A garden building residual life prediction method based on environmental factors

Abstract

The application provides a garden building residual life prediction method based on environmental factors, and belongs to the field of data prediction. The method realizes accurate prediction of the residual life of the garden building through the synergistic effect of a data collection and feature generation module, a space factor module, a space-time disturbance enhancement module and a prediction module. The data collection and feature generation module collects relevant data of the environment where the garden building is located, constructs a multi-dimensional feature data set, and generates a high-dimensional feature data set by optimizing the features using a dynamic weighting mechanism. The space factor module models the space-time dependence relationship between the building and the environment, further enhancing the feature expression capability. In the space-time disturbance enhancement module, a disturbance factor and a multi-level neighborhood update mechanism are introduced to capture the complex relationship between the features, and finally realize intelligent optimization and accurate prediction of the residual life of the garden building. The application can significantly improve the intelligent and refined level of building management and provide a scientific basis for garden building life cycle management.

Description

Technical Field

[0001] This invention belongs to the field of data prediction, specifically relating to a method for predicting the remaining lifespan of garden buildings based on environmental factors. Background Technology

[0002] With the continuous advancement of urbanization and the widespread development of landscaping, garden architecture plays an important role in landscape design and the urban environment. As facilities that are exposed to the natural environment for a long time, the remaining lifespan of garden architecture directly affects the utilization efficiency and economic costs of public resources. However, traditional methods for predicting the remaining lifespan of garden architecture mostly rely on empirical judgment and static models, lacking sufficient consideration of the dynamic impact of environmental factors on the building. This results in low accuracy in maintenance decisions and remaining lifespan predictions for garden architecture, which in turn affects the long-term operation and management of garden architecture.

[0003] With the rapid development of information technology, especially the application of big data and machine learning technologies, researchers have begun to try to incorporate environmental factors into the prediction of the remaining lifespan of garden buildings. Machine learning technology can learn from a large amount of historical data to identify complex environmental factors that affect the remaining lifespan of garden buildings, thereby providing more accurate prediction results.

[0004] However, current machine learning-based methods for predicting the remaining lifespan of garden buildings still face many challenges. First, the remaining lifespan of garden buildings is affected by multiple environmental factors, and the interactions between these factors are complex and uncertain. Second, effectively extracting the most meaningful features for predicting the remaining lifespan and reasonably fusing these features remains a major challenge in technical implementation. Therefore, environmental factor-based methods for predicting the remaining lifespan of garden buildings are still in the stage of continuous optimization and improvement, and innovative technical solutions are urgently needed to meet the needs of modern garden building management. Summary of the Invention

[0005] This invention provides a method for predicting the remaining lifespan of garden buildings based on environmental factors. It proposes a prediction model for garden building data with complex and multivariate environmental factors, which consists of a data collection and feature generation module, a spatial factor module, a spatiotemporal disturbance enhancement module, and a prediction module.

[0006] The technical solution adopted by the present invention to achieve the above objectives specifically includes the following steps:

[0007] S1. Collect relevant environmental data of the garden buildings and construct a multi-dimensional feature dataset, including temperature, humidity, wind speed, solar radiation intensity and soil moisture;

[0008] S2. Optimize the original multidimensional feature dataset using a dynamic weighting mechanism. Construct a high-dimensional feature dataset as the first dataset through multi-level feature combination and feature enhancement. The dynamic weighting mechanism is based on the building health status H. j (t) and the environmental context C k (t) Design a weight update strategy to dynamically adjust the weights between features and adaptively weight different features;

[0009] S3. Construct the spatial factor module, and calculate the spatial factor S for each feature through local windowing and global compensation. t，k Model the spatiotemporal dependencies between buildings and the environment and generate a second dataset;

[0010] S4. Construct a spatiotemporal perturbation enhancement module, update and optimize the feature state, introduce a fixed perturbation factor η to simulate the randomness of the data, design a multi-level neighborhood update mechanism, calculate the spatial and temporal similarity of features, introduce a nonlinear iterative update mechanism to update the feature state and generate a third dataset, which is divided into a training set and a prediction set.

[0011] S5. The training set is input into the garden building remaining life prediction model. The garden building remaining life prediction model adopts a regression model. The training set is used to train the model and fit the data to learn the relationship between the environmental factors of the building and the remaining life of the building.

[0012] S6. The prediction set is input into the trained garden building remaining life prediction model, and the final output is the predicted value of the garden building's remaining life.

[0013] Preferably, in step S1, data on environmental factors related to garden architecture are collected, including temperature, humidity, wind speed, solar radiation intensity, and soil moisture, to construct a multidimensional raw dataset.

