Settlement prediction method and system fused with multi-scale optimization, terminal and storage medium

Abstract

The invention provides a multi-scale optimization fused settlement prediction method and system, a terminal and a storage medium, and the method comprises the steps: carrying out the complete ensemble empirical mode decomposition of settlement monitoring data, and obtaining an intrinsic mode function; performing complexity evaluation on the intrinsic mode function to obtain a complex evaluation value, and clustering the intrinsic mode function according to the complex evaluation value to obtain a high-frequency component, an intermediate-frequency component and a low-frequency component; performing variational mode decomposition on the high-frequency component to obtain a high-frequency decomposition component, and inputting the high-frequency decomposition component, the intermediate-frequency component and the low-frequency component into a pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value; and generating a settlement prediction result according to the settlement prediction value. According to the embodiment of the invention, the complete ensemble empirical mode decomposition and the variational mode decomposition are adopted to carry out dual-scale decomposition, so that the noise in the signal can be accurately and effectively removed, the interference of the noise on the settlement prediction result is reduced, and the settlement prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of settlement prediction, and particularly to a settlement prediction method, system, terminal and storage medium integrating multi-scale optimization. Background Art

[0002] Uneven settlement of buildings or structures is one of the main reasons for damage to local components or the overall structure of buildings or structures, which may cause serious safety accidents. At the same time, if the uneven settlement of buildings or structures cannot be sensed and predicted in advance, the reinforcement and repair after the occurrence of component damage may be complex and expensive. Improving the prediction accuracy of settlement data in different scenarios has important practical significance for preventing and reducing the losses caused by uneven settlement.

[0003] In the existing settlement prediction process, generally software such as MATLAB is used for modeling and analysis. However, due to the large amount of noise data in the modeling data, the accuracy of the settlement prediction results obtained by modeling and analysis is low. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a settlement prediction method, system, terminal and storage medium integrating multi-scale optimization to solve the problem of low accuracy of settlement prediction in the prior art.

[0005] The embodiments of the present invention are implemented as follows. A settlement prediction method integrating multi-scale optimization, the method includes: Obtain settlement monitoring data, and perform complete ensemble empirical mode decomposition on the settlement monitoring data to obtain intrinsic mode functions; Evaluate the complexity of the intrinsic mode functions to obtain a complexity evaluation value, and cluster the intrinsic mode functions according to the complexity evaluation value to obtain high-frequency components, medium-frequency components and low-frequency components; Perform variational mode decomposition on the high-frequency components to obtain high-frequency decomposition components, and input the high-frequency decomposition components, the medium-frequency components and the low-frequency components into a pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value; Generate a settlement prediction result according to the settlement prediction value.

[0006] Preferably, performing complete ensemble empirical mode decomposition on the settlement monitoring data to obtain intrinsic mode functions includes: Add white noise to the settlement monitoring data to obtain a noise signal, and decompose the noise signal to obtain the intrinsic mode functions; The formula for adding white noise to the settlement monitoring data includes: Where is the white noise, is the settlement monitoring data, is the noise signal; The formula for decomposing the noise signal includes: where, is the i th intrinsic mode function, is the residual term, N represents the number of the noise signals.

[0007] Preferably, to evaluate the complexity of the intrinsic mode function to obtain a complexity evaluation value, it includes: Perform multi-scale coarse-graining processing on the time series corresponding to the intrinsic mode function to obtain a coarse-grained sequence, and construct a delay vector based on the embedding dimension and delay time for the coarse-grained sequence to obtain a delay vector; Sort the delay vector and record the permutation pattern, and calculate the permutation entropy based on the recorded result of the permutation pattern to obtain the complexity evaluation value.

[0008] Preferably, before inputting the high-frequency decomposition component, the intermediate-frequency component and the low-frequency component into the pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value, it further includes: Obtain a sample component, and input the sample component into the settlement prediction model for convolution processing to obtain a sample convolution feature; Perform batch normalization processing on the sample convolution feature to obtain a sample normalized feature, and perform non-linear transformation on the sample normalized feature to obtain a non-linear feature; Perform pooling processing on the non-linear feature to obtain a pooling feature, and perform bidirectional gated recurrent processing on the pooling feature to obtain a gated feature; Flip the gated feature to obtain a flipped feature, and perform feature splicing on the flipped feature to obtain a spliced feature; Perform fully connected processing on the spliced feature to obtain a fully connected feature, and optimize the parameters of the settlement prediction model according to the fully connected feature until the settlement prediction model converges to obtain the pre-trained settlement prediction model.

