A method and system for predicting the loss of a lightweight building material

Abstract

The present invention discloses a method and system for predicting the loss of lightweight building materials. The method includes data acquisition, data screening, loss prediction modeling, and prediction of the loss of lightweight building materials. The present invention belongs to the field of data processing, and specifically refers to a method and system for predicting the loss of lightweight building materials. In this solution, an environmental difference measurement function is introduced for sub-region merging to capture local loss characteristics; a local spatial autocorrelation index is introduced, and based on the local outlier score, abnormal loss data is optimized and identified, thereby effectively improving the quality of loss data; an environmental penalty correction factor is introduced to weight the samples in the abnormal environmental region, so that the prediction error in the region of drastic environmental change is amplified, thereby improving the sensitivity to severe loss; a compensation correction term is designed, and hierarchical penalties are implemented for different loss levels through the loss gradient, so that the model is more discriminative when predicting mild, moderate, and severe losses; and thus the final material loss prediction effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and specifically refers to a method and system for predicting the loss of lightweight building materials. Background Art

[0002] The material loss prediction method refers to the process of using technologies such as statistics, machine learning, or deep learning to model and predict the gradual degradation or loss of materials during use. Its main goal is to identify potential severely loss areas in advance through the analysis of historical data and multi-dimensional influencing factors such as environment and stress, so as to provide a scientific basis for maintenance, repair, and safety management. However, the general material loss prediction method has the problem of ignoring the interaction effect of space and environment factors, thus unable to capture local loss characteristics and effectively identify abnormal losses, resulting in poor accuracy of material loss prediction; the general material loss prediction method ignores the particularity of losses in areas with drastic environmental changes or uneven local stress, does not introduce a special compensation mechanism, and when facing severe losses, the prediction error is not taken seriously enough, resulting in poor final prediction effect. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides a method and system for predicting the loss of lightweight building materials. Aiming at the problem that the general material loss prediction method ignores the interaction effect of space and environment factors, thus unable to capture local loss characteristics and effectively identify abnormal losses, resulting in poor accuracy of material loss prediction, this solution introduces an environmental difference measurement function for sub-region merging to capture local loss characteristics; by introducing the local spatial autocorrelation index to identify the noise caused by sensor failures or external factors, and optimizing the identification of abnormal loss data based on the local outlier score, the quality of loss data is effectively improved, and thus the accuracy of the final material loss prediction is improved; aiming at the problem that the general material loss prediction method ignores the particularity of losses in areas with drastic environmental changes or uneven local stress, does not introduce a special compensation mechanism, and when facing severe losses, the prediction error is not taken seriously enough, resulting in poor final prediction effect, this solution weights the samples in abnormal environment areas by introducing an environmental penalty correction factor, so that the prediction error in the area of drastic environmental change is amplified, thereby improving the sensitivity to severe losses; for areas with uneven local stress and drastic environmental changes, an additional compensation correction term is designed to ensure that the loss prediction of lightweight building materials in these areas is additionally compensated; through the loss loss gradient, hierarchical penalties are imposed on different loss levels, so that the model is more discriminative when predicting mild, moderate, and severe losses; thus improving the final material loss prediction effect.

[0004] The technical solution adopted by the present invention is as follows: A method for predicting the loss of lightweight building materials provided by the present invention includes the following steps:

[0005] Step S1: Data acquisition;

[0006] Step S2: Data screening;

[0007] Step S3: Loss prediction modeling;

[0008] Step S4: Loss prediction of lightweight building materials.

[0009] Furthermore, in step S1, the data acquisition is to obtain historical environmental data and historical loss data of lightweight building materials; and clarify the loss level, taking the loss level as the data label; based on feature engineering, an initial loss data set is obtained.

