Indoor environment quality intelligent evaluation method based on multi-dimensional data fusion

Abstract

The invention relates to the technical field of environmental parameter measurement, in particular to an indoor environment quality intelligent evaluation method based on multi-dimensional data fusion, which comprises the following steps: step 1, performing time-space synchronous acquisition on multi-source data; 2, a dynamic correlation model is constructed, a parameter coupling matrix is generated based on inter-parameter time-delay cross-correlation analysis, the time-delay cross-correlation analysis is achieved through sequence translation in a sliding time window and correlation coefficient calculation, an environment state class is constructed according to a parameter mean vector, fluctuation intensity and abnormal event marks, and a dynamic correlation model is constructed; each environment state class is bound with an independent parameter contribution degree weight set; step 3, performing double-stage environment quality evaluation; and step 4, adaptive increment optimization: when the deviation between the comprehensive evaluation index and the subjective evaluation exceeds a threshold value, adjusting a parameter contribution degree weight set of an environment state class, and periodically reconstructing a parameter coupling matrix. The prediction capability is improved by considering the time-delay coupling between the parameters; the system has self-learning and optimization capabilities, and the stability of long-term operation of the system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental parameter measurement, and in particular to a multi-dimensional data fusion intelligent evaluation method for indoor environmental quality. Background Art

[0002] Indoor Environmental Quality (IEQ) assessment is a core technology in the fields of smart buildings and healthy living. Existing assessment methods have the following limitations: Current mainstream technologies rely on independent threshold judgments for parameters such as temperature, humidity, and PM2.5. However: The dynamic coupling effect between parameters is ignored: for example, temperature and humidity synergistically affect the microbial growth rate (when the relative humidity is >60%, the bacterial growth rate increases by 30% for every 5°C increase in temperature), but existing methods do not establish such cross-parameter correlation models; Consequences: In complex scenarios such as hospital operating rooms, controlling only a single parameter to meet the standard may still lead to the risk of microbial exceeding the standard, and the assessment results may seriously deviate from the actual environmental conditions. Reasonableness of the question: The indoor environment is a complex system with multiple parameters interacting. Independent evaluation violates the laws of environmental physics. This defect is particularly prominent in health-sensitive places. In addition, the existing technology also adopts a fixed weight fusion solution, which still has inherent defects: Unable to respond to dynamic events: When a sudden gathering of people causes a sudden increase in CO2 concentration, the CO2 weight does not increase dynamically, causing the assessment value to lag behind the actual pollution level; For example, when the CO2 concentration rises from 600ppm to 1500ppm during the use of a conference room, the fixed-weight model assessment value changes by only 15%, while the actual subjective dissatisfaction increases by 70%. However, given the frequent occurrence of emergencies in building environments, static models are unlikely to meet real-time health protection needs. Therefore, there is an urgent need for an intelligent indoor environmental quality assessment method that integrates multi-dimensional data to solve the above problems. Summary of the Invention

[0003] Based on the above objectives, the present invention provides a multi-dimensional data fusion intelligent indoor environmental quality assessment method, comprising the following steps: Step 1: Multi-source data is collected synchronously in time and space. Temperature, humidity, PM2.5 concentration, CO2 concentration, TVOC concentration, and light intensity are collected synchronously through a distributed sensor network, and spatial coordinates and synchronization timestamps are added. Step 2: Dynamic correlation model construction, generating a parameter coupling matrix based on time-lagged cross-correlation analysis between parameters. The time-lagged cross-correlation analysis is achieved by sequence translation and correlation coefficient calculation within a sliding time window. Environmental state classes are constructed based on parameter mean vectors, fluctuation intensity, and abnormal event markers. Each environmental state class is bound to an independent parameter contribution weight set. Step 3: Two-stage environmental quality assessment: Real-time status assessment: Match current data to environmental status classes and calculate the real-time quality index based on the parameter contribution weight set. Trend evolution assessment: Use the parameter coupling matrix to predict future change trends and calculate the trend risk index based on the parameter sensitivity coefficient. The real-time quality index and trend risk index are integrated to generate a comprehensive assessment index. The integration coefficient is dynamically adjusted as the environmental status changes. Step 4: Adaptive incremental optimization. When the deviation between the comprehensive evaluation index and the subjective evaluation exceeds the threshold, the parameter contribution weight set of the environmental state class is adjusted and the parameter coupling matrix is ​​periodically reconstructed.