[0009] Preferably, optimizing the parameters of the settlement prediction model according to the fully connected feature includes: Obtain the model parameters of the settlement prediction model, and generate a dung beetle population according to the model parameters; Calculate the fitness value of the dung beetle population according to the fully connected feature, and update the current optimal position of the dung beetle population according to the fitness value. During the unobstructed ball-rolling stage of the dung beetle, the position of the dung beetle is updated using the first formula, and when an obstacle is encountered during the ball-rolling process, the position of the dung beetle is updated using the second formula; The exploration dung beetle is guided to update its position according to the adaptive convergence factor formula, and the position of the reproductive dung beetle is updated using the third formula; The position of the foraging dung beetle is updated using the fourth formula, and the position of the stealing dung beetle is updated using the fifth formula; For the said dung beetle population, the current optimal position is calculated based on the updated position of the dung beetle to obtain the optimal updated position. When the number of iterations of the dung beetle population is equal to the iteration threshold, the optimal updated position of the current iteration is output; The parameters of the settlement prediction model are set according to the output optimal updated position to obtain the pre-trained settlement prediction model.

[0010] Preferably, the first formula used is: Where, is the position of the i th osprey in the j th dimension after updating the position of the dung beetle using the first formula, is the current position of the i th osprey in the j th dimension, is a random factor, is a step factor, is the inertia weight or influence factor; The second formula used is: Where, represents the position of the i th dung beetle at the t th iteration, represents the ball-rolling direction, represents the i th dung beetle at the t -1th iteration; The adaptive convergence factor formula is: Where R is the convergence factor and t is the current iteration number.

[0011] Preferably, the third formula used is: Where, represents the position of the i th egg at the t th iteration,b 1 and b 2 are independent random vectors, represents the lower limit of the spawning area, represents the upper limit of the spawning area, represents the current local best position, represents the i th egg's position at the t -1th iteration; The fourth formula used is: where, represents the T-distribution mutation perturbation with the iteration number as the degree-of-freedom parameter, is the position of the jth dimension of the dung beetle after updating using the fourth formula by the foraging dung beetle, is the current optimal position on the j th dimension; The fifth formula used is: where S is a constant, g is a random vector obeying the normal distribution, is the position of the i th individual after updating using the fifth formula at t +1 moment, is the t moment's global optimal position, is the t moment's local optimal position, is the position of the i th individual at t moment.

[0012] Another object of the embodiments of the present invention is to provide a settlement prediction system integrating multi-scale optimization. The system includes: A modal decomposition module, configured to obtain settlement monitoring data and perform complete ensemble empirical modal decomposition on the settlement monitoring data to obtain intrinsic mode functions; A clustering module, configured to evaluate the complexity of the intrinsic mode functions to obtain a complexity evaluation value, and cluster the intrinsic mode functions according to the complexity evaluation value to obtain high-frequency components, medium-frequency components, and low-frequency components; A settlement prediction module, configured to perform variational modal decomposition on the high-frequency components to obtain high-frequency decomposition components, and input the high-frequency decomposition components, the medium-frequency components, and the low-frequency components into a pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value; A prediction output module, configured to generate a settlement prediction result according to the settlement prediction value.

[0013] In an embodiment of the present invention, by performing complete ensemble empirical mode decomposition on the settlement monitoring data, noise can be effectively removed and stable intrinsic mode functions can be extracted. By calculating the complexity evaluation value of the intrinsic mode function, the complexity of the intrinsic mode function can be effectively evaluated. By clustering the intrinsic mode function based on the complexity evaluation value, the intrinsic mode function can be effectively classified according to complexity, obtaining high-frequency components, medium-frequency components, and low-frequency components. By performing dual-scale decomposition using complete ensemble empirical mode decomposition and variational mode decomposition, noise in the signal can be accurately and effectively removed, reducing the interference of noise on the settlement prediction result and improving the accuracy of settlement prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flowchart of a settlement prediction method with multi-scale optimization fusion provided by the first embodiment of the present invention; Figure 2 is a schematic structural diagram of a settlement prediction system with multi-scale optimization fusion provided by the second embodiment of the present invention; Figure 3 is a schematic structural diagram of a terminal device provided by the third embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0016] In order to illustrate the technical solutions described in the present invention, the following will be described through specific embodiments.

[0017] Embodiment 1 Please refer to Figure 1 , which is a flowchart of a settlement prediction method with multi-scale optimization fusion provided by the first embodiment of the present invention. The settlement prediction method with multi-scale optimization fusion can be applied to any device or system. The settlement prediction method with multi-scale optimization fusion includes the following steps: Step S10: Obtain settlement monitoring data, and perform complete ensemble empirical mode decomposition on the settlement monitoring data to obtain intrinsic mode functions; Among them, missing value filling and normalization processing are performed on the settlement monitoring data to ensure data quality and unify the dimension. By performing complete ensemble empirical mode decomposition on the settlement monitoring data, noise can be effectively removed and stable intrinsic mode functions can be extracted, facilitating further processing of each frequency component of the signal.

