Artificial intelligence-based climate adaptive elderly care and health care building design optimization system

Abstract

This invention discloses an AI-based climate-adaptive elderly care and health building design optimization system, relating to the fields of building design optimization and smart health management technology. It includes a state assessment module, a counterfactual diagnosis module, and an evolution path module. The system acquires building operational characteristics and dynamic meteorological parameters for sample augmentation and nonlinear mapping training. It locates failure points in the set of design variables for health and health buildings, and uses counterfactual path optimization to pinpoint the minimum intervention sequence. It then performs causal contribution measurement and semantic encapsulation to generate a decision evolution path. This invention can accurately locate environmental failure risks. Using the architectural design standards for elderly care facilities as constraints, it coordinates the control of winter solstice sunshine duration, summer shading needs, and indoor thermal environment in elderly activity spaces. This achieves a balance between sunshine compliance, age-friendly thermal safety, and environmental load in health and health buildings, improving the living comfort and climate adaptability of elderly care buildings.

Description

Technical Field

[0001] This invention relates to the fields of architectural design optimization and smart health and wellness management technology, and in particular to an artificial intelligence-based system for optimizing climate-adaptive elderly care and wellness building design. Background Technology

[0002] In recent years, with the dramatic fluctuations in global climate change and the deepening of my country's aging population, the climate adaptability of senior living buildings has become crucial for ensuring the health and well-being of the elderly and improving their quality of life. The elderly population has relatively fragile physiological functions and low tolerance to fluctuations in environmental temperature and humidity, requiring senior living buildings to possess extremely high environmental regulation and risk mitigation capabilities.

[0003] Currently, traditional methods for optimizing climate-adaptive building design primarily rely on steady-state simulations or static standard guidance. However, in practical engineering applications, these methods still suffer from the following significant drawbacks:

[0004] First, there is a scarcity of measured data and incomplete coverage of operating conditions. Due to the high cost and long cycle of data collection for the operation of health and wellness buildings, the amount of data samples used to train optimization models is often severely insufficient. Traditional simulation methods struggle to capture the nonlinear environmental patterns under extreme weather conditions, leading to inaccurate predictions from optimization systems when faced with sudden weather stresses due to a lack of typical sample support.

[0005] Second, the uncertainty of model predictions lacks quantification methods. Existing design optimization systems mostly employ deterministic mapping logic, which can only output single performance prediction values ​​and cannot detect cognitive biases in the model when dealing with complex and heterogeneous data. In a health and wellness environment, this lack of "confidence" in predictions may prevent the system from providing timely warnings of potential thermal safety risks.

[0006] Third, optimization strategies lack economic trade-offs and causal explanations. Traditional optimization algorithms (such as genetic algorithms) tend to seek the global optimum, often suggesting significant changes to the building's main structure while ignoring cost constraints (the principle of minimum intervention) during the renovation of existing buildings. Furthermore, the optimization decision-making process is often a "black box," making it difficult for architects to clearly define the contribution ratio of each design parameter to performance improvement, resulting in a lack of practical guidance and credibility in the final recommended solutions. Summary of the Invention

[0007] The technical problem addressed by this invention is that existing technologies struggle to balance high-performance prediction accuracy with low-cost intervention in the design optimization of senior living buildings. Firstly, due to the high cost and long timeframe for acquiring measured data for senior living buildings, existing technologies cannot accurately capture the nonlinear environmental patterns under extreme weather stress using a small sample size. This leads to inaccurate predictions in optimization models when dealing with sudden environmental risks due to a lack of typical sample support. Secondly, existing design optimization systems often employ deterministic mapping logic, making it difficult to perceive and quantify cognitive uncertainties in the prediction process. When faced with interference such as sensor failure or data heterogeneity, they cannot provide decision support with risk probability assessment, thus failing to guarantee thermal safety tolerance for the vulnerable physiological functions of the elderly. Finally, traditional optimization algorithms often tend to seek the global optimum, typically suggesting significant alterations to the building's main structure, ignoring cost constraints and the "minimum intervention" principle in the renovation process. Furthermore, the optimization decision-making process is often a "black box," making it difficult for architects to clearly understand the true causal contribution of each design parameter to performance improvement. This results in the final recommended scheme lacking practical guidance and credibility for construction implementation.

[0008] To address the aforementioned technical problems, this invention provides the following technical solution: an artificial intelligence-based climate-adaptive elderly care and health building design optimization system, comprising a state assessment module, a counterfactual diagnosis module, and an evolution path module;

[0009] The state assessment module is used to obtain the initial design scheme, basic operational characteristics of the health and wellness building and dynamic meteorological parameters. It performs sample augmentation, nonlinear mapping training and contribution analysis on the basic operational characteristics of the health and wellness building and dynamic meteorological parameters to obtain the performance proxy model and variable guidance matrix.

[0010] The counterfactual diagnosis module is used to perform action space abstraction and cost weight allocation on the set of design variables for health and wellness buildings, to obtain a discretized action set and action cost weight matrix, and input the initial design scheme into the performance proxy model for failure site location and counterfactual path optimization to obtain the minimum intervention action sequence.

[0011] The evolution path module is used to measure the causal contribution of the minimum intervention action sequence, extract logical association records, and encapsulate temporal semantics to obtain a decision evolution path diagram.

[0012] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system of the present invention, the state assessment module includes a sample enhancement unit, a nonlinear mapping unit, and a contribution analysis unit.

[0013] The sample augmentation unit is used to obtain the initial design scheme, basic operational characteristics of health and wellness buildings, and dynamic meteorological parameters. Through generative adversarial networks, the basic operational characteristics of health and wellness buildings and dynamic meteorological parameters are simulated for feature distribution and expanded for sampling to obtain a hybrid dataset.

[0014] The generative adversarial network includes a generation module and a discrimination module;

[0015] The nonlinear mapping unit is used to establish a weight estimation model between the set of design variables and the set of performance targets of health and wellness buildings based on a hybrid dataset. A Gaussian prior distribution is introduced to perform posterior probability estimation and hyperparameter update processing on the weights in the weight estimation model to obtain a performance proxy model.

[0016] The set of design variables for health and wellness buildings includes interface physical parameters, boundary morphology parameters, and energy supply logic parameters.

[0017] The performance target set includes environmental load indicators, age-friendly thermal safety evaluation indicators, and solar radiation compliance indicators;

[0018] The performance proxy model is used to output the distribution of the performance target set after the design variable changes, and the improvement of the distribution relative to the preset bottom line constraint is defined as the prediction gain.

[0019] The contribution analysis unit is used to perform global sensitivity analysis on the performance proxy model, calculate the contribution rate score of each health and wellness building design variable to the performance target set, sort the health and wellness building design variables in descending order of contribution rate score, and obtain the variable guidance matrix.

[0020] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system described in this invention, the processing logic of feature distribution simulation and extended sampling includes:

[0021] The basic operational characteristics and dynamic meteorological parameters of health and wellness buildings are extracted and mapped to a unified high-dimensional feature space in combination with the design variable set of health and wellness buildings in the initial design scheme, which serves as training samples for generative adversarial networks.

[0022] The joint probability distribution between design variables and dynamic meteorological parameters in the training samples is learned by the generation module, and a virtual coupled feature distribution is generated.

[0023] The discriminant module is used to verify the probability distribution of the virtual coupling feature distribution against the training samples to obtain the discriminant loss value. The distribution parameters of the generator module are then iteratively updated based on the discriminant loss value to obtain the trained generator module.