[0004] Beneficial effects of the present invention: 1. This method introduces time-lag cross-correlation analysis based on a sliding time window to establish a coupling relationship matrix between multiple parameters, effectively revealing the synergistic or inverse effects between parameters such as temperature and humidity, CO2 and TVOC.

[0005] 2. By constructing environmental state categories, combined with historical fluctuation characteristics and abnormal event markers, we achieve a refined classification of typical environmental states and assign independent parameter contribution weights to different states. This avoids a one-size-fits-all assessment model and enables targeted evaluation. This mechanism of binding environmental states to assessment weights significantly improves judgment accuracy in complex scenarios, making assessment values ​​more closely aligned with actual human perception and health risk levels.

[0006] 3. This invention uses a dynamic weighting mechanism to adaptively select parameter weights based on the currently matched environmental state. When an emergency event, such as a sudden CO2 spike, occurs, the system immediately matches the high CO2 sensitivity state category and automatically increases the assessment weight of that parameter. This allows the comprehensive assessment value to rapidly respond to pollution changes, overcoming the lag in response associated with traditional fixed-weight models.

[0007] 4. By introducing an adaptive incremental optimization mechanism, when there is a discrepancy between a user's subjective perception and the overall evaluation value, the parameter weight set is automatically corrected and the parameter coupling structure is periodically reconstructed, allowing the model to continuously converge to the true perception of individuals or groups. This optimization strategy not only enhances the personalized adaptability of the evaluation but also ensures long-term stability and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0009] Figure 1 is a flow chart of the steps of the method of the present invention; Figure 2 A flow chart showing the steps of calculating the spatial influence weight for each sensor in the method of the present invention; Figure 3 The present invention provides a flowchart of the steps for calculating the trend risk index in step 3 of the method, including calibration of the parameter sensitivity coefficient. DETAILED DESCRIPTION

[0010] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0011] See Figure 1-Figure 3 This embodiment of the present invention provides an intelligent indoor environmental quality assessment method based on multi-dimensional data fusion. First, during the data collection phase, multiple sensors deployed at various locations indoors synchronously collect environmental parameters such as temperature, humidity, PM2.5, CO2, TVOC, and light intensity in real time. Each piece of data is annotated with the specific acquisition time and spatial coordinates to achieve unified temporal and spatial identification. By establishing a unified synchronization cycle and a network clock calibration mechanism, the consistency of all data sources on the timeline is ensured, effectively supporting subsequent correlation analysis.

[0012] Subsequently, during the dynamic correlation modeling phase, a sliding time window technique is used to perform sequence analysis based on the temporal dependencies between multiple parameters. For any two parameters, the time series of one parameter is slid, and the correlation coefficient with the other parameter is calculated at each time shift position. The maximum correlation coefficient and the corresponding time offset are extracted as the coupling strength and dominant lag value for that parameter pair. The analysis results of all parameter pairs are summarized into a parameter coupling matrix. Environmental state categories are constructed by combining historical parameter means, fluctuation strengths, and anomaly markers. Each state category is associated with an independent set of parameter contribution weights to reflect the relative importance of different parameters to environmental quality under that environmental state.

[0013] A two-stage evaluation mechanism was designed for quality assessment. First, currently collected data is matched in real time to the most similar environmental state category. The weighted calculation for each parameter is performed using the weight set within that category, outputting a real-time environmental quality index. Subsequently, the parameter coupling matrix is ​​used to predict the future evolution of each parameter. Combined with parameter sensitivity indicators, a trend risk index that could potentially lead to environmental quality degradation is calculated. Finally, the real-time index and risk index are weighted and fused using a state-adaptive fusion coefficient to form a comprehensive assessment. This fusion coefficient is dynamically adjusted based on environmental stability, prioritizing real-time indicators during stable conditions and focusing on trend risks during fluctuating conditions.