[0014] Preferably, the original multidimensional feature dataset is optimized by designing a weight update strategy to dynamically adjust the weights of building health status and environmental context based on changes in building health status and environment. This ensures that the model can flexibly respond to changes in different environments and features. The mathematical model of the weight update strategy is as follows:

[0015]

[0016] In the formula, σ(x) is the Sigmoid function, representing the influence of each environmental factor, and H... j (t) represents the health status characteristics of the garden buildings, θ ij and α ik C is a weighting coefficient representing the relationship between environmental factors and the health status and environmental context of garden buildings. k(t) represents the environmental context. Through dynamic weighting, the model can better identify and focus on features that have a significant impact on the building's lifespan, thereby improving prediction accuracy. The weighted environmental data can be represented as:

[0017]

[0018] In the formula, E i (t) represents the environmental factor, indicating the i-th environmental factor collected at time t, E ω (t) represents weighted environmental data, indicating the relationships between task states, ω i (t) represents the weighting coefficients, indicating the impact of environmental factors on the health status of garden buildings. Then, the weighted environmental data, building health status, and environmental context are combined and enhanced through multi-level feature combination to generate high-dimensional features. The mathematical model is as follows:

[0019] G(t)=T(E ω (t),H(t),C(t))=W1·E ω (t)+W2·H(t)+W3·C(t)+b;

[0020] In the formula, G(t) represents the generated high-dimensional feature data, indicating the complex features that, after transformation, can reflect the environmental impact and building health status during the aging process of garden buildings. W1, W2, and W3 are weight matrices, and b is a bias term. Through multi-level feature extraction, the dimensionality of the features is enhanced layer by layer, enabling each layer of features to capture the complex relationship between buildings and the environment more meticulously. Finally, an adaptive noise filtering mechanism is introduced to filter noise. By dynamically calculating the noise filtering coefficient, the noise filtering intensity is adjusted according to changes in environmental factors, building health status, and environmental context, thereby optimizing the quality of environmental data and building features in real time and eliminating irrelevant or error-prone data noise. The mathematical model is as follows:

[0021] E f (t)=E ω (t)·(1-λ(t));

[0022] In the formula, E f λ(t) represents the environmental data after noise filtering, and λ(t) is the dynamically calculated noise filtering coefficient. The mathematical model is as follows:

[0023]

[0024] In the formula, β i and γ j To adjust the parameters of noise filtration intensity, representing the influence of environmental factors and building health status on noise filtration intensity, δ k The environmental context is used as a factor influencing the noise filtering intensity, resulting in the first dataset.

[0025] Preferably, by employing a dynamic weighting mechanism and multi-level feature combination, the original multi-dimensional feature dataset can be adaptively optimized, improving the expressive power of the features. The dynamic weighting mechanism allows feature weights to be adjusted in real time according to changes in environmental factors, building health status, and environmental context, thereby effectively focusing on the most critical features for building life prediction and avoiding the limitations of fixed weights. This flexible weighting method improves the model's ability to capture the non-linear relationship between complex environments and building features. By adopting a multi-level feature combination and feature enhancement strategy, more complex and detailed information can be extracted by gradually constructing higher-dimensional features, enhancing the model's expressive power and making the prediction results more accurate. The introduced adaptive noise filtering mechanism further optimizes data quality. By dynamically calculating the noise filtering coefficient and adjusting the noise filtering intensity in real time, it ensures that environmental data and building features maintain high quality throughout the changing process, thereby improving the robustness and prediction accuracy of the model.

[0026] Preferably, the S3 construction spatial factor module firstly calculates the local spatial factor F of feature k at time t for each garden building feature k based on the influence of surrounding environmental factors and neighboring buildings by designing a local sliding window. local，t，k By calculating the relationship between each feature and its surrounding features, a local spatial factor is generated, reflecting the spatiotemporal dependence between features. The mathematical model of the local spatial factor is as follows:

[0027]

[0028] In the formula, X t，n The environmental factors that affect a building at different points in time. N is the weighting coefficient between feature k and its neighbor features n. k Given the set of neighboring buildings of feature k, then calculate the global spatial factor F based on the local spatiotemporal factors. global,t To ensure that the impact of global environmental factors is comprehensively considered, local spatial factors are compensated by using the average environmental characteristics of all garden buildings. The specific mathematical model is as follows:

[0029]