[0024] The trained generation module is augmented with sampling to obtain virtual coupled feature samples, which are then merged with the training samples to obtain a hybrid dataset.

[0025] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system of the present invention, the processing logic of the nonlinear mapping unit is as follows:

[0026] The design variables of health and wellness buildings in the mixed dataset are mapped to the input vector of independent variables, and the set of performance targets is defined as the output target of the dependent variable.

[0027] A Gaussian prior distribution is introduced to set the initial uncertainty range for the weights in the weight estimation model, and a likelihood function is constructed using the input vector of independent variables and the output target of dependent variables.

[0028] The influence of observed data on the weight distribution is calculated by the likelihood function, and the coupling weights between variables are updated by the posterior probability estimation method to obtain the updated coupling weights.

[0029] The mean of the predicted distribution of the input vector of independent variables under the corresponding climate parameters is calculated based on the updated coupling weights, and the performance surrogate model is output.

[0030] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system described in this invention, the counterfactual diagnosis module includes an action abstraction unit, a cost measurement unit, a failure site location unit, and a path optimization unit.

[0031] The action abstraction unit is used to transform the set of design variables for health and wellness buildings into a set of discretized actions based on the variable guidance matrix. The set of discretized actions includes actions for adjusting interface physical parameters, actions for changing boundary morphology, and actions for reconstructing functional logic.

[0032] The cost measurement unit is used to establish a truncation criterion for each discretized action in the discretized action set, and to assign a corresponding preset cost weight value based on the truncation criterion to obtain the action cost weight matrix.

[0033] The failure site location unit is used to input the initial design scheme into the performance proxy model, identify the spatiotemporal distribution characteristics of the performance target set that do not meet the preset aging-friendly bottom line constraints, extract the spatial coordinates of the design variables corresponding to the failure time characteristics, define them as failure sites, and obtain the initial state vector of the design variables corresponding to the failure sites.

[0034] The path optimization unit is used to take the failure site identified by the failure site location unit as the starting node, and the discretized action set as the evolution branch. Within the limited action distance constraint domain defined by the action cost weight matrix, it calls the enhanced path-finding algorithm to extract the counterfactual candidate set of performance target set that meets the preset age-friendly bottom line constraint. The ordered action chain is locked by the cumulative evaluation cost calculation to obtain the minimum intervention action sequence.

[0035] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system described in this invention, the processing logic of the cost measurement unit is as follows:

[0036] Get the pending processing number Each action is discrete, and the corresponding preset cost weight value is retrieved from the action cost weight matrix;

[0037] When the discretization action is a parameter adjustment action that does not change the main structure, the first weight coefficient is assigned;

[0038] When the discretization action involves a change in the geometry of a component, a second weighting coefficient is assigned;

[0039] When the discretization action is a reconstruction action involving spatial functional adjacency relationships, a third weight coefficient is assigned;

[0040] When the discretization action causes the solar radiation compliance index to fall below the preset standard threshold, a penalty level weight coefficient is assigned;

[0041] Among them, the penalty level weight coefficient is greater than the third weight coefficient, the third weight coefficient is greater than the second weight coefficient, and the second weight coefficient is greater than the first weight coefficient;

[0042] Get the The execution status codes of each discretized action are used to calculate the sum of the products of the preset cost weight values ​​corresponding to each discretized action, thus obtaining the action distance.

[0043] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system described in this invention, the processing logic of the path optimization unit is as follows:

[0044] Within the manifold space defined by the discretized action set, a design space resistance diagram is constructed, and resistance cut-off walls are set in the design space resistance diagram according to the cut-off criterion.

[0045] The predicted gain output by the performance proxy model is used as a heuristic benefit guidance function to calculate the total evaluation cost of the evolution branch node;

[0046] The path with the minimum total evaluation cost is identified through iterative search.

[0047] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system of the present invention, the evolution path module includes a causal analysis unit and a prescription generation unit.

[0048] The causal analysis unit is used to extract the performance mapping components corresponding to each discretized action in the minimum intervention action sequence and calculate the causal contribution value of each discretized action.

[0049] The prescription generation unit is used to summarize the minimum intervention action sequence according to logical time sequence and perform semantic encapsulation processing to generate an intervention prescription for the target environmental stress scenario.

[0050] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system described in this invention, the formula for calculating the causal contribution value of each discretized action is as follows:

[0051] ;

[0052] in, For the first The causal contribution value of each discretized action. For discretized action sets, To extract from the discretized action set the action that does not contain the first action Any subset of discretized actions, For performance proxy models in action subsets The predicted output value is as follows. This represents the number of discretized actions contained in the action subset. This represents the total number of discretized actions contained in the discretized action set. The performance target value is the value predicted by the performance proxy model.

[0053] As a preferred embodiment of the AI-based climate-adaptive elderly care and health building design optimization system of the present invention, the evolution path module further includes a logic diagram output unit.

[0054] The logic diagram output unit is used to connect the scheme state nodes before and after the action execution, with the initial design scheme as the root node and the discrete actions in the minimum intervention action sequence as logical branches. It also marks the corresponding action distance and causal contribution value on the logical branches to generate a decision evolution path. Its processing logic is as follows:

[0055] Using the initial design scheme as the root node, and following the execution sequence of the minimum intervention action sequence, each discretized action is abstracted into a logical branch connecting the state nodes of the preceding and following schemes, thus constructing a directed acyclic graph that evolves from the initial failure state to the performance target state.

[0056] Extract the action distance and causal contribution value of the discretized action corresponding to each logical branch, and use them as weight labels to map to the corresponding logical branch;

[0057] Perform topology layout processing on the directed acyclic graph to obtain the decision evolution path graph;

[0058] The decision evolution path diagram includes an evolution path state feature sequence, an ordered set of intervention actions, and cumulative disturbance component values.

[0059] The beneficial effects of this invention are as follows: This application achieves high-quality enhancement under small sample data through generative adversarial networks, effectively solving the problem of scarce measured data for elderly care buildings and significantly reducing computational costs and time. Simultaneously, the introduction of a Gaussian prior distribution probability prediction mechanism endows the performance proxy model with the ability to quantify uncertainty, enabling it to issue reliable early warnings in the face of extreme or rare climate fluctuations and output robust predictive gains, thus possessing a higher safety tolerance rate when ensuring key indicators such as age-friendly thermal safety. By establishing a weight matrix with a penalty term through a cost metric model, the generated decision path prioritizes the action combination with the lowest cost and least disturbance to existing buildings, effectively resolving the contradiction between performance optimization and regulatory conflicts, as well as the cost of renovation. Combined with the introduced Shapley value calculation formula, this invention can overcome the nonlinear barriers in building physics processes, accurately extracting the true causal contribution of each action to the repair of environmental failures, eliminating coupling interference between variables and the phenomenon of "credit fraud," and providing a rigorous evaluation index with priority reference for prescription generation. Through logical graph output units and semantic encapsulation processing, abstract AI algorithm decisions are transformed into intuitive, expert-level evidence-chain-based visual path diagrams and structured intervention prescriptions. This approach transforms complex discrete coding into industry-standard professional terminology, bridging the technical gap between computer languages ​​and engineering practices. It enables non-professionals to clearly grasp the complete logic of the evolution from the initial failure state to the performance target state, greatly enhancing the guiding value and widespread application significance of climate adaptability optimization technology in the field of health and wellness buildings. Attached Figure Description

[0060] Figure 1 A schematic diagram of the basic process of an artificial intelligence-based climate-adaptive elderly care and health building design optimization system provided in one embodiment of the present invention;

[0061] Figure 2 This is a schematic diagram of the internal processing flow of the status assessment module;

[0062] Figure 3 This is a schematic diagram of the counterfactual diagnosis module's processing flow;

[0063] Figure 4 This is a schematic diagram of the evolution path module processing flow. Detailed Implementation

[0064] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0065] Example, refer to Figure 1 It provides an AI-based climate-adaptive elderly care and health building design optimization system, including a status assessment module, a counterfactual diagnosis module, and an evolution path module;

[0066] The status assessment module is used to obtain the initial design scheme, basic operational characteristics of health and wellness buildings, and dynamic meteorological parameters. It performs sample augmentation, nonlinear mapping training, and contribution analysis on the basic operational characteristics and dynamic meteorological parameters of health and wellness buildings to obtain the performance proxy model and variable guidance matrix.