[0014] Finally, if there's a significant discrepancy between the comprehensive evaluation and the user's subjective experience (e.g., if the deviation exceeds a set threshold), a parameter weight adjustment mechanism is triggered to further refine the fit to the subjective experience. Simultaneously, changes in environmental state categories and the stability of the parameter coupling structure are periodically checked, and the coupling matrix is ​​rebuilt when necessary to improve overall evaluation accuracy and adaptability.

[0015] The present invention realizes dynamic perception and intelligent evaluation of complex indoor environmental conditions, can accurately reflect the current status and warn of potential risks; by considering the time-lag coupling between parameters, it improves the prediction ability; at the same time, it has self-learning and optimization capabilities, which improves the long-term stability of the system and user satisfaction.

[0016] In one possible implementation, the spherical neighborhood radius of each sensor is first calculated based on the indoor space's volume and the sensor density. Larger volumes and sparser sensor distribution correspond to larger neighborhood radiuses; conversely, higher sensor density results in smaller neighborhood radiuses. This calculation process ensures that the influence of each sensor is appropriately considered across different spatial layouts, avoiding inaccurate assessments caused by overly concentrated or overly dispersed sensor configurations.

[0017] Next, based on the Euclidean distance principle, the influence weights of other sensors in the neighborhood on the target point are calculated. The weights decay exponentially with increasing distance, meaning that sensors farther from the target point have less influence on the data, and vice versa. This decay pattern aligns with the signal transmission characteristics of real physical environments and accurately reflects the sensor's range of influence in space.

[0018] Because sensors may experience data anomalies during operation due to failures, environmental factors, or other reasons, a signal quality factor is introduced to correct the initial weight components. This factor is determined by calculating the anomaly rate of the sensor's historical data. A higher anomaly rate indicates a smaller correction factor, indicating lower reliability of the sensor's data and a corresponding decrease in its influence on the target point. This correction mechanism ensures that the model more accurately reflects the actual environmental conditions in the presence of unreliable sensors, avoiding misleading evaluation results caused by unreliable data.

[0019] Ultimately, the calculated spatial influence weights are used in subsequent data restoration and spatial interpolation. During the restoration process, the sensor spatial influence weights help select appropriate sensor data for fusion and correction, ensuring data accuracy and consistency. Furthermore, during spatial interpolation, the weightings help accurately estimate environmental data at locations not directly measured, enhancing the comprehensiveness and accuracy of the assessment method.

[0020] The implementation of this technical feature effectively solves the evaluation errors caused by uneven sensor distribution or differences in signal quality. By introducing spatial influence weights, not only the data repair accuracy is improved, but also the adaptability and reliability of the evaluation method in complex spatial environments are enhanced. For dynamically changing environments, this weight calculation method based on physical distance and sensor quality significantly improves the authenticity and accuracy of the evaluation, avoids the problem of "data distortion far away from the sensor location" that may exist in traditional evaluation methods, and effectively addresses the uncertainty caused by sensor failure or poor signal, thereby improving the stability and accuracy of the entire evaluation process.

[0021] In one possible implementation, two sets of parameter sequences, A and B, are first processed. During this process, sequence A is shifted within a sliding time window at a fixed step size. The length of the sliding time window and the step size are crucial, directly impacting the accuracy of the analysis results and computational efficiency. By shifting sequence A, the correlation between sequences A and B can be gradually assessed at different time points, ensuring a comprehensive analysis of potential time lags between the two.

[0022] After each translation, the correlation coefficient between the translated sequences A and B is calculated. The correlation coefficient is a measure of the linear relationship between two sequences, ranging from -1 to 1, with values ​​closer to 1 indicating a closer relationship. By calculating the correlation coefficient for all translation steps, the maximum value is selected as the coupling strength value for that parameter pair, reflecting the strongest coupling relationship between sequences A and B at that moment. This maximum value indicates the strongest coupling strength at that moment, which helps further identify interaction patterns between sequences.

[0023] After finding the maximum correlation coefficient, record the number of translation steps at that point and convert it to time units. This time value is the dominant lag. The dominant lag represents the time difference between the two series when the maximum correlation occurs, reflecting the time delay in the impact of changes in one parameter on the other. This lag is an important parameter for describing the temporal relationship between the two series and provides a deeper understanding of their temporal coupling characteristics.