[0010] Furthermore, in step S2, the data screening specifically includes the following steps:

[0011] Step S21: Sub-region merging process; introduce the environmental difference measurement function D(·); continuously merge regions based on the similarity between regions until the number of sub-regions is reached. The initial sub-regions are data points; the similarity between regions is expressed as: ; ; where and represent the loss data sets of two adjacent regions; and represent the weight coefficients of two adjacent regions; represents the comprehensive loss effect after region merging; and are adjustment parameters; where and are expressed as the environmental data vectors of two regions; T is the transpose operation; is a diagonal matrix, and the diagonal elements represent the sensor reliability of each environmental indicator; is the covariance matrix of environmental data;

[0012] Step S22: Optimize the number of sub-regions, introduce a penalty for small sample regions; the optimal number of sub-regions P is expressed as: ; where K is the total number of sub-regions, k is the sub-region index; N k is the number of samples in the k-th sub-region; is the variance of the loss data within the k-th sub-region; N is the total number of samples; is the variance of the overall data; is an additional penalty coefficient;

[0013] Step S23: Identify abnormal losses; calculate the local spatial autocorrelation index which is expressed as: ; where , and are the i-th, j-th, and s-th loss data respectively; i and j are data indices; n is the total number of measurement points in the currently investigated area; is the average loss value of all measurement points; is the spatial weight of the point; is the neighborhood of the loss data, and k is the neighborhood data index; is the smoothing term; is the median of the loss data of all measurement points in the neighborhood; represents the median of the absolute deviations of all measurement points in the neighborhood;

[0014] Step S24: Outlier score optimization; perform deviation adjustment based on local loss and introduce a local sensor reliability factor ; the outlier score is expressed as: ; ; ; where is the local loss deviation of the i-th measurement point; is the average loss deviation of the measurement point; is the inherent standard error of the sensor; is the local dispersion degree of the loss data in the current sub-region; preset a threshold, identify the data with an outlier score higher than the threshold as abnormal loss data; eliminate the abnormal loss data to obtain a filtered loss prediction data set.

[0015] Furthermore, in step S3, the loss prediction modeling specifically includes the following steps:

[0016] Step S31: Feature construction; first perform principal component analysis on the filtered data to extract the global feature vector ; use local linear embedding for the data in the neighborhood to extract local non-linear features ; fuse the two parts of information to form the final feature vector;

[0017] Step S32: Dual-branch deep learning model design; perform loss prediction based on the deep learning model. One branch captures the spatial distribution characteristics of material loss, and the other branch focuses on environmental variables and global features. The two branches jointly predict the loss after fusion; the input of the model is the extracted feature vector, and the output is the loss prediction value of the lightweight building material; use the filtered data as the normal loss data set and divide it into a training set and a validation set; optimize the model parameters using the gradient descent and backpropagation algorithms; use the K-fold cross-validation method to evaluate the generalization performance of the model;

[0018] Step S33: Design the loss function; design the penalty loss function LE and introduce the environmental penalty correction factor , which is expressed as: ; ; where N is the total number of samples; is the predicted loss value of the sample; is the environmental data vector; is the mean of the overall environmental vector; is the environmental adjustment coefficient; c is the critical threshold of the loss level; is the penalty exponent, which is used to adjust the weights of loss errors at different levels; and the loss loss gradient is defined as ; introduce the loss correction amount, and the final loss function LC is expressed as: ; ; where is the loss correction amount; and are the environmental compensation weight coefficient and the stress compensation weight coefficient respectively; and are the exponential parameters; is the standard deviation of the environmental deviation; is the average load stress within the neighborhood; is the average value of the load stress in the entire area.

[0019] Furthermore, in step S4, the loss prediction of the lightweight building material is to collect environmental data and lightweight building material loss data in real time, input them into the loss prediction model after feature engineering processing, and use the loss level output by the loss prediction model as the loss prediction result.

[0020] A loss prediction system for lightweight building materials provided by the present invention includes a data acquisition module, a data screening module, a loss prediction modeling module, and a lightweight building material loss prediction module;

[0021] The data acquisition module acquires historical environmental data and lightweight building material loss data, and constructs an initial loss data set based on feature engineering;

[0022] The data screening module screens the initial data based on local outlier detection and eliminates the abnormal data;

[0023] The loss prediction modeling module constructs a global and local features, designs a double-branch deep learning model and a dedicated loss function based on the screened data to realize the modeling of material loss prediction;

[0024] The lightweight building material loss prediction module uses the loss prediction modeling to realize the loss prediction of lightweight building materials for the environmental and loss data collected in real time.