[0067] The counterfactual diagnosis module is used to perform action space abstraction and cost weight allocation on the set of design variables for health and wellness buildings, to obtain a discretized action set and action cost weight matrix. The initial design scheme is then input into the performance proxy model for failure site location and counterfactual path optimization to obtain the minimum intervention action sequence.

[0068] The evolution path module is used to measure the causal contribution of the minimum intervention action sequence, extract logical association records, and encapsulate temporal semantics to obtain the decision evolution path diagram.

[0069] In practice, the state assessment module includes a sample enhancement unit, a nonlinear mapping unit, and a contribution analysis unit.

[0070] The sample augmentation unit is used to obtain the initial design scheme, basic operational characteristics of health and wellness buildings, and dynamic meteorological parameters. Through generative adversarial networks, the basic operational characteristics of health and wellness buildings and dynamic meteorological parameters are simulated for feature distribution and augmented sampling to obtain a hybrid dataset.

[0071] Generative adversarial networks consist of a generation module and a discriminator module;

[0072] The nonlinear mapping unit is used to establish a weight estimation model between the set of design variables and the set of performance targets of health and wellness buildings based on a mixed dataset. A Gaussian prior distribution is introduced to perform posterior probability estimation and hyperparameter update processing on the weights in the weight estimation model to obtain a performance proxy model.

[0073] The set of design variables for health and wellness buildings includes interface physical parameters, boundary morphology parameters, and energy supply logic parameters;

[0074] The performance target set includes environmental load indicators, age-friendly thermal safety evaluation indicators, and solar radiation compliance indicators;

[0075] The performance proxy model is used to output the distribution of the performance target set after the design variable changes, and the improvement of the distribution relative to the preset bottom line constraint is defined as the prediction gain;

[0076] The contribution analysis unit is used to perform global sensitivity analysis on the performance proxy model, calculate the contribution rate score of each health and wellness building design variable to the performance target set, sort the health and wellness building design variables in descending order of contribution rate score, and obtain the variable guidance matrix.

[0077] Specifically, the initial design scheme refers to the original digital baseline of the building before optimization, typically composed of a feature vector consisting of interface physical parameters, boundary morphological parameters, and energy supply logic parameters. The basic operational characteristics of senior living buildings refer to the physical feedback data generated during actual use, such as the temporal distribution of indoor temperature and humidity, the energy efficiency curves of senior living equipment, and load data during elderly activities. Dynamic meteorological parameters represent environmental stress variables outside the building, including hourly solar radiation intensity, outdoor dry-bulb temperature, relative humidity, and wind direction and speed. The processing logic for acquiring this data mainly relies on calling multiple source interfaces: the initial design scheme is extracted and normalized through the API interface of parametric modeling software (such as BIM); basic operational characteristics are obtained through IoT monitoring terminals deployed on-site or existing building operation and maintenance databases; dynamic meteorological parameters are obtained through standard meteorological data interfaces to acquire local typical meteorological year data or through real-time transmission from on-site micro-weather stations.

[0078] The Generative Adversarial Network (GAN) architecture employs a conditional generation model, consisting of a generation module and a discriminator module. The generation module is responsible for inputting random noise vectors, design variables for health and wellness buildings, and dynamic meteorological constraints. By learning the coupling correlation between design variables, dynamic meteorological parameters, and performance targets, it generates virtual coupled feature samples that conform to the physical logic of the building. These virtual coupled feature samples include the design variables, dynamic meteorological parameters, and corresponding performance target correlation features. The discriminator module is responsible for comparing the consistency probability of these virtual samples with real physical operational data. Through adversarial competition between the two modules, the generation module is forced to produce a hybrid dataset that more closely reflects real physical laws, thereby addressing the problem of scarce measured data for health and wellness buildings.

[0079] The processing logic for establishing the weight estimation model is based on a Bayesian inference framework. First, the design variables in the mixed dataset are mapped as independent input variables, the performance target is defined as the dependent output variable, and a Gaussian prior distribution is introduced to the model weights to set the initial uncertainty range. Then, the likelihood function is used to calculate the impact of observed data on the weight distribution, and the coupling weights between variables are iteratively updated using a posterior probability estimation method. Finally, a performance surrogate model that expresses the confidence level of the prediction results is output. Based on this, the processing logic for global sensitivity analysis of the performance surrogate model employs variance decomposition technology, sampling across the entire space of design variables and calculating the proportion of the total variance of the performance target caused by fluctuations in a single variable. After quantifying this proportion into contribution rate scores, the system sorts the variables in descending order of scores, generating a variable guidance matrix to identify the key variables with the greatest impact on age-friendly environments.

[0080] Compared to traditional exhaustive simulations based on static regression or time-consuming physics engines, this application achieves high-quality enhancement with small sample data through generative adversarial networks, significantly reducing computational costs and time. Simultaneously, the introduction of a probabilistic prediction mechanism based on Gaussian prior distribution enables the model to cope with the uncertainties brought about by climate fluctuations, outputting robust predictive gains. Accurate identification of key variables through contribution analysis ensures that the system generates optimized paths with minimal intervention in subsequent steps, avoiding the drawbacks of blindly undertaking large-scale demolition and alteration of the main building structure in traditional solutions. This achieves an optimal balance between renovation costs and environmental impact while ensuring age-friendly thermal safety and solar energy compliance.

[0081] In specific implementation, the processing logic for feature distribution simulation and expanded sampling includes:

[0082] The basic operational characteristics and dynamic meteorological parameters of health and wellness buildings are extracted and mapped to a unified high-dimensional feature space in combination with the design variable set of health and wellness buildings in the initial design scheme, which serves as training samples for generative adversarial networks.

[0083] The joint probability distribution between design variables and dynamic meteorological parameters in the training samples is learned by the generation module, and a virtual coupled feature distribution is generated.

[0084] The discriminant module is used to verify the probability distribution of the virtual coupling feature distribution against the training samples to obtain the discriminant loss value. The distribution parameters of the generator module are then iteratively updated based on the discriminant loss value to obtain the trained generator module.

[0085] The trained generation module is augmented with sampling to obtain virtual coupled feature samples, which are then merged with the training samples to obtain a hybrid dataset.