[0018] Optionally, performing complete ensemble empirical mode decomposition on the settlement monitoring data to obtain intrinsic mode functions includes: Add white noise to the settlement monitoring data to obtain a noise signal, and decompose the noise signal to obtain the intrinsic mode functions. Specifically, in each decomposition, different white noises are added to the settlement monitoring data to obtain a noise signal. This white noise is Gaussian-distributed noise. By performing complete ensemble empirical mode decomposition on the noise signal, the intrinsic mode functions are obtained. Perform an averaging process on the intrinsic mode functions to obtain the intrinsic mode functions. The formula for adding white noise to the settlement monitoring data includes: Where is the white noise, is the settlement monitoring data, is the noise signal; The formula for decomposing the noise signal includes: Where is the i th intrinsic mode function, is the residual term, N represents the number of the noise signals; Preferably, in this step, an averaging process can also be performed on the intrinsic mode functions. The formula for the averaging process includes: Where is the m th intrinsic mode function obtained after the averaging process for the i th decomposition, M is the number of times of adding noise to the settlement monitoring data.

[0019] In this step, by introducing multiple reconstructions and noise weighting, the signal decomposition accuracy is improved, especially in dealing with non-stationary signals, and a stable decomposition result of the intrinsic mode functions can be obtained. Using the complete ensemble empirical mode decomposition with adaptive noise to decompose the varying signal, it suppresses the mode mixing problem by introducing an adaptive noise, ensuring the reliability of the intrinsic mode function components.

[0020] Step S20: Evaluate the complexity of the intrinsic mode functions to obtain a complexity evaluation value, and cluster the intrinsic mode functions according to the complexity evaluation value to obtain high-frequency components, medium-frequency components, and low-frequency components. Among them, by calculating the complexity evaluation values of each intrinsic mode function, the complexity of each intrinsic mode function can be effectively evaluated. By clustering the intrinsic mode functions according to the complexity evaluation values, the complexity classification of the intrinsic mode functions can be effectively carried out. In this step, by calculating the sample entropy and fuzzy entropy for each intrinsic mode function respectively, the complexity evaluation values are obtained, and the intrinsic mode functions are divided into high-frequency components (including details and noise), medium-frequency components and low-frequency components by using the K-means clustering algorithm.

[0021] Optionally, evaluating the complexity of the intrinsic mode function to obtain a complexity evaluation value includes: Performing multi-scale coarse-graining processing on the time series corresponding to the intrinsic mode function to obtain a coarse-grained sequence, and constructing a delay vector according to the embedding dimension and delay time for the coarse-grained sequence; wherein, performing multi-scale coarse-graining processing on the time series corresponding to the intrinsic mode function means dividing the time series into multiple non-overlapping windows and calculating the window mean under the scale factor to generate a coarse-grained sequence. For a time series, select a scale factor τ , and construct a coarse-grained sequence: wherein, represents the j th coarse-grained sequence, x i represents the i th element in the original time series; Set the embedding dimension m and delay time d, and construct a delay vector v of length m for the coarse-grained sequence i : Sort the delay vector and record the permutation pattern, and calculate the permutation entropy according to the recorded result of the permutation pattern to obtain the complexity evaluation value; Among them, sort each delay vector and record its permutation pattern (Permutation Pattern). Denote the probability of π appearing in each pattern as p ( π ).

[0022] wherein, PE ( m ) represents the calculated permutation entropy, τ is the scale factor, indicating the degree of coarse-graining of the time series. Different scale factors correspond to different time scales; The multi-scale permutation entropy is defined as the set of permutation entropies under different scale factors τ : Among them, MPE is the multi-scale permutation entropy, is the time series after coarse-graining. The coarse-graining process is achieved by dividing the original time series into non-overlapping windows and calculating the average value within each window. is the permutation entropy at the scale factor τ which is used to quantify the complexity of the time series.

[0023] The multi-scale permutation entropy (MPE) is used to evaluate the complexity of each IMF component, and K-means is used for clustering to obtain high-frequency components, medium-frequency components, and low-frequency components.

[0024] Step S30: Perform variational mode decomposition on the high-frequency components to obtain high-frequency decomposition components, and input the high-frequency decomposition components, the medium-frequency components, and the low-frequency components into the pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value; Among them, since the high-frequency components may contain noise and short-term perturbations, variational mode decomposition (VMD) is used to decompose the high-frequency components twice to enhance the ability to extract key information and suppress residual noise, so as to improve the data structure analysis ability while maintaining the essential characteristics of the signal, providing high-quality input for the subsequent steps. The principle of the VMD method is to determine the center frequency and bandwidth of the decomposition components through iterative search, which is an adaptive and completely non-recursive signal processing method. As part of the VMD process, the high-frequency component x(t) is decomposed into K modal signals uk(t) (k = 1, 2,..., K). The spectral distribution of the modal signals is concentrated in a specific frequency range. In addition, the modal signals are orthogonal to each other, which means that the frequency domain is fully separated, and the obtained high-frequency decomposition components have non-overlapping frequency bands, improving the separability of signal features.