[0030] In the formula, N represents the total number of features. Through an adaptive weight adjustment mechanism, the influence weights of spatial factors are dynamically adjusted, allowing the model to automatically optimize the weights of spatial factors based on the correlation between different features. This ensures more accurate feature fusion under different environmental conditions. The mathematical model is as follows:

[0031]

[0032] In the formula, To control the hyperparameters of environmental similarity sensitivity, the final spatial factor S of feature k at time t is generated by fusing compensations from local and global spatial factors. t，k To ensure that the spatiotemporal relationship between architectural features and their surrounding environment is fully considered, the mathematical model is as follows:

[0033] S t,k =F local,t,k +Ψ t，k ·F global,t ;

[0034] Finally, denormalization is used to transform the spatial factors back into the original environmental feature space of the building, restoring the generated spatial factors to the original feature space, ensuring that all calculation results are consistent with the actual data of the building and environment. The mathematical model is as follows:

[0035]

[0036] In the formula, S is the standard deviation matrix of the feature space, and the second dataset is finally obtained.

[0037] Preferably, the spatial factor module combines local and global spatial factors and employs an adaptive weight adjustment mechanism to accurately model the spatiotemporal dependency between buildings and the environment. This method allows the model to flexibly adjust feature weights dynamically based on changes in environmental factors, improving the accuracy and robustness of building lifespan prediction. The generated spatial factors not only enhance the model's ability to express building features and adapt to the impact of different environmental conditions on building aging, thus improving the accuracy and robustness of remaining lifespan prediction, but also provide refined and rich input for subsequent prediction tasks. Finally, through a denormalization step, the generated spatial factors are restored to the original feature space, ensuring that all calculation results are consistent with the actual data of buildings and the environment, making subsequent predictions more realistic and interpretable.

[0038] Preferably, a perturbation factor is introduced to simulate the randomness of building features and environmental changes, and the complex relationships between features are simulated to construct the spatiotemporal perturbation enhancement module. This allows the model to better adapt to the uncertainty in the data during training, enhancing the model's robustness. Simultaneously, the nonlinear interactions between features are considered. The mathematical model is as follows:

[0039]

[0040] In the formula, To introduce the characteristics of nonlinear perturbations and interactions, where η is a fixed perturbation factor and τ is a characteristic of the interaction. tkWe assign interaction weights to features k and n. To avoid overfitting and improve the model's generalization ability, we design a multi-level neighborhood update mechanism to capture the spatiotemporal dependence of buildings on their surroundings. The neighborhood influence is determined through multiple levels of integration, allowing the neighborhood weight of each node to be dynamically adjusted based on node similarity and global importance. We then calculate the similarity between features, including spatial and temporal similarity. The mathematical model for spatial similarity calculation is as follows:

[0041] The mathematical model for calculating time similarity is:

[0042]

[0043] Combining spatial similarity and temporal similarity, the mathematical model is as follows:

[0044]

[0045] In the formula, ζ is a hyperparameter used to adjust the weights of feature space similarity and temporal similarity. Finally, the multi-level weighted neighborhood weights are calculated, and the mathematical model is as follows:

[0046]

[0047] In the formula, distance(k, n) is the physical distance between feature k and feature n, ρ is a hyperparameter controlling the influence of the physical distance between features in the weighting, and I k The mathematical model for measuring feature importance is as follows:

[0048]

[0049] In the formula, deg(k) is the degree of feature k. As a global mean feature, to further capture complex nonlinear relationships, a nonlinear iterative mechanism is introduced. The feature is continuously updated through multiple iterations, enabling the model to gradually adjust the state of the nodes. The mathematical model for the iterative update rule is as follows:

[0050]

[0051] In the formula, ξ1 is the hyperparameter that controls the update magnitude of each iteration, and ξ2 is the weight coefficient of the neighborhood update. The nonlinear iterative mechanism helps the model gradually adjust the node features, which can capture complex nonlinear relationships and finally obtain the third dataset. At the same time, the third dataset is divided into training set and prediction set according to 7:3.

[0052] Preferably, the spatiotemporal perturbation enhancement module introduces perturbation factors, multi-level neighborhood update mechanisms, dynamic similarity adjustment, feature importance measurement, and nonlinear iteration mechanisms. This method can effectively capture the complex spatiotemporal dependencies between buildings and the environment, enhance the robustness and accuracy of the model, and improve the prediction accuracy and generalization ability of the model by dynamically adjusting weights and optimizing features. Ultimately, it provides high-quality predictions of the remaining lifespan of garden buildings. Through this comprehensive modeling method, it can cope with changing environmental conditions and building characteristics, thereby providing more accurate building health status assessments in practical applications.