[0024] Finally, the trigger conditions for matrix updates are defined. Specifically, if the rate of change in the number of environmental state classes exceeds a set proportional threshold, or if the fluctuation amplitude of the dominant lag value within the same class exceeds a stability threshold, the analysis results are reconstructed. This triggering mechanism ensures that the analysis model can be adjusted promptly when the environmental state changes significantly, thereby improving the model's adaptability and accuracy.

[0025] By using a sliding time window and a fixed-step shift, the temporal correlations between sequences can be captured at multiple time points, making the analysis of time lags more detailed and comprehensive. Secondly, the selection of the maximum correlation coefficient effectively reflects the strongest coupling relationship between parameters, ensuring the accuracy of the time lag analysis. By recording the dominant time lag value, the temporal relationship between parameters can be effectively revealed, providing a reliable basis for subsequent dynamic adjustment and prediction. In addition, the triggering conditions for matrix updates ensure the flexibility and dynamic responsiveness of the analysis in the face of environmental changes, avoiding data distortion caused by environmental fluctuations and making the overall analysis results more stable and reliable. These steps work together to greatly enhance the practicality of the time lag analysis method and can effectively improve the modeling and prediction capabilities of complex environmental systems in practical applications.

[0026] In one possible implementation, the fluctuation intensity of each parameter in the historical data is first calculated. Fluctuation intensity generally refers to the degree of change in a parameter over time. After calculating the fluctuation intensity of each parameter in the historical data, quantile analysis is then performed. This involves sorting these fluctuation intensities by size and dividing them into several intervals. This quantile analysis can clarify the distribution of parameter fluctuation intensities, providing a data basis for the subsequent determination of cluster radius.

[0027] In the density clustering algorithm, the cluster radius is an important parameter that determines whether data points belong to the same class. In order to determine the cluster radius, this method uses the quantile method to select the initial radius. The specific steps are to select the fluctuation intensity value of the Nth quantile and multiply it by a spatial topological complexity coefficient as the initial radius of the cluster. The Nth quantile value here can be adjusted according to different needs to adapt to different data distribution characteristics. The spatial topological complexity coefficient compensates for the complexity of the data space structure, which can help adjust the adaptability of the initial radius so that the density clustering can better reflect the true structure of the data.

[0028] During density clustering, it's necessary to dynamically evaluate whether a new environmental state class should be added. This decision is based on whether the weighted Euclidean distance between the current data vector and all existing class centers is greater than a specified multiple of the maximum intra-class distance. Weighted Euclidean distance measures the distance between data points, where parameter weights are weighted according to the inverse of their historical fluctuations. This means that parameters with larger historical fluctuations are assigned smaller weights when calculating distances. This effectively minimizes their impact on the clustering results, ensuring more stable and accurate clustering.

[0029] Through the above steps, the environmental state classes can be dynamically updated each time new data arrives. A large weighted Euclidean distance between the current data vector and the existing class center indicates that the data differs significantly from the existing environmental state classes, and therefore can be considered for classification as a new environmental state class. This ensures that the clustering results adapt to data changes, avoids excessive convergence between classes, and ensures that each environmental state class accurately reflects specific environmental characteristics.

[0030] By introducing a density clustering algorithm and determining the cluster radius based on the fluctuation intensity quantile, the reliance on parameter selection and radius setting in traditional clustering methods is avoided. The combination of the fluctuation intensity value calculated by the quantile and the topological complexity coefficient makes the cluster radius more adaptable and can automatically adjust according to the actual distribution of the data, thereby effectively improving the clustering accuracy and stability. At the same time, the weighted Euclidean distance strategy is adopted to minimize the impact of parameters with large historical fluctuations on the clustering results, ensuring that the clustering process is not interfered with by individual abnormal data, and the clustering effect is more in line with actual needs. In addition, the mechanism of dynamically updating the environmental state class ensures that the method can adjust the clustering results in a timely manner when facing changing data, maintaining the accuracy and timeliness of data classification.

[0031] In one possible implementation, a dataset of extreme events is first extracted from historical data. Extreme events are defined as events with the largest fluctuations or unusual changes in past historical data. These events typically cause dramatic changes in the system's state, so a detailed analysis can help identify which parameters play a key role in the system. In this step, the maximum fluctuation of each parameter is recorded, along with the corresponding subjective evaluation drop for each event. This subjective evaluation drop is typically provided by experts or experienced evaluators to describe the overall impact of the event on the system.