[0025] VMD can finely separate the frequency components in the signal. Especially for signals with complex frequency components, the variational method is used to optimize the frequency band of each mode. By adjusting the bandwidth parameter, the signal can be adaptively decomposed according to the frequency characteristics of the signal, avoiding over- or under-decomposition of the frequency band, so as to effectively extract the fine features of each frequency band.

[0026] In this embodiment, a settlement prediction model is constructed using a CNN-BiGRU structure. The CNN network is used to perform convolutional processing on each decomposed component sequence to extract local temporal features, and then connected to the BiGRU network for bidirectional temporal modeling to capture historical and future dependence information. The Flatten layer is used to connect the output of the CNN to the BiGRU to form a spatio-temporal feature fusion path. CNN is good at capturing local features, especially short-term dependencies and local temporal patterns. Through convolutional operations, CNN can effectively capture local changes such as periodic fluctuations, spikes, and valleys in signals. BiGRU is an improvement of GRU. Through a bidirectional recursive network structure, it can extract temporal information from both the forward and backward directions simultaneously. Compared with unidirectional GRU or LSTM, it can capture historical and future information in the sequence more comprehensively, and due to fewer parameters, the calculation speed is faster. Combining CNN and BiGRU can more effectively extract local and global features in temporal data, thereby improving prediction accuracy.

[0027] Optionally, before inputting the high-frequency decomposition component, the intermediate-frequency component, and the low-frequency component into the pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value, it further includes: Obtain a sample component, and input the sample component into the settlement prediction model for convolutional processing to obtain sample convolutional features; Perform batch normalization processing on the sample convolutional features to obtain sample normalized features, and perform non-linear transformation on the sample normalized features to obtain non-linear features; Perform pooling processing on the non-linear features to obtain pooling features, and perform bidirectional gated recurrent processing on the pooling features to obtain gated features; Flip the gated features to obtain flipped features, and perform feature splicing on the flipped features to obtain spliced features; Perform fully connected processing on the spliced features to obtain fully connected features, and optimize the parameters of the settlement prediction model according to the fully connected features until the settlement prediction model converges to obtain the pre-trained settlement prediction model; In this embodiment, the sample component first enters the CNN layer for convolution operation (conv_1) to extract the features of the input data, and then performs batch normalization (bn_1) on the output of the convolutional layer to accelerate the training process and improve the stability of the model. Then, the ReLU activation function (rule_1) is applied to introduce non-linearity, enabling the model to learn more complex features. After that, a pooling operation (maxpool_1) is performed to reduce the dimension of the data while retaining important features. The multi-dimensional data is converted into one-dimensional (flatten) for input into the fully connected layer. The output of the GRU layer is flipped (flip) and gated (gru / gru_1), which may be used to change the order of the data or perform other processing. Then, the output of the GRU layer after flipping and gating is concatenated (concat) with other data to form a new feature vector. Finally, the output of the concatenation layer is input into the fully connected layer (fc) for final feature integration and classification, generating the final output (output) of the settlement prediction model.

[0028] Further, parameter optimization of the settlement prediction model is performed according to the fully connected features, including: Obtain the model parameters of the settlement prediction model and generate a dung beetle population according to the model parameters; Calculate the fitness value of the dung beetle population according to the fully connected features and update the current optimal position of the dung beetle population according to the fitness value; During the stage when the dung beetle rolls the ball without obstacles, update the position of the dung beetle using the first formula, and when encountering an obstacle during the ball-rolling process, update the position of the dung beetle using the second formula; Among them, the global exploration strategy of the osprey optimization algorithm in the first stage of the dung beetle population is used to replace the position update formula of the ball-rolling stage of the original dung beetle algorithm. The global exploration strategy of the osprey optimization algorithm is introduced during the stage when the dung beetle rolls the ball without obstacles. Since there are many parameters involved in the update strategy of the ball-rolling dung beetle in the original dung beetle optimization algorithm, and the dung beetles are isolated from each other without information exchange, the optimization algorithm converges prematurely and falls into a local optimal solution without being able to find the global optimal solution. The global exploration strategy formula (the first formula) of the osprey optimization algorithm in the first stage is used to replace the position update formula of the ball-rolling stage of the original dung beetle algorithm. The global exploration strategy formula (the first formula) of the osprey optimization algorithm in the first stage is as follows: Among them, is the position of the i th osprey updated using the first formula in the j th dimension, is the current position of the i th osprey in the j th dimension, is a random factor, is the step factor, is the inertia weight or influence factor.

[0029] When encountering an obstacle during the process of rolling the ball, the dung beetle will re-determine the ball-rolling direction by dancing, and the dancing behavior is completed through the tangent function. The ball-rolling direction is within (0, π ). After the ball-rolling direction is re-determined, the dung beetle will continue to roll the dung ball. At this time, its position update formula (the second formula) is as follows: where represents the position of the i th dung beetle at the t th iteration, represents the ball-rolling direction, represents the i th dung beetle at the t -1th iteration.