[0025] The beneficial effects achieved by the present invention using the above solution are as follows:

[0026] (1) Regarding the problem that general material loss prediction methods ignore the interaction effects of space and environmental factors, thus unable to capture local loss characteristics, unable to effectively identify abnormal losses, and resulting in poor accuracy of material loss prediction, this solution introduces an environmental difference measurement function for sub-region merging to capture local loss characteristics; by introducing a local spatial autocorrelation index to identify noise caused by sensor failures or external factors, and optimizing the identification of abnormal loss data based on local outlier scores, thereby effectively improving the quality of loss data and further enhancing the accuracy of the final material loss prediction.

[0027] (2) Regarding the problem that general material loss prediction methods ignore the particularity of losses in areas with drastic environmental changes or uneven local forces, without introducing a dedicated compensation mechanism, and when facing severe losses, the prediction errors are not given sufficient attention, resulting in poor final prediction effects, this solution weights the samples in abnormal environmental regions by introducing an environmental penalty correction factor, so that the prediction errors in regions with drastic environmental changes are amplified, thereby enhancing the sensitivity to severe losses; for regions with uneven local forces and drastic environmental changes, an additional compensation correction term is designed to ensure that the loss prediction of lightweight building materials in these regions is additionally compensated; by implementing hierarchical penalties for different loss levels through the loss gradient, the model becomes more discriminative when predicting mild, moderate, and severe losses; thus improving the final material loss prediction effect. Description of the Drawings

[0028] Figure 1 It is a schematic flowchart of a method for predicting the loss of a lightweight building material provided by the present invention;

[0029] Figure 2 It is a schematic diagram of a system for predicting the loss of a lightweight building material provided by the present invention;

[0030] Figure 3 It is a schematic flowchart of step S2.

[0031] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. Detailed Embodiments

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0034] Example 1. Refer to Figure 1 , a method for predicting the loss of a lightweight building material provided by the present invention, the method comprising the following steps:

[0035] Step S1: Data acquisition; acquiring historical environmental data and lightweight building material loss data, and constructing an initial loss data set based on feature engineering;

[0036] Step S2: Data screening; screening the initial data based on local outlier detection and removing the outlier data;

[0037] Step S3: Loss prediction modeling; based on the screened data, by constructing global and local features, designing a dual-branch deep learning model and a dedicated loss function, realizing the modeling of material loss prediction;

[0038] Step S4: Prediction of lightweight building material loss; using the loss prediction modeling to realize the prediction of lightweight building material loss for the real-time collected environmental and loss data.

[0039] Example 2. Refer to Figure 1 , based on the above example, in step S1, the data acquisition is based on sensors to acquire historical environmental data and historical lightweight building material loss data; and the loss level is clarified and used as the data label; an initial loss data set is obtained based on feature engineering; sensors are installed at the key parts of the walls and beam-slab of the building structure to collect environmental data and the loss data of the lightweight building material itself; the historical environmental data includes humidity, temperature, precipitation, wind speed, air quality, ultraviolet radiation intensity and load stress; the historical lightweight building material loss data includes material type, crack width and material shedding rate; the loss level includes no loss, mild loss, moderate loss and severe loss.

[0040] Example 3. Refer to Figure 1 and Figure 3 , based on the above example, in step S2, the data screening is to introduce sub-region merging and local outlier detection to screen the initial loss data set to ensure that the normal loss data set is screened out, providing high-quality data for subsequent modeling; specifically including the following steps:

[0041] Step S21: Sub-region merging process; Divide the material loss data into several sub-regions according to the spatial distribution, so that the losses within the same region have high consistency, which is convenient for capturing local loss characteristics caused by environmental factors such as humidity, load, and temperature gradient; Introduce the environmental difference measurement function D(·), so that the sub-region merging not only considers the material loss data but also fully reflects the environmental characteristics of each sub-region to improve the discrimination ability of local loss characteristics; Continuously merge regions based on the similarity between regions until the number of sub-regions is reached. The initial sub-regions are data points; The similarity between regions is expressed as: ; ; where, and represent the loss data sets of two adjacent regions; and represent the weight coefficients of two adjacent regions, reflecting the importance of the loss data within each region; represents the comprehensive loss effect after region merging; and are adjustment parameters that balance the influence of loss data and environmental data on the sub-region merging result respectively; where, and represent the environmental data vectors of two regions; T is the transpose operation; is a diagonal matrix, and the diagonal elements represent the sensor reliability of each environmental index; is the covariance matrix of environmental data, reflecting the statistical correlation between environmental indicators;