[0053] Preferably, the training set is input into the garden building remaining life prediction model. Through regression methods, based on the environmental characteristics and building health status in the training set, the remaining life of the building is learned. The core objective of the regression model is to establish the relationship between environmental characteristics and the remaining life of the building. The mathematical model for predicting the remaining life of the garden building is as follows:

[0054]

[0055] In the formula, f(·,θ) is the regression function. Let θ represent the building's remaining lifespan predicted by the model, and θ be the parameters of the regression model, representing the weights of environmental factors and building health status on the remaining lifespan prediction. During training, the model is trained by minimizing a loss function, which uses mean squared error. The mathematical model is as follows:

[0056]

[0057] In the formula, M is the number of samples in the training set, and X... k Let θ represent the actual remaining life of the building, i.e., the labels in the training set. Loss(θ) is the loss function, representing the error between the model's predictions and the actual values. The loss function is optimized using backpropagation, and the model parameters θ are gradually adjusted to reduce the prediction error. Gradient descent is used to optimize the parameters. The mathematical model is as follows:

[0058] In the formula, θ i Here, represents the model parameters in the i-th iteration, and u is the learning rate, controlling the step size for each update. The gradient of the loss function with respect to the parameter θ is used as the basis for training and adjustment. The prediction set is then input into the trained model for predicting the remaining lifespan of garden buildings, ultimately outputting a predicted value for the remaining lifespan of the garden buildings.

[0059] Preferably, by using regression models, loss function optimization, gradient descent algorithms, and multiple rounds of training and adjustment, the remaining service life of garden buildings can be accurately predicted. This method continuously improves the prediction accuracy by gradually adjusting the model parameters, making the final output life prediction results more accurate and reliable, providing effective support for the management and maintenance of garden buildings. By optimizing the training process, the robustness and generalization ability of the model are enhanced, ensuring that the model can adapt to complex environmental conditions and provide high-quality predictions.

[0060] In summary, this invention proposes a method for predicting the remaining lifespan of garden buildings based on environmental factors. The method comprises the following modules: a data collection and feature generation module, a spatial factor module, a spatiotemporal perturbation enhancement module, and a prediction module. First, by collecting relevant data on the environment in which the garden building is located, a multidimensional feature dataset is constructed, and a dynamic weighting mechanism is used to optimize the features, generating a high-dimensional feature dataset. Then, the spatial factor module models the spatiotemporal dependency between the building and its environment, further enhancing the expressive power of the features. In the spatiotemporal perturbation enhancement module, perturbation factors and a multi-level neighborhood update mechanism are introduced to capture the complex relationships between features. Finally, a regression model is trained, using the training set to learn the relationship between the building's environmental features and its remaining lifespan. Ultimately, the predicted remaining lifespan of the garden building is obtained through the prediction set. Compared with traditional rule-based and human experience-based prediction methods, this invention can fully consider the dynamic relationship between the building and its environment, providing more accurate and reliable building lifespan predictions, more accurately simulating the impact of environmental factors on the remaining lifespan of garden buildings, and improving prediction accuracy and robustness. This provides more precise support for the intelligent management and decision-making of garden buildings. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the steps of a method for predicting the remaining lifespan of garden buildings based on environmental factors.

[0062] Figure 2 This is a structural diagram of a model for predicting the remaining lifespan of garden buildings.

[0063] Figure 3 Structure diagram of the high-dimensional feature generation module.

[0064] Figure 4 This is a structural diagram of the space factor module.

[0065] Figure 5 The image shows the fitting effect of the prediction model on the remaining lifespan of the garden building. Detailed Implementation

[0066] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Please see Figures 1-5 This invention provides a technical solution: a method for predicting the remaining lifespan of garden buildings based on environmental factors, comprising a data collection and feature generation module, a spatial factor module, a spatiotemporal perturbation enhancement module, and a prediction module. First, by collecting relevant data on the environment in which the garden building is located, a multidimensional feature dataset is constructed, and the features are optimized using a dynamic weighting mechanism to generate a high-dimensional feature dataset. Then, the spatial factor module models the spatiotemporal dependency between the building and its environment, further enhancing the expressive power of the features. In the spatiotemporal perturbation enhancement module, perturbation factors and a multi-level neighborhood update mechanism are introduced to capture the complex relationships between features. Finally, a regression model is trained, using the training set to learn the relationship between the building's environmental features and its remaining lifespan. Ultimately, the predicted remaining lifespan of the garden building is obtained through the prediction set. Specific steps are as follows: Figure 1 As shown.