[0030] When = 0, π / 2 , π , tan is 0 or does not exist, and the position of the dung beetle is not updated.

[0031] According to the adaptive convergence factor formula, guide the exploration of the dung beetle to update its position, and use the third formula to update the position of the reproductive dung beetle; among them, in this embodiment, the dung beetle optimization algorithm (ZTDBO) algorithm that fuses the scaling factor and the adaptive t-distribution is used for hyperparameter optimization.

[0032] In the original dung beetle reproduction behavior, linear decay may cause the weight to drop to a relatively low level prematurely, which means that the algorithm may start to focus on the detailed search in the local area without fully exploring the entire search space. Insufficient exploration may lead to suboptimal solutions. Secondly, the gradual neglect of boundary individuals may lead to missing the opportunity to identify better solutions located near the edge or outside the search space. Finally, the predictability and certainty of the linear decay formula limit the adaptability of the algorithm to dynamically respond to newly discovered promising areas. To address these limitations, a Nonlinear Dynamic Adjustment Factor is proposed, which introduces both nonlinear decay and random perturbation to enhance the flexibility and adaptability of the algorithm. The specific adaptive convergence factor formula is as follows: Among them, R is the convergence factor and t is the current number of iterations. Compared with the original linear attenuation, this non-linear strategy ensures that R does not decrease monotonically or predictably. By incorporating randomness, the non-linear dynamic adjustment factor ensures that the reduction of weights is no longer monotonic or predictable, thus enhancing the adaptability of the algorithm to different optimization stages.

[0033] A suitable spawning site for the reproductive behavior is very important for the dung beetle to breed offspring. Therefore, the dung beetle will roll the dung ball to a safe and suitable area for spawning and hide it. Based on this behavior of the dung beetle, a boundary selection strategy is given to simulate the spawning area of the reproductive dung beetle: and represent the lower and upper limits of the spawning area respectively. is the current local best position, R = 1 - t / T is the linear convergence factor, T is the maximum number of iterations, Lb and Ub are the lower and upper limits of the problem to be optimized respectively; When the spawning area is determined, the reproductive dung beetle will start to lay eggs, and only one egg is laid in each iteration. Since its spawning area is dynamically changing, the position of the egg is also dynamically updated. At this time, its position update formula (the third formula) is as follows: Among them, represents the position of the i th egg at the t th iteration, b 1 and b 2 are independent random vectors, represents the lower limit of the spawning area, represents the upper limit of the spawning area, represents the current local best position, represents the i th egg at the t -1th iteration.

[0034] Update the position of the foraging dung beetle using the fourth formula and update the position of the stealing dung beetle using the fifth formula; Among them, when the larva hatches from the egg and grows into a small dung beetle, it usually comes out to forage; During the foraging stage of the dung beetle, a distribution perturbation is imposed on the foraging behavior of the small dung beetle. To further enhance the balance ability of the dung beetle optimization algorithm between global exploration and local exploitation, an adaptive T-distribution perturbation is introduced during the foraging stage of the small dung beetle. is the number of iterations The T-distribution variation perturbation with the degree of freedom parameter is used to perturb the foraging behavior of the small dung beetles. At the beginning of the iteration, the T-distribution perturbation is similar to the Cauchy variation, making the dung beetle algorithm have a strong global exploration ability at this time. In the later stage of the iteration, the T-distribution perturbation is similar to the Gaussian variation. At this time, the local development ability of the algorithm is more prominent, and the convergence speed of the algorithm is also improved. The update method of the new position (the fourth formula) is as follows: Among them, represents the T-distribution variation perturbation with the iteration number as the degree of freedom parameter, is the position of the j-th dimension of the foraging dung beetle after updating using the fourth formula, is the j current optimal position on the

[0035] There is a behavior of stealing dung balls in the dung beetle population. Its position update formula (the fifth formula) is as follows: Among them, S is a constant, g is a random vector obeying the normal distribution, is the i position of the i-th individual after updating using the fifth formula t at time +1, t is the global optimal position at time is the t local optimal position at time is the i position of the i-th individual at time t . In the population, the ratio of ball-rolling dung beetles, breeding dung beetles, foraging dung beetles, and stealing dung beetles is 6:6:7:11.

[0036] For the said dung beetle population, calculate the current optimal position according to the updated position of the dung beetles, obtain the optimal updated position, and until the iteration number of the said dung beetle population is equal to the iteration threshold, output the optimal updated position of the current iteration; Set the parameters of the settlement prediction model according to the output optimal updated position to obtain the pre-trained settlement prediction model.

[0037] In this embodiment, by introducing a non-linear dynamic adjustment factor, the flexibility and adaptability of the algorithm are significantly enhanced. At the same time, by introducing the exploration strategy of the osprey optimization algorithm, the global search ability is improved. By introducing the adaptive T-distribution perturbation, the ability of the algorithm to jump out of the local optimum is further improved.