[0086] Specifically, the set of design variables for senior living buildings includes interface physical parameters (such as the thermal coefficient of the building envelope and the thermal inertia of the materials), boundary morphology parameters (such as the window-to-wall ratio, the size of the shading components, and the building azimuth), and energy supply logic parameters (such as the opening threshold of the air conditioning system and the frequency of fresh air exchange). These variables together constitute the human-controlled dimensions that affect the performance of senior living buildings.

[0087] The processing logic of mapping these heterogeneous data to a unified high-dimensional feature space is essentially a standardization and feature alignment process. The system first normalizes numerical design variables, time-series operational characteristics (such as indoor temperature), and meteorological parameters (such as solar radiation) to eliminate dimensional differences. Then, a multidimensional encoder concatenates these parameters into a high-dimensional feature vector. In this space, the physical properties of the building and the dynamic disturbances of the external environment are transformed into points in a coordinate system, enabling the generative adversarial network to capture the nonlinear relationships between them within a unified mathematical dimension.

[0088] The processing logic for learning the joint probability distribution in the generation module utilizes the fitting capability of deep neural networks. The processing logic for learning the joint coupling distribution in the generation module utilizes the nonlinear fitting capability of deep neural networks. The generation module typically consists of multiple transposed convolutional layers or fully connected layers, taking into account random noise vectors, design variable constraints, and dynamic meteorological condition constraints. The learning process approximates the hidden conditional probability density function in the training samples: P(dynamic meteorological parameters, performance target | design variables). Through the backpropagation algorithm, the generation module continuously adjusts the parameters of its internal neurons, thereby generating a virtual coupling feature distribution consistent with the actual building operation patterns. This virtual coupling feature distribution includes not only the dynamic meteorological parameters corresponding to the target design variables but also the performance target association features formed under corresponding meteorological conditions, used to construct a hybrid dataset that can directly participate in the training of the performance proxy model.

[0089] The processing logic of the discrimination module's verification and iterative update is a typical "adversarial game" process. The discrimination module, acting as a binary classifier, takes as input real training samples and a virtual coupled feature distribution. The similarity between the distributions is measured by calculating the cross-entropy loss between the two; this result is the discrimination loss value. The system uses the gradient information generated by this loss value to perform bidirectional optimization: on the one hand, it trains the discrimination module to become more sensitive and able to detect fake data; on the other hand, it feeds the gradient back to the generation module, prompting the generation module to modify its distribution parameters until the discrimination module can no longer distinguish between real and fake data. At this point, the trained generation module is obtained.

[0090] The system utilizes the trained generation module to perform augmented sampling by inputting different random noise seeds and dynamic meteorological disturbance conditions, while keeping the target design variable constraints unchanged. This generates virtual coupled feature samples covering extreme climate conditions and typical operating conditions. The virtual coupled feature samples include dynamic meteorological parameters, building operation characteristics, and corresponding performance target correlation features. Finally, the system performs feature fusion processing, concatenating the generated virtual coupled feature samples with the original training samples to obtain a hybrid dataset containing design variables, dynamic meteorological parameters, and performance target correlation features. This dataset is then used for supervised training of the performance proxy model.

[0091] In specific implementation, the processing logic of the nonlinear mapping unit is as follows:

[0092] The design variables of health and wellness buildings in the mixed dataset are mapped to the input vector of independent variables, and the set of performance targets is defined as the output target of the dependent variable.

[0093] A Gaussian prior distribution is introduced to set the initial uncertainty range for the weights in the weight estimation model, and a likelihood function is constructed using the input vector of independent variables and the output target of dependent variables.

[0094] The influence of observed data on the weight distribution is calculated by the likelihood function, and the coupling weights between variables are updated by the posterior probability estimation method to obtain the updated coupling weights.

[0095] The mean of the predicted distribution of the input vector of independent variables under the corresponding climate parameters is calculated based on the updated coupling weights, and the performance surrogate model is output.

[0096] Specifically, the processing logic that maps the design variables of senior living buildings in the mixed dataset to independent variable input vectors involves multidimensional feature encoding and standardization of the data. The system extracts interface physical parameters, boundary morphology parameters, and energy supply logic parameters, eliminates dimensional differences through feature normalization and one-heat encoding, and encapsulates them into tensors conforming to the input format of a neural network. Simultaneously, performance targets such as environmental load, age-friendly thermal safety, and solar radiation compliance are defined as dependent variable output targets, thus establishing a mathematical mapping foundation from the design space to the performance space.

[0097] Introducing a Gaussian prior distribution into the weights of the weight estimation model sets an initial uncertainty boundary for the model parameters. The system presupposes that the weights follow a Gaussian distribution with a mean of zero, using variance to characterize the cognitive ambiguity regarding the coupling relationship between design variables and performance targets in the initial state—that is, setting an initial uncertainty range. Subsequently, a likelihood function is constructed using the already constructed input vector of independent variables and the output target of the dependent variable. The probability density of the current mixed data sample is observed, transforming the laws of building physics into a statistical likelihood probability expression. In this embodiment, the presupposed weights are set to follow a Gaussian prior distribution with a mean of 0 and a small variance (e.g., 0.01 or 0.1). This setting does not provide a fixed value, but rather initializes a "unbiased" probability boundary with an uncertainty range for the system. This effectively prevents overfitting of the model with small sample data through mathematical regularization, ensuring that the weights always remain within a reasonable physical value range. More importantly, this distribution-based weight setting gives the performance proxy model the ability to quantify "uncertainty," allowing the system to characterize the confidence level of the prediction result through the magnitude of the variance while outputting the mean of the prediction distribution. When faced with extreme or rare climate fluctuations, the model can issue reliability warnings by expanding the output probability range, thus exhibiting higher robustness and safety tolerance than traditional fixed-weight deterministic models when ensuring key performance indicators such as age-friendly thermal safety.

[0098] The processing logic for calculating the impact of observed data on the weight distribution using the likelihood function follows the Bayesian posterior update principle. The system compares the predicted values ​​of the prior distribution with the actual observed values ​​in the mixed dataset, using the distribution density of the observed data to correct the bias of the prior distribution. Using posterior probability estimation methods, such as variational inference or Laplace approximation, the prior distribution is multiplied by the likelihood function and normalized, causing the probability distribution of the weights to shift and shrink towards the direction of the actual observed data, thus obtaining the updated coupling weights. This process essentially involves using data evidence to refine and correct the model's understanding, removing falsehoods and retaining truth. In this embodiment, the observed data refers to the sample data in the mixed dataset used to train the weight estimation model, including the health and wellness building design variables and dynamic meteorological parameters corresponding to the input vector of the independent variable, and the performance target set data corresponding to the output target of the dependent variable.

[0099] The processing logic for calculating the predicted distribution mean based on the updated coupled weights involves passing the probability distribution attribute of the weights to the output. For the input design vector, the system no longer calculates a single scalar result, but instead performs a full probability integration of the posterior weight distribution. By weighted averaging over all possible weight values, the predicted distribution mean of the independent variable under the corresponding climate parameters is calculated. The final output performance proxy model uses the mean as the core prediction reference and combines it with the distribution variance to characterize the confidence level of the prediction results, thereby achieving a probabilistic early warning of the performance failure risk of health and wellness buildings.

[0100] In practice, the counterfactual diagnosis module includes an action abstraction unit, a cost measurement unit, a failure point location unit, and a path optimization unit.