[0068] Construct a model for predicting the remaining lifespan of garden buildings, the structure of which is as follows: Figure 2 As shown, the specific steps are as follows:

[0069] S1. Collect relevant environmental data of the garden buildings and construct a multi-dimensional feature dataset, including temperature, humidity, wind speed, solar radiation intensity and soil moisture.

[0070] Furthermore, the dataset of this invention contains environmental-related data of 1000 garden buildings, and the dataset is divided into training set and test set in a 7:3 ratio.

[0071] S2. Optimize the original multidimensional feature dataset using a dynamic weighting mechanism. Construct a high-dimensional feature dataset as the first dataset through multi-level feature combination and feature enhancement. The dynamic weighting mechanism is based on the building health status H. j (t) and the environmental context C k (t) Design a weight update strategy to dynamically adjust the weights between features and adaptively weight different features.

[0072] Furthermore, the original multidimensional feature dataset is optimized to generate the first dataset, whose structure is as follows: Figure 3 As shown, based on changes in building health status and environment, a weight update strategy is designed to dynamically adjust the weights of building health status and environmental context. The mathematical model of the weight update strategy is as follows:

[0073]

[0074] In the formula, E is the Sigmoid function, used to map the calculation results to the interval [0,1], representing the influence of each environmental factor, and H... j (t) represents the health status characteristics of the garden buildings, indicating the current health status of the buildings, θ ij and α ik C represents the weighting coefficients, indicating the relationship between environmental factors and the health status and environmental context of the garden buildings. Initially set to 0, these weights are adaptively adjusted during training, increasing when the health status and environmental context have a significant impact on the environmental factors. k (t) represents the environmental context. Through dynamic weighting, the model can better identify and focus on features that have a significant impact on the building's lifespan, thereby improving prediction accuracy. The weighted environmental data can be represented as:

[0075]

[0076] In the formula, E i (t) represents the environmental factor, indicating the i-th environmental factor collected at time t, E ω (t) represents weighted environmental data, indicating the relationships between task states, ω i (·) represents the weighting coefficient, indicating the impact of environmental factors on the health status of garden buildings. Then, high-dimensional features are generated from the weighted environmental data, building health status, and environmental context. The mathematical model is as follows:

[0077] G(t)=T(E ω (t),H(t),C(t))=W1·E ω (t)+W2·H(t)+W3·C(t)+b;

[0078] In the formula, G(t) represents the generated high-dimensional feature data, indicating the complex features that, after transformation, can reflect the environmental impact and building health status during the aging process of garden buildings. W1, W2, and W3 are weight matrices, and b is a bias term, initialized to 0, representing the offset of the generation function. Finally, noise filtering is performed. The mathematical model is as follows:

[0079] E f (t)=E ω (t)·(1-λ(t));

[0080] In the formula, E fλ(t) represents the environmental data after noise filtering, and λ(t) is the dynamically calculated noise filtering coefficient. The initial value can be set to 0.1, indicating that the initial noise is relatively low. During model training, the noise filtering intensity will be dynamically adjusted according to changes in environmental factors and building health status. The mathematical model is as follows:

[0081]

[0082] In the formula, β i and γ j To adjust the parameters of noise filtration intensity, representing the influence of environmental factors and building health status on noise filtration intensity, δ k The environmental context influences the noise filtering intensity, resulting in the first dataset. The implementation code is as follows:

[0083]

[0084]

[0085]

[0086] S3. Construct the spatial factor module, and calculate the spatial factor S for each feature through local windowing and global compensation. t，k The modeling of the spatiotemporal dependencies between buildings and the environment generates a second dataset.

[0087] Furthermore, the spatial factor module structure is as follows: Figure 4 As shown, for each garden building feature k at time t, a local sliding window is designed to calculate the local spatial factor F of feature k. local，t，k The mathematical model is:

[0088]

[0089] In the formula, X t，n The environmental factors that affect a building at different points in time. N is the weighting coefficient between feature k and its neighbor features n, initially set to 0.45. k Given the set of neighboring buildings of feature k, then calculate the global spatial factor F based on the local spatiotemporal factors. global，t To ensure that the impact of global environmental factors is comprehensively considered, local spatial factors are compensated by using the average environmental characteristics of all garden buildings. The mathematical model is as follows:

[0090]

[0091] In the formula, N is the total number of features, which is 15, including environmental features and features of the building itself.