[0042] Step S22: Optimize the number of sub-regions. While preventing the loss of details caused by excessive sub-region merging, the losses at different positions can be reasonably divided to ensure that the classification details are retained; And introduce a penalty for small sample regions to ensure sufficient data and reliable statistics within each sub-region; The optimal number of sub-regions P is expressed as: ; where, K is the total number of sub-regions, k is the sub-region index; N k is the number of samples in the k-th sub-region; is the variance of the loss data within the k-th sub-region; N is the total number of samples; is the variance of the overall data; is the additional penalty coefficient;

[0043] Step S23: Abnormal loss identification; By measuring the spatial correlation of the material loss region, identify the abnormal loss points caused by statistical noise and actual structural defects, so as to eliminate the abnormal data; Calculate the local spatial autocorrelation index which is expressed as: ; where, , and They are the i-th, j-th, and s-th loss data respectively; i and j are data indices; n is the total number of measurement points in the currently investigated area; is the average loss value of all measurement points; is the spatial weight of the point; is the loss data neighborhood, and k is the neighborhood data index; is the smoothing term; is the median of the loss data of all measurement points within the neighborhood; represents the median of the absolute deviations of all measurement points within the neighborhood;

[0044] Step S24: Outlier score optimization; for areas with small data fluctuations, enhance the outlier scoring ability and improve the recognition accuracy of abnormal losses; perform deviation adjustment based on local losses and introduce a local sensor reliability factor such that the score can reflect the credibility of sensor data, thereby more accurately identifying abnormal losses; the outlier score is expressed as: ; ; ; where is the local loss deviation of the i-th measurement point; is the average loss deviation of the measurement points; is the inherent standard error of the sensor; is the local dispersion degree of the loss data within the current sub-region; preset a threshold, and identify the data with an outlier score higher than the threshold as abnormal loss data; eliminate the abnormal loss data to obtain a filtered loss prediction data set.

[0045] By performing the above operations, for the problem that the general material loss prediction method ignores the interaction effect of spatial and environmental factors, thus unable to capture local loss characteristics, unable to effectively identify abnormal losses, and further resulting in poor accuracy of material loss prediction, this solution introduces an environmental difference measurement function for sub-region merging to capture local loss characteristics; by introducing a local spatial autocorrelation index to identify the noise caused by sensor failures or external factors, and based on local outlier score optimization to identify abnormal loss data, thereby effectively improving the quality of loss data, and further improving the accuracy of the final material loss prediction.

[0046] Example 4, refer to Figure 1 Based on the above example, in step S3, the loss prediction modeling uses the filtered data to establish a loss prediction model. The loss prediction model not only considers the input of environmental variables, but also incorporates the spatial distribution characteristics of local loss data, and finely penalizes the prediction error through a designed loss function to ensure greater sensitivity to the identification of severe losses; specifically includes the following steps:

[0047] Step S31: Feature construction: Perform principal component analysis on the filtered data to extract the global feature vector ; Local linear embedding is used for the data in the neighborhood to extract local nonlinear features ; The two parts of information are combined to form the final feature vector;

[0048] Step S32: dual-branch deep learning model design; loss prediction is performed based on the deep learning model, one branch captures the spatial distribution characteristics of material loss, and the other branch focuses on environmental variables and global features. The two branches are fused to jointly predict the loss; the model input is the extracted feature vector, and the output is the loss prediction value of lightweight building materials; the screened data is used as a normal loss data set and divided into a training set and a validation set; the model parameters are optimized using gradient descent and back propagation algorithms; the K-fold cross-validation method is used to evaluate the generalization performance of the model;