[0101] The action abstraction unit is used to transform the set of design variables for health and wellness buildings into a set of discretized actions based on the variable guidance matrix. The set of discretized actions includes actions for adjusting interface physical parameters, actions for changing boundary forms, and actions for reconstructing functional logic.

[0102] The cost measurement unit is used to establish a truncation criterion for each discretized action in the discretized action set, and to assign a corresponding preset cost weight value based on the truncation criterion to obtain the action cost weight matrix.

[0103] The failure site location unit is used to input the initial design scheme into the performance proxy model, identify the spatiotemporal distribution characteristics of the performance target set that do not meet the preset aging-friendly bottom line constraints, extract the spatial coordinates of the design variables corresponding to the failure time characteristics, define them as failure sites, and obtain the initial state vector of the design variables corresponding to the failure sites.

[0104] The path optimization unit uses the failure site identified by the failure site location unit as the starting node and the discretized action set as the evolution branch. Within the limited action distance constraint domain defined by the action cost weight matrix, it calls the enhanced pathfinding algorithm to extract the counterfactual candidate set of performance target set that meets the preset age-friendly bottom line constraint. By calculating the cumulative evaluation cost, it locks the ordered action chain and obtains the minimum intervention action sequence.

[0105] Specifically, the processing logic that transforms the design variables of health and wellness buildings into a set of discretized actions essentially involves differentiating and encapsulating actions within a continuous parameter space. The system first acquires the current values ​​of interface physical parameters, boundary morphology parameters, and power supply logic parameters. Based on a preset step size (e.g., a gradient every 5mm increase in insulation thickness) or on / off state (e.g., switching of air conditioning operation logic), the changes in these parameters are defined as independent "operators." Each operator is an "action," such as "increasing the external wall thermal resistance coefficient" or "reducing the south-facing window-to-wall ratio." Through this discretization process, the originally complex and continuous design adjustment process is transformed into a finite set of combinable action instructions, providing discrete decision nodes for subsequent graph-based path optimization.

[0106] The spatiotemporal distribution characteristics of preset age-friendly baseline constraints refer to the specific manifestations of performance targets (such as thermal safety indicators and solar radiation indicators) failing within the target time and space range. Spatially, this manifests as excessively low temperatures or insufficient sunlight in certain specific areas within the building (such as the north-facing bedroom for the elderly or the area near exterior windows); temporally, it manifests as performance indicators failing to meet the physiological safety thresholds (baseline constraints) that the elderly can withstand under the target climate scenario (such as the early morning of an extremely cold winter day or the afternoon of a hot summer day). This spatiotemporal distribution characteristic can accurately pinpoint "when and where" environmental failure occurred, rather than providing a vague overall assessment.

[0107] The processing logic for extracting failure sites and initial state vectors is a multi-dimensional coordinate locking process. The system inputs the initial design scheme into the performance proxy model for simulation. When the model's output prediction mean does not meet the preset baseline, the system immediately records the corresponding meteorological parameters, indoor location coordinates, and the values ​​of various building design variables at that moment. The system defines the intersection of this set of "time-space-performance values" as the failure site and simultaneously extracts the original values ​​of various design variables at that point (such as the current shading angle, wall heat transfer coefficient, etc.), encapsulating them into an initial state vector. The purpose of this logic is to determine a clear "logical starting point" and "initial boundary conditions" for subsequent counterfactual optimization.

[0108] Compared to traditional building design optimization based on genetic algorithms or full-scale iteration, the advantages of this application are mainly reflected in two aspects: First, "minimal intervention." Traditional optimization often involves rebalancing the entire building, which may lead to significant changes in the design. However, this application, through failure site location and action cost constraints, can lock in the lowest-cost and least disruptive action combination, protecting the stability of the existing design. Second, "interpretability of decisions." By transforming design optimization into a perceptible sequence of actions (discrete action set) and causal path (counterfactual optimization), architects can clearly see the specific contribution of each adjustment action to repairing environmental failures, thus making the optimization scheme more practical and valuable for implementation.

[0109] In practice, the processing logic of the cost measurement unit is as follows:

[0110] Get the pending processing number Each action is discrete, and the corresponding preset cost weight value is retrieved from the action cost weight matrix;

[0111] When the discretization action is a parameter adjustment action that does not change the main structure, the first weight coefficient is assigned;

[0112] When the discretization action involves a change in the geometry of a component, a second weighting coefficient is assigned;

[0113] When the discretization action is a reconstruction action involving spatial functional adjacency relationships, a third weight coefficient is assigned;

[0114] When the discretization action causes the solar radiation compliance index to fall below the preset standard threshold, a penalty level weight coefficient is assigned;

[0115] Among them, the penalty level weight coefficient is greater than the third weight coefficient, the third weight coefficient is greater than the second weight coefficient, and the second weight coefficient is greater than the first weight coefficient;

[0116] Get the Execution status codes of each discretized action The sum of the products of the preset cost weights corresponding to each discretized action is used to obtain the action distance, and the calculation formula is as follows:

[0117] ;

[0118] in, The distance of the action. The total dimension of candidate actions, The preset cost weight value;

[0119] When the When a discretization action is selected and executed The value is 1;

[0120] When the When a discretization action is not executed in parallel The value is 0.

[0121] Specifically, the processing logic for retrieving preset cost weight values ​​from the action cost weight matrix is ​​a dynamic mapping process based on action attribute tags. The system first identifies the physical attributes of the action to be processed and matches them with four preset weight levels: if it involves only non-structural adjustments to equipment or material parameters, it is mapped to the first weight coefficient; if it involves geometric changes such as window dimensions or eaves length, it is mapped to the second weight coefficient; and if it involves adjustments to functional layout or floor plan logic, it is mapped to the third weight coefficient. This retrieval mechanism achieves a quantitative transformation from "physical alteration" to "resource consumption," enabling the system to perceive the implementation difficulty behind each design change.

[0122] In this embodiment, the preset cost weight values ​​are set using orders-of-magnitude differences. For example, the first weight coefficient (parameter adjustment) is set to 1, the second weight coefficient (geometric change) is set to 10, the third weight coefficient (functional reconfiguration) is set to 100, and the penalty-level weight coefficient is set to 1000. This gives the system a strong "low-intervention tendency." (The formula is used to...) By accumulating these parameters, the system automatically excludes high-weight actions that involve restructuring the main structure or functions, prioritizing low-cost parameter fine-tuning solutions. This not only ensures the economic efficiency of the renovation costs but also preserves the original design intent of the existing health and wellness buildings to the greatest extent possible, aligning with the technical approach of "micro-renewal."

[0123] In this embodiment, the preset standard thresholds are typically set strictly in accordance with the mandatory clauses in the "Unified Standard for Civil Building Design" or the "Standard for Architectural Design of Elderly Care Facilities." For example, the threshold for the duration of sunlight in the ground floor elderly bedroom on the winter solstice is usually set to no less than 2 hours, establishing a legal and compliance red line. If a counterfactual optimization action (such as adding a sunshade) improves the thermal environment but causes the duration of sunlight to fall below this threshold, the system will immediately apply a penalty-level weighting coefficient, rapidly increasing the action distance, thereby automatically filtering out these non-compliant solutions during the path optimization stage.

[0124] Compared to traditional optimization methods that are solely performance-oriented, this application establishes a cost metric model with a penalty term. This ensures that the generated decision path not only addresses environmental failure issues but also represents the lowest-cost and most compliant solution under constrained conditions. This design effectively resolves the conflict between performance optimization and regulations, as well as the contradiction between optimization results and renovation costs, providing practical decision support for the intelligent operation and maintenance and low-energy consumption renovation of senior living buildings.