[0092] Furthermore, through an adaptive weight adjustment mechanism, the influence weights of spatial factors are dynamically adjusted, enabling the model to automatically optimize the weights of spatial factors based on the correlation between different features. This ensures more accurate feature fusion under different environmental conditions. The mathematical model is as follows:

[0093]

[0094] In the formula, To control the hyperparameter of environmental similarity sensitivity, the initial value is set to 0.5. Then, by fusing compensation from local and global spatial factors, the final spatial factor S of feature k at time t is generated. t，k To ensure that the spatiotemporal relationship between architectural features and their surrounding environment is fully considered, the mathematical model is as follows:

[0095] S t,k =F local,t,k +Ψ t，k ·F global,t

[0096] Finally, denormalization is used to transform the spatial factors back into the original environmental feature space of the building, restoring the generated spatial factors to the original feature space, ensuring that all calculation results are consistent with the actual data of the building and environment. The mathematical model is as follows:

[0097] In the formula, S is the standard deviation matrix of the feature space. The final result is the second dataset. The implementation code for the spatial factor module is as follows: #Step 1: Calculate the local spatial factor

[0098]

[0099] #Step 2: Calculate the global spatial factor

[0100] def calculate_global_space_factor(X_env,N):

[0101] F_global = sum(X_env) / N # Average of all building environment factors

[0102] return F_global

[0103] #Step 3: Adaptive Weight Adjustment Mechanism

[0104]

[0105] #Step 4: Final Space Factor Generation

[0106]

[0107] #Step 5: Denormalization

[0108]

[0109] S4. Construct a spatiotemporal perturbation enhancement module to update and optimize the feature state. Introduce a fixed perturbation factor η to simulate the randomness of the data. At the same time, design a multi-level neighborhood update mechanism to calculate the spatial and temporal similarity of the features. Introduce a nonlinear iterative update mechanism to update the feature state and generate a third dataset, which is divided into a training set and a prediction set.

[0110] Furthermore, a perturbation factor is first introduced to simulate the randomness of building features and environmental changes, and the complex relationships between features are simulated to construct the spatiotemporal perturbation enhancement module. The mathematical model is as follows:

[0111]

[0112] In the formula, To introduce nonlinear perturbations and the resulting characteristics, η is a fixed perturbation factor, initially set to 0.1, indicating that the impact of the nonlinear perturbation is small in the initial stage and can be dynamically adjusted through training. τ tk Let be the interaction weight between feature k and feature n, initially set to 1, indicating that the similarity weight between each feature and its neighboring features is equal. Then, a multi-level neighborhood update mechanism is designed to capture the spatiotemporal dependence of buildings on their surrounding environment. The neighborhood influence is determined through multiple levels of integration, so that the neighborhood weight of each node is dynamically adjusted according to node similarity and global importance. The similarity between features is calculated, including spatial similarity and temporal similarity. The mathematical model for calculating spatial similarity is as follows:

[0113]

[0114] The mathematical model for calculating time similarity is:

[0115]

[0116] Combining spatial similarity and temporal similarity, the mathematical model is as follows:

[0117]

[0118] In the formula, ζ is a hyperparameter used to control different weights of similarity. The initial value is set to 0.5, indicating that the weights of spatial similarity and temporal similarity are equal. Finally, the multi-level weighted average of the neighborhood weights is calculated. The mathematical model is as follows:

[0119]

[0120] In the formula, distance(k, n) is the physical distance between feature k and feature n, ρ is a hyperparameter controlling the influence of the physical distance between features in the weighting, and its initial value is set to 0.8. k The mathematical model for measuring feature importance is as follows:

[0121]

[0122] In the formula, deg(k) is the degree of feature k. To capture the global mean feature, a nonlinear iterative mechanism is introduced to further capture complex nonlinear relationships, enabling the model to gradually adjust the state of nodes. The mathematical model for updating the rule iteration is as follows:

[0123] In the formula, ξ1 is a hyperparameter controlling the update magnitude of each iteration, with an initial value of 0.5; ξ2 is the weight coefficient for neighborhood updates, with an initial value of 0.5. The third dataset is then obtained and divided into a training set and a prediction set in a 7:3 ratio. The implementation code is as follows:

[0124] # Define the disturbance factor and introduce the function

[0125] def perturbation(F_denormalized,alpha,F_denormalized_neighbors,beta):

[0126] perturbation_result = F_denormalized + alpha * F_denormalized # Fixed perturbation factor # For neighbor in F_denormalized_neighbors: # Add neighbor interactions

[0127] perturbation_result+=beta*F_denormalized*neighbor

[0128] return perturbation_result

[0129] # Calculate spatial similarity between features

[0130] def space_similarity(F_denormalized_k,F_denormalized_n):