[0032] After obtaining a dataset of historical extreme events, a machine learning algorithm is used to fit the mapping between parameter fluctuations and subjectively assessed declines. This step aims to uncover patterns between the intensity of each parameter's fluctuations and its actual impact on the system (i.e., decline). Typically, regression analysis or other supervised learning algorithms are used to model the relationship between fluctuations and declines, generating a baseline sensitivity value for each parameter. This baseline value reflects the sensitivity of each parameter to changes in system state during fluctuations and serves as the basis for calculating the trend risk index.

[0033] After generating the baseline sensitivity value, a weighted correction is performed based on the coupling relationships between the various parameters. The coupling matrix reflects the strength and direction of the mutual influence between different parameters. This process first calculates the average correlation strength of each parameter in the coupling matrix—that is, the degree of correlation between a parameter and other parameters. Then, a weighted correction is applied to the baseline sensitivity value and the average correlation strength for that parameter to obtain a more accurate sensitivity value. This weighted correction takes into account the interactions between different parameters and further improves prediction accuracy.

[0034] Finally, the rate of change of trend risk is predicted using a chain propagation model. In this process, the dominant lag value is used to capture the time-delayed effects of parameter changes in the system. The chain propagation model simulates the impact of changes in various system parameters on the final trend risk index. By accounting for the influence, sensitivity, and time lag effects between these parameters, it can predict the system's future trend risk. This approach better captures the complex time dependencies and interactions within the system, providing more accurate risk predictions.

[0035] This method accurately calibrates the sensitivity of each parameter, making trend risk predictions more accurate. By extracting historical extreme events and leveraging machine learning techniques, it can reveal the deep connection between parameter fluctuations and the actual impact on the system, thereby optimizing the sensitivity calculation and correction process. The weighted correction of the sensitivity base value takes into account the interaction between parameters, making the risk prediction more consistent with the actual operation of the system. The use of chain propagation models and dominant lag values ​​further optimizes the time accuracy of risk prediction and the speed of response to changing trends. In summary, this method can provide an efficient, dynamic, and accurate solution for trend risk prediction, which is of great significance for identifying potential risks in advance and adjusting decisions in a timely manner.

[0036] In one possible implementation, one of the conditions for increasing the real-time evaluation item coefficient is that the duration of the environmental state class exceeds a stability threshold determined by the historical average duration of that class. An environmental state class refers to the state characteristics exhibited by an environment under specific conditions, while duration refers to the length of time the environment maintains that state. To determine whether an environmental state has stabilized, its historical average duration can be calculated and a stability threshold set. If the current environmental state duration exceeds this threshold, the current environmental state is considered stable, and the coefficient can be further adjusted to make the evaluation item more consistent with the changing trends of the current environment.

[0037] Secondly, the growth rate of the trend risk index must be below the sensitivity threshold set by the environmental state feature. This condition determines the speed of change in trend risk. If the growth rate of the trend risk index is low, it indicates that the system's risk is changing slowly and no longer experiencing drastic fluctuations. In this case, the real-time assessment coefficient can be appropriately increased to strengthen the focus on the current state and thus more accurately predict future risk changes.

[0038] In practice, the increase conditions for the trend assessment coefficient are opposite to those for the real-time assessment coefficient. That is, when the real-time assessment coefficient increases, the trend assessment coefficient decreases accordingly, and vice versa. Furthermore, the sum of these two factors always remains equal to 1. This means that during the adjustment process, the weights of the assessment items are always balanced, preventing any one coefficient from being overly weighted. This ensures that the weights of the assessment items are dynamically adjusted under varying environmental conditions to meet diverse risk assessment needs.

[0039] By setting the conditions for increasing the real-time assessment coefficient, this method can dynamically adjust the assessment coefficient to more accurately reflect changes in environmental conditions and the evolution of trend risks. Specifically, when environmental conditions are relatively stable and risks change slowly, increasing the real-time assessment coefficient helps maintain high sensitivity to existing trends, thereby improving the accuracy of forecasts of future changes. In the case of rapidly changing trend risks, increasing the trend assessment coefficient helps quickly capture environmental changes and effectively respond to sudden risks. Because the sum of the real-time and trend assessment coefficients is always 1, the problem of unbalanced assessment item weights is avoided, making the entire assessment process more scientific and stable.

