved K-means clustering algorithm, and process the lag effect features by using a Kalman filter algorithm to generate a parameter prediction adjustment vector; A deviation calculation module is configured to calculate the deviation between the parameter prediction adjustment vector and a preset target production parameter set to generate predicted deviation data; 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. An adjustment instruction generation module is configured to call a preset parameter adjustment rule library based on the predicted deviation data to generate a square pile production parameter adjustment instruction. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 7. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, which is executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images