[0038] Step S40, generate a settlement prediction result according to the said settlement prediction value; Among them, the complete ensemble empirical mode decomposition is performed on the settlement monitoring data to obtain the intrinsic mode functions ( , ). The complexity of the intrinsic mode functions is evaluated (Calculate SE) to obtain the complexity evaluation values. Based on the complexity evaluation values and the K-means clustering algorithm, the intrinsic mode functions are clustered to obtain high-frequency components (Co- ), medium-frequency components (Co- ), and low-frequency components (Co- ). The Co- is decomposed by VMD to obtain high-frequency decomposition components ( , ). The high-frequency decomposition components, medium-frequency components, and low-frequency components are input into the pre-trained settlement prediction model (CNN-BiGRU), and the settlement prediction is performed based on the Attention mechanism to obtain the settlement prediction values. The prediction results of each IMF are fused to obtain the settlement prediction result. In the settlement prediction result, the average value of the settlement prediction values is calculated to obtain the predicted average value. If the predicted average value is greater than the average threshold, it is determined that the object corresponding to the settlement monitoring data has a settlement phenomenon.

[0039] In this embodiment, by performing the complete ensemble empirical mode decomposition on the settlement monitoring data, the noise can be effectively removed and the stable intrinsic mode functions can be extracted. By calculating the complexity evaluation values of the intrinsic mode functions, the complexity of the intrinsic mode functions can be effectively evaluated. By clustering the intrinsic mode functions based on the complexity evaluation values, the complexity classification of the intrinsic mode functions can be effectively performed to obtain high-frequency components, medium-frequency components, and low-frequency components. By adopting the complete ensemble empirical mode decomposition and variational mode decomposition for double-scale decomposition, the noise in the signal can be accurately and effectively removed, the interference of the noise on the settlement prediction result is reduced, and the accuracy of the settlement prediction is improved.

[0040] Embodiment 2 Please refer to Figure 2 , which is a schematic structural diagram of the settlement prediction system 100 integrating multi-scale optimization provided by the second embodiment of the present invention, including: The modal decomposition module 10 is used to obtain the settlement monitoring data and perform the complete ensemble empirical mode decomposition on the settlement monitoring data to obtain the intrinsic mode functions.

[0041] Optionally, the modal decomposition module 10 is further used for: adding white noise to the settlement monitoring data to obtain a noise signal, and decomposing the noise signal to obtain the intrinsic mode functions; The formula for adding white noise to the settlement monitoring data includes: Among them, is the white noise, is the settlement monitoring data, is the noise signal; The formula for decomposing the noise signal includes: where is the i th intrinsic mode function, is the residual term.

[0042] The clustering module 11 is used to evaluate the complexity of the intrinsic mode function to obtain a complexity evaluation value, and cluster the intrinsic mode function according to the complexity evaluation value to obtain high-frequency components, medium-frequency components, and low-frequency components.

[0043] Optionally, the clustering module 11 is further used to: perform multi-scale coarse-graining processing on the time series corresponding to the intrinsic mode function to obtain a coarse-grained sequence, and construct a delay vector according to the embedding dimension and delay time for the coarse-grained sequence; Sort the delay vectors and record the permutation pattern, and calculate the permutation entropy according to the permutation pattern record result to obtain the complexity evaluation value.

[0044] The settlement prediction module 12 is used to perform variational mode decomposition on the high-frequency components to obtain high-frequency decomposition components, and input the high-frequency decomposition components, the medium-frequency components, and the low-frequency components into a pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value.

[0045] Optionally, the settlement prediction module 12 is further used to: obtain sample components, and input the sample components into the settlement prediction model for convolution processing to obtain sample convolution features; Perform batch normalization processing on the sample convolution features to obtain sample normalized features, and perform non-linear transformation on the sample normalized features to obtain non-linear features; Perform pooling processing on the non-linear features to obtain pooling features, and perform bidirectional gated recurrent processing on the pooling features to obtain gated features; Flip the gated features to obtain flipped features, and perform feature splicing on the flipped features to obtain spliced features; Perform fully connected processing on the spliced features to obtain fully connected features, and optimize the parameters of the settlement prediction model according to the fully connected features until the settlement prediction model converges to obtain the pre-trained settlement prediction model.

[0046] Further, the settlement prediction module 12 is further configured to: obtain the model parameters of the settlement prediction model, and generate a dung beetle population according to the model parameters; Calculate the fitness value of the dung beetle population according to the fully connected features, and update the current optimal position of the dung beetle population according to the fitness value; In the stage where the dung beetle rolls the ball without obstacles, update the position of the dung beetle using the first formula, and when encountering an obstacle during the ball rolling process, update the position of the dung beetle using the second formula; Guide the exploration dung beetle to update the position according to the adaptive convergence factor formula, and use the third formula to update the position of the reproductive dung beetle; Use the fourth formula to update the position of the foraging dung beetle, and use the fifth formula to update the position of the stealing dung beetle; For the dung beetle population, calculate the current optimal position according to the updated position of the dung beetle to obtain the optimal updated position. When the iteration times of the dung beetle population is equal to the iteration threshold, output the optimal updated position of the current iteration; Set the parameters of the settlement prediction model according to the output optimal updated position to obtain the pre-trained settlement prediction model.