[0125] In practice, the processing logic of the path optimization unit is as follows:

[0126] Within the manifold space defined by the discretized action set, a design space resistance diagram is constructed, and resistance cut-off walls are set in the design space resistance diagram according to the cut-off criterion.

[0127] The predicted gain output by the performance proxy model is used as a heuristic benefit guidance function to calculate the total evaluation cost of the evolution branch node. The calculation formula is as follows:

[0128] ;

[0129] in, For the total assessment cost, The accumulated action distance, For adjustment coefficients, The residual performance deviation value is calculated using the predicted gain;

[0130] The calculation logic for the residual performance deviation value obtained through the predicted gain is as follows:

[0131] ;

[0132] in, To preset the age-appropriate baseline constraint value, This is the predicted value of the performance proxy model for the current node;

[0133] When the performance proxy model predicts values ​​that meet the preset age-appropriate baseline constraint... When the value is 0, the system marks the current node as a performance-compliant node and terminates the expansion processing of the current evolution branch.

[0134] The path with the minimum total evaluation cost is identified through iterative search.

[0135] Specifically, within the manifold space defined by the discretized action set, the change dimension of each design variable is regarded as an axis of the space, and the combination of all actions constitutes a high-dimensional, nonlinear topological structure. In this space, the transformation process of the building from a "failure state" to a "compliance state" is mapped as the movement between two points in the space.

[0136] The processing logic of constructing a design space resistance map and setting resistance cut-off walls is to assign "physical resistance" to this abstract space. The system utilizes the action cost weight matrix in claim 6 to render high-cost action areas (such as areas involving the reconstruction of the main structure) as high resistance values; simultaneously, based on cut-off criteria (i.e., those red lines that are absolutely inviolable in terms of laws, regulations, cost budgets, or construction conditions), it generates resistance cut-off walls with infinite resistance in the manifold space. This is equivalent to pre-setting "no-go zones" within the design search space, ensuring that the algorithm automatically avoids design combinations that are too costly or non-compliant during the search.

[0137] The residual performance deviation value calculated through the predicted gain quantifies the "distance from the endpoint of the current state". At each search node (i.e., each intermediate design state), the system calls the performance proxy model in claim 2 to calculate the predicted gain (i.e., performance improvement) of the current scheme. The system calculates the residual performance deviation value according to the optimization direction of different performance objectives. When the performance objective is a gain-type indicator such as sunshine duration or age-friendly thermal comfort, where higher values ​​are preferred, the system uses the difference between the preset age-friendly baseline constraint value and the predicted mean of the current state to characterize the degree of deviation of the current scheme from the performance target state. When the performance objective is a cost-type indicator such as environmental load or building energy consumption, where lower values ​​are preferred, the system uses the difference between the predicted mean of the current state and the preset age-friendly baseline constraint value to characterize the degree of exceeding the limit of the current scheme.

[0138] When the remaining performance deviation is large, it indicates that the current solution is still far from meeting the performance standards. When the remaining performance deviation approaches zero or is less than zero, it indicates that the corresponding performance target has met the preset age-friendly baseline constraint requirements. When all performance targets meet the corresponding constraints, the system marks the current node as a performance-compliant node and terminates the expansion processing of the current evolution branch. This indicator, as the "benefit guide" of the heuristic function, forces the search algorithm to move in the direction that can eliminate performance defects the fastest. The preset age-friendly baseline constraint value is used to define the minimum age-friendly safety performance boundary that the health and wellness building must meet under the target climate conditions. Its specific value is set according to the climate zone where the building is located, the comfort and safety standards for the elderly, and the building code requirements. In this embodiment, the preset age-friendly baseline constraint value may include the indoor operating temperature range of the elderly activity area, the predicted average thermal perception index (PMV) range, the energy consumption threshold per unit area, and the winter solstice sunshine duration threshold. For example, the indoor operating temperature in the activity area for the elderly can be set to 20℃~26℃, the predicted average thermal perception index (PMV) can be limited to the range of [-0.5, +0.5], and the continuous sunshine duration of the main activity room on the winter solstice can be limited to no less than 2 hours. The system compares the prediction results of the performance surrogate model with the preset age-friendly baseline constraint value, calculates the remaining performance deviation value under the current design state, and determines whether the current scheme meets the age-friendly safety requirements. By introducing the preset age-friendly baseline constraint value, the abstract performance goal in the traditional building optimization process can be transformed into a quantified safety boundary with engineering constraint significance, so that the path optimization process always converges and controls around the age-friendly safety goal, avoiding the problem that the algorithm only pursues local performance improvement and ignores the physiological safety needs of the elderly. At the same time, the preset age-friendly baseline constraint value can also serve as the target determination condition in the counterfactual path search process. When the prediction result of the performance surrogate model reaches the corresponding constraint requirement, the system automatically marks the current node as a performance-compliant node and terminates the expansion of the corresponding evolution branch, thereby reducing the computational overhead caused by invalid search and improving the generation efficiency of the minimum intervention action sequence. Furthermore, by establishing unified constraint boundaries for thermal safety, energy consumption, and solar radiation standards, it is possible to avoid conflicts between different performance objectives, enabling the generated intervention schemes to simultaneously meet the requirements of age-friendly comfort, building compliance, and low renovation costs, thereby improving the engineering feasibility and reliability of the climate-adaptive health and wellness building design optimization process.

[0139] The system employs a heuristic cost-balancing mechanism to iteratively search for the path with the minimum total evaluation cost. Starting from the failure point, the system continuously evaluates the total evaluation cost of each evolutionary branch node. This cost is composed of the accumulated physical action distance and the weighted residual performance deviation. This logic balances the conflict between "modification cost" and "repair efficiency" by adjusting coefficients. The system uses iterative search to continuously expand the node with the minimum total evaluation cost until it captures the target node whose performance fully meets the constraints, thereby backtracking to lock in an ordered action chain that is technically compliant and has the lightest physical intervention.

[0140] Compared to traditional optimization methods based on genetic algorithms or blind trial and error, this application significantly improves search efficiency and avoids ineffective design attempts by introducing AI-predicted gain as a guiding function. Simultaneously, through deep coupling of the resistance graph and cost function, it ensures that the final generated sequence of minimum intervention actions achieves an optimal balance between economic cost, compliance, and construction disturbance, providing a highly practical automated decision support system for the micro-renovation of climate-adaptive healthcare buildings.

[0141] In practice, the evolution path module includes a causal analysis unit and a prescription generation unit;

[0142] The causal analysis unit is used to extract the performance mapping components corresponding to each discretized action in the minimum intervention action sequence and calculate the causal contribution value of each discretized action.

[0143] The prescription generation unit is used to summarize the minimum intervention action sequence according to logical time sequence and perform semantic encapsulation processing to generate an intervention prescription for the target environmental stress scenario.

[0144] Specifically, the system aggregates the minimum intervention action sequence according to logical time sequence and performs semantic encapsulation processing. First, based on the output order of the path optimization unit, the system arranges each action in the minimum intervention action sequence in time sequence. Then, the system performs semantic encapsulation, mapping the discrete action codes to a pre-stored terminology library. For example, the "interface physical parameter adjustment action" is encapsulated as "increasing the thickness of the external insulation layer of the south-facing bedroom wall"; the "energy supply logic reconfiguration action" is encapsulated as "adjusting the air conditioning heating on threshold from 18°C ​​to 20°C". Finally, the system combines the current simulated dynamic meteorological conditions to generate a structured document containing action names, execution parameters, expected contribution, and execution order, i.e., an intervention prescription for the target environmental stress scenario.