[0131] # Calculate spatial similarity using Euclidean distance

[0132] return 1 / (1+np.linalg.norm(F_denormalized_k-F_denormalized_n)**2)

[0133] # Calculate the temporal similarity between features

[0134] def time_similarity(F_denormalized_k,F_denormalized_k_minus1,F_denormalized_n):

[0135] return 1 / (1+np.linalg.norm(F_denormalized_k-F_denormalized_k_minus1)**2+np.linalg.norm(F_denormalized_k-F_denormalized_n)**2)

[0136] #Combined similarity of spatial and temporal similarity

[0137] def combined_similarity(space_sim,time_sim,zeta):

[0138] return zeta*space_sim+(1-zeta)*time_sim

[0139] #Multi-level weighting for calculating neighborhood weights

[0140] def multi_level_weight(F_denormalized_k,F_denormalized_n,physical_distance,delta,degree_k,global_mean_feature):

[0141] return(1 / (1+physical_distance**delta))*degree_k / (degree_k+global_mean_feature)

[0142] #Nonlinear Iterative Mechanism: Updating Feature States

[0143] def nonlinear_iteration(F_denormalized_k,perturbation_factor,learning_rate,neighborhood_weight):

[0144] return F_denormalized_k+learning_rate*(perturbation_factor*F_denormalized_k-F_denormalized_k)+learning_rate*neighborhood_weight*F_denormalized_k.

[0145] S5. The training set is input into the garden building remaining life prediction model. The garden building remaining life prediction model adopts a regression model. The training set is used to train the model and fit the data to learn the relationship between the environmental factors of the building and the remaining life of the building.

[0146] Furthermore, the training set was input into the remaining lifespan prediction model for garden buildings. The model uses the PyTorch deep learning framework, runs on a Linux operating system, and is accelerated using an NVIDIA V100 32GB GPU. During training, the batch size was set to 64. Using a regression method, the remaining lifespan of the buildings is learned based on the environmental characteristics and building health status in the training set. The core objective of the regression model is to establish the relationship between environmental characteristics and the remaining lifespan of the buildings. The mathematical model for predicting the remaining lifespan of garden buildings is as follows:

[0147]

[0148] In the formula, f(·,θ) is the regression function. Let θ represent the building's remaining lifespan predicted by the model, and θ be the parameters of the regression model, representing the weights of environmental factors and building health status on the remaining lifespan prediction. During training, the model is trained by minimizing a loss function, which uses mean squared error. The mathematical model is as follows:

[0149]

[0150] In the formula, M is the number of samples in the training set, and X... k Let θ represent the actual remaining life of the building, i.e., the labels in the training set. Loss(θ) is the loss function, representing the error between the model's predictions and the actual values. The loss function is optimized using backpropagation, and the model parameters θ are gradually adjusted to reduce the prediction error. Gradient descent is used to optimize the parameters. The mathematical model is as follows:

[0151]

[0152] In the formula, θ i Here, represents the model parameters in the i-th iteration, and u is the learning rate, controlling the step size for each update. The initial value is set to 0.001. The gradient of the loss function with respect to the parameter θ is used to obtain the prediction model for the remaining lifespan of garden buildings through multiple rounds of training and adjustment.

[0153] S6. The prediction set is input into the trained garden building remaining life prediction model, and the final output is the predicted value of the garden building's remaining life.

[0154] Furthermore, the remaining life prediction model for garden buildings achieves the following fitting effect diagram: Figure 5 As shown in the figure, the horizontal axis represents the date, the vertical axis represents the predicted remaining life, the gray dashed line and crosses represent the actual values, and the black solid line and dots represent the predicted values. It can be seen from the figure that the changing trends of the actual values ​​and the predicted values ​​are roughly similar, and the two change in the same direction. The experimental results show that the remaining life prediction model for garden buildings can effectively capture the changing trend of the remaining life of buildings, especially in the period of large fluctuations, it can predict the remaining life well.