[0047] The first formula used is: Where, is the position of the i th osprey updated by the first formula in the j th dimension, is the current position of the i th osprey in the j th dimension, is a random factor, is a step size factor, is an inertia weight or influence factor; The second formula used is: Where, represents the position of the i th dung beetle at the t th iteration, represents the ball rolling direction; The adaptive convergence factor formula is: Where, R is the convergence factor and t is the current iteration number.

[0048] The third formula used is: Where, Indicates the i position of the t th egg at the b th iteration, b where 1 and 2 are independent random vectors, represents the lower limit of the spawning area, represents the upper limit of the spawning area, represents the current local best position; where, represents the T-distribution mutation perturbation with the iteration number as the degree-of-freedom parameter, is the position of the jth dimension of the foraging dung beetle after updating using the fourth formula, is the j current optimal position on the th dimension; where S is a constant, g is a random vector following a normal distribution, is the i position of the t th individual at the +1 moment after updating using the fifth formula, t is the global optimal position at the t moment, is the i local optimal position at the t moment,

[0049] The prediction output module 13 is used to generate a settlement prediction result according to the settlement prediction value.

[0050] In this embodiment, by performing complete ensemble empirical mode decomposition on the settlement monitoring data, noise can be effectively removed and stable intrinsic mode functions can be extracted. By calculating the complexity evaluation value of the intrinsic mode function, the complexity of the intrinsic mode function can be effectively evaluated. By clustering the intrinsic mode function using the complexity evaluation value, the complexity classification of the intrinsic mode function can be effectively performed, obtaining high-frequency components, medium-frequency components, and low-frequency components. By performing dual-scale decomposition using complete ensemble empirical mode decomposition and variational mode decomposition, noise in the signal can be accurately and effectively removed, reducing the interference of noise on the settlement prediction result and improving the accuracy of the settlement prediction.

[0051] Embodiment III Figure 3 is a structural block diagram of a terminal device 2 provided in the third embodiment of the present application. AsFigure 3 As shown, the terminal device 2 of this embodiment includes: a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program for the sedimentation prediction method integrating multi-scale optimization. When the processor 20 executes the computer program 22, the steps in each of the above embodiments of the sedimentation prediction method integrating multi-scale optimization are implemented.

[0052] Exemplarily, the computer program 22 can be divided into one or more modules. The one or more modules are stored in the memory 21 and executed by the processor 20 to complete this application. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, the processor 20 and the memory 21.

[0053] The so-called processor 20 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0054] The memory 21 may be an internal storage unit of the terminal device 2, such as the hard disk or memory of the terminal device 2. The memory 21 may also be an external storage device of the terminal device 2, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 2. Further, the memory 21 may also include both the internal storage unit and the external storage device of the terminal device 2. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 may also be used to temporarily store data that has been output or is to be output.

[0055] In addition, each functional module in the various embodiments of the present application may be integrated into one processing unit, may exist separately as individual physical units, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0056] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Among them, the computer-readable storage medium may be non-volatile or volatile. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present application, it may also be completed by instructing relevant hardware through a computer program. The computer program may be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments may be implemented. Among them, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0057] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A settlement prediction method integrating multi-scale optimization, characterized in that The method includes: Obtaining settlement monitoring data, and performing complete ensemble empirical mode decomposition on the settlement monitoring data to obtain intrinsic mode functions; Evaluating the complexity of the intrinsic mode functions to obtain complexity evaluation values, and clustering the intrinsic mode functions according to the complexity evaluation values to obtain high-frequency components, medium-frequency components, and low-frequency components; Performing variational mode decomposition on the high-frequency components to obtain high-frequency decomposition components, and inputting the high-frequency decomposition components, the medium-frequency components, and the low-frequency components into a pre-trained settlement prediction model for settlement prediction to obtain settlement prediction values; Generating a settlement prediction result according to the settlement prediction values.

2. The sedimentation prediction method integrating multi-scale optimization according to claim 1, wherein Performing complete ensemble empirical mode decomposition on the settlement monitoring data to obtain intrinsic mode functions, including: Adding white noise to the settlement monitoring data to obtain a noise signal, and decomposing the noise signal to obtain the intrinsic mode functions; The formula for adding white noise to the settlement monitoring data includes: wherein, is the white noise, is the settlement monitoring data, is the noise signal; The formula for decomposing the noise signal includes: Among them, is the i th intrinsic mode function, is the residual term, N represents the number of the noise signals.