[0040] In one possible implementation, first, within the same environmental state class, the data is divided into multiple subintervals based on chronological order. These subintervals are typically divided based on the continuity of time and its impact on changes in the dominant time lag value. These subintervals represent different sections of time, and the time lag values ​​within each subinterval have a certain degree of stability. Within each subinterval, the environmental state remains essentially unchanged, and the changes in various indicators are relatively regular. Therefore, dividing time into multiple subintervals enables independent evaluation of the fluctuation amplitude of the dominant time lag value in different time periods, thereby preventing global data fluctuations from interfering with local evaluations.

[0041] Next, calculate the standard deviation of the dominant lag values ​​within each subinterval. Standard deviation is a statistic that measures the degree of data fluctuation or dispersion. In this step, the standard deviation quantifies the magnitude of fluctuation in the dominant lag values ​​within each subinterval. A large standard deviation of the dominant lag values ​​within a subinterval indicates significant fluctuations in the dominant lag values ​​within that interval, possibly reflecting some instability. Conversely, a small standard deviation indicates minimal fluctuation in the lag values ​​within that interval, indicating a relatively stable system.

[0042] Finally, when the standard deviation of the dominant lag value within multiple consecutive subintervals exceeds a preset threshold, the fluctuation amplitude is considered abnormal. This criterion helps promptly identify abnormal fluctuations in the dominant lag value. Abnormal fluctuation amplitudes may indicate unusual changes in environmental conditions or abnormal system responses. If the fluctuation amplitudes in certain subintervals remain abnormal, it indicates that factors in the environment or system are affecting the changes in the dominant lag value, which requires sufficient attention.

[0043] This fluctuation amplitude assessment method enables precise monitoring of changes in the dominant time lag value, promptly identifying fluctuations that could impact system stability. In practical applications, the dominant time lag value often plays a crucial role in risk assessment and environmental forecasting. By quantifying the fluctuation amplitude and setting thresholds to identify anomalies, the system can quickly respond to abnormal fluctuations, avoiding erroneous judgments or decisions caused by excessive fluctuations. Therefore, this technical feature can effectively improve the accuracy of environmental status assessments and risk forecasts, enhance the ability to respond to emergencies, reduce risks in system decision-making, and ensure stable system operation in complex environments.

[0044] In one possible implementation, to address the problem of missing parameters at a point P in space due to a lack of sensors, this method proposes the following multi-level estimation process: First, a spherical neighborhood with an adaptively adjustable radius is dynamically constructed, centered around the point to be estimated, P. The radius is determined based on the density of sensors in the area and data availability. If sensors are sparsely distributed, the radius is appropriately increased to ensure that the neighborhood contains sufficient valid sensor data sources. This approach avoids the problem of a fixed radius failing in sparse areas and achieves flexibility in spatial scope.

[0045] After determining the neighborhood, all sensors within it are retrieved and their corresponding measurements are extracted. Next, the spatial influence weight of each sensor on point P is calculated based on factors such as the spatial distance between each sensor and point P, environmental similarity, and historical stability. This weighting can be applied using methods such as the inverse distance principle or exponential decay model. Finally, an estimated value for point P is obtained by taking the weighted sum of each sensor's measurements and their corresponding weights.

[0046] If no valid sensors exist within the designated neighborhood, and even expanding the neighborhood radius fails to generate a valid data source, a cross-region compensation mechanism is activated. This mechanism first analyzes the movement trajectory data of individuals within the area and extracts the percentage of time spent in the target area. It then extracts corresponding parameter data from other areas with high correlations with the target area's behavior. Using the dwell time in each area as a weight, these measurements are combined to infer the parameters of the target area. This compensation method not only fills the gap in spatial data but also incorporates crowd behavior as a key reference dimension, enhancing the semantic rationality of the data inference.