[0049] Step S33: Design a loss loss function; To ensure the model's ability to distinguish different loss levels, especially the sensitivity to severe loss, design a special loss function and its gradient penalty term; Design a penalty loss function LE and introduce an environmental penalty correction factor , so that the loss of samples in abnormal environmental areas is amplified, expressed as: ; ; Where N is the total number of samples; is the predicted loss value of the sample; is the environmental data vector; is the mean of the overall environment vector; is the environmental adjustment coefficient; c is the critical threshold of the loss level; is the penalty index, which is used to adjust the weights of different levels of loss errors; and the loss gradient is defined as ; In order to deal with the uneven stress area of ​​building materials, it is necessary to improve the ability to distinguish different loss levels and introduce loss correction to compensate for the loss prediction of lightweight building materials in the area of ​​local uneven stress and environmental drastic changes. The final loss loss function LC is expressed as: ; ;in, It is the loss correction amount, which provides additional compensation for areas where the material is subjected to uneven forces or where the environment changes drastically; and They are the environmental compensation weight coefficient and the force compensation weight coefficient; and is the exponential parameter; is the standard deviation of the environmental deviation; is the average load stress in the neighborhood; is the average value of the load stress over the entire area.

[0050] By performing the above operations, in view of the problem that the general material loss prediction method ignores the particularity of losses in areas with drastic environmental changes or uneven local stress, and does not introduce a special compensation mechanism. When facing severe losses, the prediction error is not given enough attention, resulting in a poor final prediction effect. In this solution, an environmental penalty correction factor is introduced to weight the samples in abnormal environmental areas, so that the prediction error in areas with drastic environmental changes is amplified, thereby improving the sensitivity to severe losses. For areas with uneven local stress and drastic environmental changes, an additional compensation correction term is designed to ensure that the loss prediction of lightweight building materials in these areas is additionally compensated. By implementing hierarchical penalties for different loss levels through the loss gradient, the model is made more discriminative when predicting mild, moderate, and severe losses, thereby improving the final material loss prediction effect.

[0051] Example Five, refer to Figure 1 , based on the above example, in step S4, the loss prediction of lightweight building materials is to collect environmental data and lightweight building material loss data in real time. After being processed by feature engineering, they are input into the loss prediction model, and the loss level output by the loss prediction model is used as the loss prediction result. When the loss level is severe loss or moderate loss, a warning is given to the management personnel.

[0052] Example Six, refer to Figure 2 , based on the above example, a loss prediction system for lightweight building materials provided by the present invention includes a data acquisition module, a data screening module, a loss prediction modeling module, and a lightweight building material loss prediction module;

[0053] The data acquisition module acquires historical environmental data and lightweight building material loss data, constructs an initial loss data set based on feature engineering, and sends the data to the data screening module;

[0054] The data screening module screens the initial data based on local outlier detection, eliminates abnormal data, and sends the data to the loss prediction modeling module;

[0055] The loss prediction modeling module, based on the screened data, realizes the modeling of material loss prediction by constructing global and local features, designing a dual-branch deep learning model, and a dedicated loss function, and sends the data to the lightweight building material loss prediction module;

[0056] The lightweight building material loss prediction module uses loss prediction modeling to realize the loss prediction of lightweight building materials for the environment and loss data collected in real time.

[0057] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0058] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention.