Claims

1. A method for predicting the remaining lifespan of garden buildings based on environmental factors, characterized in that, Includes the following steps: S1. Collect relevant environmental data of the garden buildings and construct a multi-dimensional feature dataset, including temperature, humidity, wind speed, solar radiation intensity and soil moisture; S2. Optimize the original multidimensional feature dataset using a dynamic weighting mechanism. Construct a high-dimensional feature dataset as the first dataset through multi-level feature combination and feature enhancement. The dynamic weighting mechanism is based on building health status. and environment context The design employs a weight update strategy to dynamically adjust the weights among various features, adaptively weighting different features. S3. Construct a spatial factor module to calculate the spatial factor of each feature through local windowing and global compensation. Model the spatiotemporal dependencies between buildings and the environment and generate a second dataset. Design a local sliding window to calculate the spatiotemporal dependency of each feature, and then calculate the spatiotemporal dependency based on each feature. In time Changes on the surface, calculation features Local space factor The spatial factor model is constructed as follows: ； In the formula, For features Related environmental factors, Features and neighbor characteristics The weighting coefficients between them The feature set of the building is then used to calculate the global spatial factor. The mathematical model is: ； In the formula, To determine the total number of features, an adaptive weight adjustment mechanism is proposed to automatically optimize the weights of local and global spatial factors. The mathematical model is as follows: ； In the formula, To control the hyperparameters of sensitivity to environmental similarity, features are generated by fusing compensations from local and global spatial factors. In time The final space factor Finally, the spatial factors are reduced through denormalization. Transforming back to the original environmental characteristics of the building, we obtain This yields the second dataset; S4. Construct a spatiotemporal perturbation enhancement module, update and optimize feature states, and introduce a fixed perturbation factor. The randomness of the simulated data is simulated, and a multi-level neighborhood update mechanism is designed to calculate the spatial and temporal similarity of features. A nonlinear iterative update mechanism is introduced to update the node state and generate a third dataset, which is divided into a training set and a prediction set. S5. The training set is input into the garden building remaining life prediction model. The garden building remaining life prediction model adopts a regression model. The training set is used to train the model and fit the data to learn the relationship between the environmental factors of the building and the remaining life of the building. S6. The prediction set is input into the trained garden building remaining life prediction model, and the final output is the predicted value of the garden building's remaining life. .

2. The method for predicting the remaining lifespan of garden buildings based on environmental factors according to claim 1, characterized in that, The specific process of constructing the dynamic weighting mechanism is as follows: based on the building health status and environment context The design of a weight update strategy dynamically adjusts the weights of different environmental factors. The mathematical model of the weight update strategy is as follows: ； In the formula, For the Sigmoid function, and These are the weighting coefficients. To establish an environmental context, the impact of environmental factors on building health is represented by weighting environmental data, resulting in weighted environmental data. .

3. The method for predicting the remaining lifespan of garden buildings based on environmental factors according to claim 2, characterized in that, The method for constructing the first dataset in S2 is as follows: using a feature generation function to weight the environmental data. Combined with building health status and environment context Generate high-dimensional feature data The mathematical model is: ； In the formula, For the generated high-dimensional feature data, , and This is the weight matrix. As a bias term, an adaptive noise filtering mechanism is then introduced for noise filtering. The noise filtering intensity is adjusted according to changes in environmental factors, building health status, and environmental context by dynamically calculating the noise filtering coefficient, ultimately obtaining the noise-filtered environmental data. .

4. The method for predicting the remaining lifespan of garden buildings based on environmental factors according to claim 3, characterized in that, Introducing a fixed disturbance factor The spatiotemporal perturbation enhancement module is constructed based on the randomness of the simulated data, by representing each feature. Nonlinear perturbation processing is performed, considering the nonlinear interaction between building health status and environmental factors, to obtain... Then, a multi-level neighborhood update mechanism is designed to calculate the similarity between features, including spatial similarity. and time similarity By fusing information at multiple levels, the aging process space of each feature is comprehensively evaluated, and spatial similarity and temporal similarity are combined to obtain the comprehensive similarity of each feature. Multi-level weighting is obtained by calculating neighborhood weights using a multi-level weighting mechanism. Finally, a nonlinear iterative update mechanism is introduced to update the feature states to obtain the third dataset. Each iteration adjusts the update rules to control the feature update magnitude. The mathematical model is as follows: ； In the formula, To control the hyperparameters for updating the magnitude of each iteration, The weight coefficients for neighborhood updates ultimately divide the third dataset into a training set and a prediction set.

5. The method for predicting the remaining lifespan of garden buildings based on environmental factors according to claim 4, characterized in that, The remaining lifespan of garden buildings is predicted using regression functions, and the loss function is minimized using mean squared error. Training is performed, and the loss function is optimized through backpropagation algorithm to gradually adjust the model parameters. The gradient descent algorithm is used to optimize the parameters. After multiple rounds of training and adjustment, the trained model for predicting the remaining lifespan of garden buildings is obtained and used as input for subsequent prediction sets. Finally, the model outputs the predicted remaining lifespan values ​​of the garden buildings. .

Images