[0047] In one possible implementation, similarity retrieval technology is first used to analyze historical environmental data for categories similar to the current environmental state. By comparing the attributes of the current environment and historical state categories, the similarity between the two is calculated, and the state category closest to the current environment is selected. For these historical state categories, the weights of their relevant parameters are extracted and used as initial weights. This approach leverages the experience and patterns of historical data to provide reasonable weight initialization for the current environment, avoiding the instability caused by random initialization.

[0048] In some cases, no historical categories similar to the current environment state may be found. In these cases, the method instead uses a global regression model of parameters and subjective evaluations to generate initial weights. Through global analysis and regression modeling of historical data, combined with subjective evaluation feedback, initial parameter weights appropriate for the current environment are generated. This strategy ensures that even without direct reference to similar categories, appropriate weights can be obtained through the global model, avoiding initialization issues caused by a lack of historical data.

[0049] In environments with new classes, the initial sample size may be small, leading to potential inaccurate weight initialization. To address this issue, in the early stages, the weight adjustment step size is dynamically reduced based on the number of samples. Specifically, when the sample size is small, the weight adjustment is larger to quickly adapt to new data; as the sample size increases, the step size is gradually reduced to avoid instability caused by over-adjustment. This approach ensures sufficient flexibility for new classes in the early stages, and gradually stabilizes as data accumulates, ultimately achieving more precise weight distribution.

[0050] In one possible implementation, during each round of weight update, the weight change trend from the previous iteration or iterations is retained and introduced as a momentum term into the current weight calculation. The introduction of the momentum term can effectively mitigate the dramatic fluctuations caused by sample discreteness or sudden changes in data, thereby achieving "inertia control" during the update process. The initial value of the momentum coefficient is set high to fully influence the historical trend, but as the number of such samples continues to increase, the momentum coefficient decreases according to a preset strategy and is generally negatively correlated with the total number of samples. This mechanism can achieve a natural transition from initial reliance on historical trends to a later reliance on current data.

[0051] To prevent newly introduced data from shifting the internal structure of the weight vector, this method imposes an intra-class consistency constraint on the weight gradients corresponding to the new data. This constraint constructs a local consistency regularization term by analyzing the distribution of samples within the class in the feature space. This constrains the new data to maintain consistency with the existing data in terms of direction and scale when adjusting its parameter contribution. The strength of this constraint is designed to be inversely proportional to the square root of the number of samples in the class; the greater the number of samples, the weaker the constraint. This allows for greater freedom when data is abundant and enforces local consistency when data is scarce.

[0052] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0053] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A multi-dimensional data fusion intelligent indoor environment quality assessment method, characterized by: The following steps are involved: Step 1: Multi-source data is collected synchronously in time and space. Temperature, humidity, PM2.5 concentration, CO2 concentration, TVOC concentration, and light intensity are collected synchronously through a distributed sensor network, and spatial coordinates and synchronization timestamps are added. Step 2: Dynamic correlation model construction, generating a parameter coupling matrix based on time-lagged cross-correlation analysis between parameters. The time-lagged cross-correlation analysis is achieved by sequence translation and correlation coefficient calculation within a sliding time window. Environmental state classes are constructed based on parameter mean vectors, fluctuation intensity, and abnormal event markers. Each environmental state class is bound to an independent parameter contribution weight set. Step 3: Two-stage environmental quality assessment: Real-time status assessment: Match current data to environmental status classes and calculate the real-time quality index based on the parameter contribution weight set. Trend evolution assessment: Use the parameter coupling matrix to predict future change trends and calculate the trend risk index based on the parameter sensitivity coefficient. The real-time quality index and trend risk index are integrated to generate a comprehensive assessment index. The integration coefficient is dynamically adjusted as the environmental status changes. Step 4: Adaptive incremental optimization. When the deviation between the comprehensive evaluation index and the subjective evaluation exceeds the threshold, the parameter contribution weight set of the environmental state class is adjusted and the parameter coupling matrix is ​​periodically reconstructed.

2. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 1, characterized in that: Step 1 also includes calculating the spatial influence weight for each sensor. The calculation process is: a. Determine the spherical neighborhood radius based on the indoor space volume and sensor distribution density; b. Calculate the initial weight components of each sensor in the neighborhood for the target point. The initial weight components decay according to the negative exponential law as the Euclidean distance increases. c. Introduce a signal quality factor to correct the initial weight component. The signal quality factor is calculated based on the anomaly rate of the sensor's historical data. The higher the anomaly rate, the smaller the correction coefficient. The spatial influence weight is used for subsequent data repair and spatial interpolation.

3. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 1, characterized in that: The specific process of time-lag cross-correlation analysis in step 2 includes: a. For two parameter sequences A and B, shift sequence A with a fixed step size within a sliding time window; b. Calculate the correlation coefficient between sequence A and sequence B after translation, and take the maximum correlation coefficient as the coupling strength value of the parameter pair; c. Record the number of translation steps when the maximum correlation coefficient is reached and convert it into time units as the dominant lag value; Matrix update triggering condition: Reconstruction is initiated when the rate of change of the number of environmental state classes exceeds the set ratio threshold, or the fluctuation amplitude of the dominant lag value within the same class exceeds the stability threshold.

4. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 1, characterized in that: The clustering process when constructing the environmental state class in step 2 adopts the density clustering algorithm. The clustering radius in the density clustering algorithm is determined as follows: a. Calculate the quantile of the parameter fluctuation intensity in historical data; b. Take the Nth quantile value multiplied by the spatial topological complexity coefficient as the initial radius; Conditions for adding a new environmental state class: The weighted Euclidean distance between the current data vector and all existing class centers is greater than a set multiple of the maximum distance within the class, and the parameter weight in the weighted Euclidean distance is the inverse of its historical fluctuation amplitude.

5. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 1, characterized in that: The calculation of the trend risk index in step 3 includes the calibration of the parameter sensitivity coefficient. The specific process includes: a. Extract historical extreme event data sets and record the maximum fluctuation range of each parameter and the corresponding subjective evaluation drop; b. Use machine learning to fit the mapping relationship between parameter fluctuations and subjective evaluation reductions to generate the basic sensitivity values ​​of each parameter; c. The sensitivity base value is weighted and modified according to the average correlation strength of the parameter in the coupling matrix, and the chain propagation model based on the dominant time lag value is used to calculate the predicted change rate.

6. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 1, characterized in that: The dynamic adjustment of the fusion coefficient as the environmental state changes in step 3 also includes the determination of the increase conditions of the real-time evaluation item coefficient, specifically including: a. The duration of the current environmental state exceeds the stability threshold determined by the historical average duration of the state; b. The growth rate of the trend risk index is lower than the sensitivity threshold set by the environmental state characteristics; The increase condition of the trend evaluation item coefficient is opposite to that of the real-time evaluation item coefficient, and the sum of the two is always 1.

7. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 3, characterized in that: The fluctuation range of the leading lag value is evaluated as follows: In the same environmental state class, multiple sub-intervals are divided in chronological order; Calculate the standard deviation of the dominant lag value in each subinterval; When the standard deviation exceeds the set threshold number of times continuously, it is determined that the fluctuation amplitude is abnormal.

8. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 2, characterized in that: The specific process of using spatial influence weights for spatial interpolation includes: Estimation of the parameter value of the sensorless position point P: a. Retrieve all sensors within a spherical neighborhood with P as the center and an adaptively determined radius; b. Calculate the weighted sum of the spatial influence of each sensor measurement value, the weights calculated according to claim 2; c. When there are no valid sensors in the neighborhood, cross-region compensation is initiated, including: Obtain the proportion of time people's movement trajectories spend in the area; The data of related areas are fused using the residence time as the weight.

9. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 4, characterized in that: The parameter weight initialization in weighted Euclidean distance includes: The weights of historically similar environmental state classes are retrieved based on similarity as the initial values; If there is no similar class, the global regression model of parameters and subjective evaluation is used to generate initial weights; When the initial samples of the new class are insufficient, the weight adjustment step is reduced in proportion to the number of samples.

10. The method for intelligent indoor environmental quality assessment based on multi-dimensional data fusion according to claim 1, characterized in that: The parameter contribution weight set for adjusting the environmental state class in step 4 also includes a momentum constraint mechanism: The momentum term of the historical weight set is retained, and the momentum coefficient decreases as the amount of such data increases; An intra-class consistency constraint is imposed on the weight gradient of new data, and the constraint strength is inversely proportional to the square root of the number of samples in that class.

Images