3. The sedimentation prediction method integrating multi-scale optimization according to claim 1, characterized in that Evaluating the complexity of the intrinsic mode functions to obtain complexity evaluation values, including: Performing multi-scale coarse-graining processing on the time series corresponding to the intrinsic mode functions to obtain a coarse-grained sequence, and constructing vectors from the coarse-grained sequence according to the embedding dimension and delay time to obtain delay vectors; Sorting the delay vectors and recording the permutation patterns, and calculating the permutation entropy according to the recorded results of the permutation patterns to obtain the complexity evaluation values.

4. The sedimentation prediction method integrating multi-scale optimization according to claim 1, wherein Before inputting the high-frequency decomposition components, the medium-frequency components, and the low-frequency components into a pre-trained settlement prediction model for settlement prediction to obtain settlement prediction values, it further includes: Obtaining sample components, and inputting the sample components into the settlement prediction model for convolutional processing to obtain sample convolutional features; Performing batch normalization processing on the sample convolutional features to obtain sample normalized features, and performing non-linear transformation on the sample normalized features to obtain non-linear features; Performing pooling processing on the non-linear features to obtain pooling features, and performing bidirectional gated recurrent processing on the pooling features to obtain gated features; Flipping the gated features to obtain flipped features, and performing feature splicing on the flipped features to obtain spliced features; Performing fully connected processing on the spliced features to obtain fully connected features, and optimizing the parameters of the settlement prediction model according to the fully connected features until the settlement prediction model converges to obtain the pre-trained settlement prediction model.

5. The sedimentation prediction method integrating multi-scale optimization according to claim 4, wherein Optimizing the parameters of the settlement prediction model according to the fully connected features, including: Obtaining the model parameters of the settlement prediction model, and generating a dung beetle population according to the model parameters; Calculating the fitness values of the dung beetle population according to the fully connected features, and updating the current optimal position of the dung beetle population according to the fitness values; In the stage where the dung beetle rolls the ball without obstacles, updating the position of the dung beetle using the first formula, and when encountering an obstacle during the ball-rolling process, updating the position of the dung beetle using the second formula; Guiding the exploration dung beetle to update its position according to the adaptive convergence factor formula, and using the third formula to update the position of the reproductive dung beetle; Update the position of the foraging dung beetles using the fourth formula and update the position of the stealing dung beetles using the fifth formula; For the said dung beetle population, calculate the current optimal position according to the updated positions of the dung beetles to obtain the optimal updated position. Until the number of iterations of the dung beetle population is equal to the iteration threshold, output the optimal updated position of the current iteration; Set the parameters of the settlement prediction model according to the output optimal updated position to obtain the pre-trained settlement prediction model.

6. The sedimentation prediction method integrating multi-scale optimization according to claim 5, wherein The first formula used is: Among them, is the updated position of the dung beetle using the first formula by the i -th osprey at the j -th dimension, is the current position of the i -th osprey at the j -th dimension, is a random factor, is a step factor, is an inertia weight or influence factor; The second formula used is: in, Indicates i A dung beetle in the t The position at the iteration, Indicates the direction of the rolling ball. Indicates i A dung beetle in the t -The position at iteration 1; The self-adaptive convergence factor formula is: Where R is the convergence factor and t is the current number of iterations.

7. The sedimentation prediction method integrating multi-scale optimization according to claim 5, characterized in that The third formula used is: Among them, represents the position of the i th egg at the t th iteration, b 1 and b 2 are independent random vectors, represents the lower limit of the spawning area, represents the upper limit of the spawning area, represents the current local best position, represents the i th egg at the t -1th iteration; The fourth formula used is: Among them, represents the T-distribution mutation perturbation with the iteration number as the degree-of-freedom parameter, is the position of the j-th dimension of the foraging dung beetle after updating using the fourth formula, is the j current optimal position on the j-th dimension; The fifth formula used is: where S is a constant, g is a random vector following a normal distribution, is the position at the i -th individual after updating using the fifth formula at t +1 time step, is the t time step's global optimal position, is the t time step's local optimal position, is the i -th individual's position at t time step.

8. A settlement prediction system integrating multi-scale optimization, characterized in that, The said system includes: A mode decomposition module, used to obtain settlement monitoring data and perform complete ensemble empirical mode decomposition on the settlement monitoring data to obtain intrinsic mode functions; A clustering module, used to evaluate the complexity of the intrinsic mode functions to obtain a complexity evaluation value, and cluster the intrinsic mode functions according to the complexity evaluation value to obtain high-frequency components, medium-frequency components, and low-frequency components; A settlement prediction module, used to perform variational mode decomposition on the high-frequency components to obtain high-frequency decomposition components, and input the high-frequency decomposition components, the medium-frequency components, and the low-frequency components into the pre-trained settlement prediction model for settlement prediction to obtain a settlement prediction value; A prediction output module, used to generate a settlement prediction result according to the settlement prediction value.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the said processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the said computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images