[0059] The above describes the present invention and its embodiments, which description is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In summary, if those of ordinary skill in the art are inspired by it and design, without creative efforts, structural modes and embodiments similar to the technical solution without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A method for predicting the loss of lightweight building materials, characterized in that: The method comprises the following steps: Step S1: Data acquisition: historical environmental data and lightweight building material loss data are acquired, and an initial loss data set is constructed based on feature engineering; Step S2: data screening; Step S3: loss prediction modeling; Step S4: lightweight building material loss prediction: using loss prediction modeling to predict lightweight building material loss based on real-time collected environmental and loss data; Step S2 includes step S21: sub-region merging process; introducing environmental difference measurement function D(·); continuously merging regions based on inter-region similarity until the number of sub-regions is reached, and the initial sub-region is a data point; inter-region similarity It is expressed as: ; ;in, and Represents a set of loss data for two adjacent regions; and Represents the weight coefficient of two adjacent regions; represents the comprehensive loss effect after regional merger; and is the adjustment parameter; where and Represented as environmental data vectors of two regions; T is the transpose operation; It is a diagonal matrix, and the diagonal elements represent the sensor reliability of each environmental indicator; is the covariance matrix of the environmental data; In step S3, the loss prediction modeling specifically includes the following steps: Step S31: Feature construction: Perform principal component analysis on the filtered data to extract the global feature vector ; Local linear embedding is used for the data in the neighborhood to extract local nonlinear features ; The two parts of information are combined to form the final feature vector; Step S32: dual-branch deep learning model design; loss prediction is performed based on the deep learning model, one branch captures the spatial distribution characteristics of material loss, and the other branch focuses on environmental variables and global features. The two branches are fused to jointly predict the loss; the model input is the extracted feature vector, and the output is the loss prediction value of lightweight building materials; the screened data is used as a normal loss data set and divided into a training set and a validation set; the model parameters are optimized using gradient descent and back propagation algorithms; the K-fold cross-validation method is used to evaluate the generalization performance of the model; Step S33: Design the loss function; design the penalty loss function LE and introduce the environmental penalty correction factor , expressed as: ; ; Where N is the total number of samples; is the predicted loss value of the sample; is the environmental data vector; is the mean of the overall environment vector; is the environmental adjustment coefficient; c is the critical threshold of the loss level; is the penalty index, which is used to adjust the weights of different levels of loss errors; and the loss gradient is defined as ; Introducing the loss correction, the final loss loss function LC is expressed as: ; ;in, is the loss correction; and They are the environmental compensation weight coefficient and the force compensation weight coefficient; and is the exponential parameter; is the standard deviation of the environmental deviation; is the average load stress in the neighborhood; is the average value of the load stress over the entire area.

2. The method for predicting loss of lightweight building materials according to claim 1, characterized in that: In step S2, the data screening further includes the following steps: Step S22: Optimize the number of sub-regions and introduce penalties for small sample regions; the optimal number of sub-regions P is expressed as: ; Where K is the total number of sub-regions, k is the sub-region index; N k is the number of samples in the kth sub-region; is the variance of loss data in the kth sub-region; N is the total sample size; is the variance of the overall data; is the additional penalty coefficient; Step S23: Abnormal loss identification; calculation of local spatial autocorrelation index , expressed as: ;in, , and are the i-th, j-th and s-th loss data respectively; i and j are data indexes; n is the total number of measurement points in the current investigation area; is the average loss value of all measurement points; is the spatial weight of the point; is the loss data neighborhood, k is the neighborhood data index; is a smoothing term; yes The median of the loss data of all measuring points in the neighborhood; Represents the median of the absolute deviations of all measurement points in the neighborhood; Step S24: outlier score optimization.

3. The method for predicting loss of lightweight building materials according to claim 2, characterized in that: In step S2, the outlier score optimization is based on local loss to adjust the deviation and introduce the local sensor reliability factor ; Outlier score It is expressed as: ; ; ;in, is the local loss deviation of the ith measuring point; is the average loss deviation of the measuring point; is the standard error inherent in the sensor; It is the local dispersion degree of loss data in the current sub-region; a threshold is set in advance, and data with an outlier score higher than the threshold is identified as abnormal loss data; the abnormal loss data is eliminated to obtain a screened loss prediction data set.

4. The method for predicting loss of lightweight building materials according to claim 3, characterized in that: In step S4, the lightweight building material loss prediction is to collect environmental data and lightweight building material loss data in real time, input them into the loss prediction model after feature engineering processing, and use the loss level output by the loss prediction model as the loss prediction result.

5. A loss prediction system for lightweight building materials, used to implement a loss prediction method for lightweight building materials as claimed in any one of claims 1 to 4, characterized in that: It includes a data acquisition module, a data screening module, a loss prediction modeling module and a lightweight building material loss prediction module; The data acquisition module acquires historical environmental data and lightweight building material loss data, and constructs an initial loss data set based on feature engineering; The data screening module screens the initial data based on local anomaly detection to remove abnormal data; The loss prediction modeling module realizes the modeling of material loss prediction based on the screened data by constructing global and local features, designing a dual-branch deep learning model and a dedicated loss loss function; The lightweight building material loss prediction module uses loss prediction modeling to predict the loss of lightweight building materials based on real-time collected environmental and loss data.

Images