[0145] Compared to traditional optimization systems that only output a final design value (such as "the window-to-wall ratio should be 0.2"), this application, through causal analysis units, can clearly inform architects of the specific contribution weight of each fine-tuning action to solving age-friendly thermal safety issues. This approach not only makes the optimization scheme transparent and credible but also allows designers to selectively implement high-contribution actions based on budget constraints. Simultaneously, semantic encapsulation processing transforms the complex AI algorithm output into professional expressions commonly used in the construction industry, bridging the technical gap between computer language and engineering language. This enables non-AI professionals in elderly care facility operations or junior architects to directly adjust schemes or optimize equipment based on the generated "intervention prescriptions," greatly enhancing the widespread application value of climate adaptability optimization technology in the field of elderly care buildings.

[0146] In practice, the formula for calculating the causal contribution value of each discretization action is as follows:

[0147] ;

[0148] in, For the first The causal contribution value of each discretized action. For discretized action sets, To extract from the discretized action set the action that does not contain the first action Any subset of discretized actions, For performance proxy models in action subsets The predicted output value is as follows. This represents the number of discretized actions contained in the action subset. This represents the total number of discretized actions contained in the discretized action set. The performance target value predicted by the performance proxy model;

[0149] To further explain, To execute only a subset of actions based on the initial state vector at the failure site Then, the predicted gain value output by the performance proxy model, To execute a subset of actions based on the initial state vector at the failure site With the The predicted gain value output by the performance proxy model after each discretization action;

[0150] Specifically, by introducing the Shapley value calculation formula, its core value lies in its ability to overcome the nonlinear barriers in building physics processes and achieve a "fair distribution" of the effects of design adjustments. In the optimization of senior living buildings, there is often a strong coupling effect between interface physical parameters, morphological parameters, and energy supply logic. For example, the combined effect of adding insulation and improving airtightness is not simply the sum of their effects. The formula's processing logic involves iterating through all subsets of the discretized action set, examining the marginal performance improvement of a specific action when added to different action groups, and performing a weighted average. This logic ensures that regardless of how building variables interact, the final causal contribution value accurately identifies the true contribution of each action to repairing environmental failures. It transforms the complex mathematical logic within the performance proxy model into a contribution weight perceptible to architects. Compared to traditional linear regression coefficients or partial correlation coefficients, the Shapley value formula has a rigorous axiomatic basis, effectively identifying and eliminating "free-rider" variables that are statistically correlated but have weak actual contributions, thus ensuring that the generated intervention prescription directly addresses the pain points of environmental failures. This causal decoupling capability is the technical guarantee for achieving the "minimal intervention" principle, avoiding unnecessary engineering actions in the micro-renovation of senior living buildings. Furthermore, the formula provides priority-based evaluation indicators for the prescription generation unit. Because the algorithm's output is unique and additiveable, the system can clearly tell maintenance personnel the specific percentage of each action in the total gain of addressing age-friendly thermal safety issues. When faced with limited budgets or construction timelines, managers can prioritize and execute the highest-contributing "key decisions" from the action sequence based on causal contribution values, thereby maximizing environmental remediation benefits under constrained conditions.

[0151] In practice, the evolution path module also includes a logic diagram output unit;

[0152] The logic diagram output unit is used to connect the scheme state nodes before and after the action execution, with the initial design scheme as the root node and the discrete actions in the minimum intervention action sequence as logical branches. It also marks the corresponding action distance and causal contribution value on the logical branches to generate a decision evolution path. Its processing logic is as follows:

[0153] Using the initial design scheme as the root node, and following the execution sequence of the minimum intervention action sequence, each discretized action is abstracted into a logical branch connecting the state nodes of the preceding and following schemes, thus constructing a directed acyclic graph that evolves from the initial failure state to the performance target state.

[0154] Extract the action distance and causal contribution value of the discretized action corresponding to each logical branch, and use them as weight labels to map to the corresponding logical branch;

[0155] Perform topology layout processing on the directed acyclic graph to obtain the decision evolution path graph;

[0156] The decision evolution path diagram includes the evolution path state feature sequence, the ordered intervention action set, and the cumulative disturbance component value.

[0157] Specifically, the processing logic for constructing the directed acyclic graph involves encapsulating state nodes and mapping operators: the system defines the initial design scheme as the root node containing the original performance defects, and based on the minimum intervention action sequence output by the optimization algorithm, abstracts each discretized action (such as changing the window-to-wall ratio or adjusting the fresh air logic) into a logical branch that triggers state transitions. Each time an action operator is executed, the system generates a new scheme state node, recording the physical parameters and predicted performance indicators for that stage, thus forming a topological path that evolves unidirectionally from the initial failure state to the performance-compliant state, ensuring the continuity and irreversibility of the design logic.

[0158] The processing logic of extracting attributes and mapping them to weight labels is a key step in achieving data alignment across decision dimensions. During the generation of the logic graph, the system retrieves the technical parameters corresponding to each branch from the backend in real time: on the one hand, it obtains the action distance of the action from the cost measurement unit to characterize the difficulty and cost of engineering implementation; on the other hand, it obtains the causal contribution value of the action from the causal analysis unit to quantify its contribution to performance repair. The system associates and maps these two values ​​to the corresponding logic branches in the form of weight labels, so that each design decision in the graph not only has a sequential order but also a clear "cost-benefit" attribute.

[0159] Compared to traditional "black box" models that only provide the final optimization result, this application offers architects a rigorous, expert-level chain of evidence by outputting a decision evolution path diagram that includes a sequence of state features and cumulative disturbance component values. This visual representation not only allows designers to clearly grasp the logical basis for each step in achieving performance targets but also enables them to quickly find suboptimal alternative paths when actual construction is constrained, based on the action distance and contribution value on the branches. This approach greatly improves the feasibility of optimization solutions, eliminates the communication gap between computer language and engineering practice, and provides a practically valuable intelligent navigation for the climate-adaptive transformation of health and wellness buildings.

[0160] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. An AI-based climate-adaptive elderly care and wellness building design optimization system, characterized in that, It includes a status assessment module, a counterfactual diagnosis module, and an evolution path module; The state assessment module is used to obtain the initial design scheme, basic operational characteristics of the health and wellness building and dynamic meteorological parameters. It performs sample augmentation, nonlinear mapping training and contribution analysis on the basic operational characteristics of the health and wellness building and dynamic meteorological parameters to obtain the performance proxy model and variable guidance matrix. The counterfactual diagnosis module is used to perform action space abstraction and cost weight allocation on the set of design variables for health and wellness buildings, to obtain a discretized action set and action cost weight matrix, and input the initial design scheme into the performance proxy model for failure site location and counterfactual path optimization to obtain the minimum intervention action sequence. The evolution path module is used to measure the causal contribution of the minimum intervention action sequence, extract logical association records, and encapsulate temporal semantics to obtain a decision evolution path diagram.

2. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 1, characterized in that, The state assessment module includes a sample enhancement unit, a nonlinear mapping unit, and a contribution analysis unit. The sample augmentation unit is used to obtain the initial design scheme, basic operational characteristics of health and wellness buildings, and dynamic meteorological parameters. Through generative adversarial networks, the basic operational characteristics of health and wellness buildings and dynamic meteorological parameters are simulated for feature distribution and expanded for sampling to obtain a hybrid dataset. The generative adversarial network includes a generation module and a discrimination module; The nonlinear mapping unit is used to establish a weight estimation model between the set of design variables and the set of performance targets of health and wellness buildings based on a hybrid dataset. A Gaussian prior distribution is introduced to perform posterior probability estimation and hyperparameter update processing on the weights in the weight estimation model to obtain a performance proxy model. The set of design variables for health and wellness buildings includes interface physical parameters, boundary morphology parameters, and energy supply logic parameters. The performance target set includes environmental load indicators, age-friendly thermal safety evaluation indicators, and solar radiation compliance indicators; The performance proxy model is used to output the distribution of the performance target set after the design variable changes, and the improvement of the distribution relative to the preset bottom line constraint is defined as the prediction gain. The contribution analysis unit is used to perform global sensitivity analysis on the performance proxy model, calculate the contribution rate score of each health and wellness building design variable to the performance target set, sort the health and wellness building design variables in descending order of contribution rate score, and obtain the variable guidance matrix.

3. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 2, characterized in that, The processing logic for feature distribution simulation and augmented sampling includes: The basic operational characteristics and dynamic meteorological parameters of health and wellness buildings are extracted and mapped to a unified high-dimensional feature space in combination with the design variable set of health and wellness buildings in the initial design scheme, which serves as training samples for generative adversarial networks. The joint probability distribution between design variables and dynamic meteorological parameters in the training samples is learned by the generation module, and a virtual coupled feature distribution is generated. The discriminant module is used to verify the probability distribution of the virtual coupling feature distribution against the training samples to obtain the discriminant loss value. The distribution parameters of the generator module are then iteratively updated based on the discriminant loss value to obtain the trained generator module. The trained generation module is augmented with sampling to obtain virtual coupled feature samples, which are then merged with the training samples to obtain a hybrid dataset.

4. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 3, characterized in that, The processing logic of the nonlinear mapping unit is as follows: The design variables of health and wellness buildings in the mixed dataset are mapped to the input vector of independent variables, and the set of performance targets is defined as the output target of the dependent variable. A Gaussian prior distribution is introduced to set the initial uncertainty range for the weights in the weight estimation model, and a likelihood function is constructed using the input vector of independent variables and the output target of dependent variables. The influence of observed data on the weight distribution is calculated by the likelihood function, and the coupling weights between variables are updated by the posterior probability estimation method to obtain the updated coupling weights. The mean of the predicted distribution of the input vector of independent variables under the corresponding climate parameters is calculated based on the updated coupling weights, and the performance surrogate model is output.

5. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 4, characterized in that, The counterfactual diagnosis module includes an action abstraction unit, a cost measurement unit, a failure point location unit, and a path optimization unit. The action abstraction unit is used to transform the set of design variables for health and wellness buildings into a set of discretized actions based on the variable guidance matrix. The set of discretized actions includes actions for adjusting interface physical parameters, actions for changing boundary morphology, and actions for reconstructing functional logic. The cost measurement unit is used to establish a truncation criterion for each discretized action in the discretized action set, and to assign a corresponding preset cost weight value based on the truncation criterion to obtain the action cost weight matrix. The failure site location unit is used to input the initial design scheme into the performance proxy model, identify the spatiotemporal distribution characteristics of the performance target set that do not meet the preset aging-friendly bottom line constraints, extract the spatial coordinates of the design variables corresponding to the failure time characteristics, define them as failure sites, and obtain the initial state vector of the design variables corresponding to the failure sites. The path optimization unit is used to take the failure site identified by the failure site location unit as the starting node, and the discretized action set as the evolution branch. Within the limited action distance constraint domain defined by the action cost weight matrix, it calls the enhanced path-finding algorithm to extract the counterfactual candidate set of performance target set that meets the preset age-friendly bottom line constraint. The ordered action chain is locked by the cumulative evaluation cost calculation to obtain the minimum intervention action sequence.

6. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 5, characterized in that, The processing logic of the cost metric unit is as follows: Get the pending processing number Each action is discrete, and the corresponding preset cost weight value is retrieved from the action cost weight matrix; When the discretization action is a parameter adjustment action that does not change the main structure, the first weight coefficient is assigned; When the discretization action involves a change in the geometry of a component, a second weighting coefficient is assigned; When the discretization action is a reconstruction action involving spatial functional adjacency relationships, a third weight coefficient is assigned; When the discretization action causes the solar radiation compliance index to fall below the preset standard threshold, a penalty level weight coefficient is assigned; Among them, the penalty level weight coefficient is greater than the third weight coefficient, the third weight coefficient is greater than the second weight coefficient, and the second weight coefficient is greater than the first weight coefficient; Get the The execution status codes of each discretized action are used to calculate the sum of the products of the preset cost weight values ​​corresponding to each discretized action, thus obtaining the action distance.

7. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 6, characterized in that, The processing logic of the path optimization unit is as follows: Within the manifold space defined by the discretized action set, a design space resistance diagram is constructed, and resistance cut-off walls are set in the design space resistance diagram according to the cut-off criterion. The predicted gain output by the performance proxy model is used as a heuristic benefit guidance function to calculate the total evaluation cost of the evolution branch node; The path with the minimum total evaluation cost is identified through iterative search.

8. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 7, characterized in that, The evolution path module includes a causal analysis unit and a prescription generation unit; The causal analysis unit is used to extract the performance mapping components corresponding to each discretized action in the minimum intervention action sequence and calculate the causal contribution value of each discretized action. The prescription generation unit is used to summarize the minimum intervention action sequence according to logical time sequence and perform semantic encapsulation processing to generate an intervention prescription for the target environmental stress scenario.

9. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 8, characterized in that, The formula for calculating the causal contribution value of each discretization action is as follows: ； in, For the first The causal contribution value of each discretized action. For discretized action sets, To extract from the discretized action set the action that does not contain the first action Any subset of discretized actions, For performance proxy models in action subsets The predicted output value is as follows. This represents the number of discretized actions contained in the action subset. This represents the total number of discretized actions contained in the discretized action set. The performance target value is the value predicted by the performance proxy model.

10. The AI-based climate-adaptive elderly care and health building design optimization system as described in claim 9, characterized in that, The evolution path module also includes a logic diagram output unit; The logic diagram output unit is used to connect the scheme state nodes before and after the action execution, with the initial design scheme as the root node and the discrete actions in the minimum intervention action sequence as logical branches. It also marks the corresponding action distance and causal contribution value on the logical branches to generate a decision evolution path. Its processing logic is as follows: Using the initial design scheme as the root node, and following the execution sequence of the minimum intervention action sequence, each discretized action is abstracted into a logical branch connecting the state nodes of the preceding and following schemes, thus constructing a directed acyclic graph that evolves from the initial failure state to the performance target state. Extract the action distance and causal contribution value of the discretized action corresponding to each logical branch, and use them as weight labels to map to the corresponding logical branch; Perform topology layout processing on the directed acyclic graph to obtain the decision evolution path graph; The decision evolution path diagram includes an evolution path state feature sequence, an ordered set of intervention actions, and cumulative disturbance component values.

Images