Building energy-saving roof heat insulation monitoring system and heat insulation monitoring method

Abstract

The invention relates to the technical field of building energy-saving monitoring, and discloses a building energy-saving roof heat insulation monitoring system and a heat insulation monitoring method. The method comprises the following steps: acquiring real-time heat insulation monitoring data of a building roof of a plurality of sensors, comparing and analyzing the real-time heat insulation monitoring data with a pre-stored roof heat insulation historical database, and extracting a system change component and a random interference component; detecting an unexpected temperature fluctuation phenomenon in the real-time data by adopting a mode recognition technology; mining a causal relationship by using a graph model analysis method, performing mode co-occurrence association processing, and identifying a core influence variable; taking the core influence variable as input, combining reference performance information of the same heat insulation structure to establish a standard heat insulation performance reference line, comparing the offset of the standard heat insulation performance reference line with the offset of an actually measured heat insulation performance curve, and fusing a statistical inference model to evaluate the heat insulation performance and judge the risk level of a heat insulation weak point; according to the related information, a sequence prediction technology is used for estimating an area where heat insulation failure possibly occurs in a specified time interval, and a heat insulation optimization maintenance scheme document is generated.

Description

Technical Field

[0001] This invention relates to the field of building energy conservation monitoring technology, specifically to a building energy conservation roof insulation monitoring system and insulation monitoring method. Background Technology

[0002] In the field of building energy conservation, the quality of roof insulation performance directly affects the stability of the building's indoor thermal environment and energy consumption levels. With the continuous improvement of energy-saving standards in the construction industry, the need for monitoring roof insulation systems is becoming increasingly urgent. Currently, traditional roof insulation monitoring methods mostly rely on manual inspections or single-point sensor data collection, which has significant limitations. Manual inspections not only consume a lot of manpower and resources but are also limited by the inspection cycle, making it difficult to capture real-time temperature changes, especially under extreme weather conditions, often failing to detect anomalies in the insulation system in a timely manner. Data collected by single-point sensors also has limitations, failing to comprehensively reflect the insulation status of the entire roof and easily leading to missed detections or misjudgments.

[0003] While existing monitoring methods incorporate multi-point sensing technology, they suffer from shortcomings in data processing and analysis. Most methods perform only simple statistical analysis of the collected data, lacking effective separation of systematic variation components and random disturbance components, making it difficult to accurately identify unexpected temperature fluctuations. Furthermore, the analysis of the causes of temperature fluctuations often remains at the level of judging superficial factors, lacking in-depth exploration of core influencing variables, resulting in inaccurate evaluations of thermal insulation performance and failing to provide a reliable basis for maintenance plan development.

[0004] Traditional monitoring methods also have shortcomings in predicting areas of insulation failure and developing optimized maintenance plans. Due to the lack of effective data fusion and predictive models, it is impossible to accurately predict potential future problems based on historical data and real-time monitoring results. This leads to a reactive approach to maintenance, increasing costs and potentially impacting the building's energy efficiency and lifespan due to untimely handling of insulation issues. Therefore, developing a comprehensive, accurate, and efficient method for monitoring building roof insulation performance has become an urgent problem to be solved in the field of building energy conservation. Summary of the Invention

[0005] The purpose of this invention is to provide a building energy-saving roof insulation monitoring system and method to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a building energy-saving roof insulation monitoring system and a method for monitoring insulation, the method comprising:

[0007] Real-time thermal insulation monitoring data of building roofs is collected from multiple sensors. The real-time thermal insulation monitoring data is compared and analyzed with a pre-stored historical database of roof thermal insulation to extract the system variation component and random interference component in the real-time thermal insulation monitoring data.

[0008] Based on the system variation component and the random interference component, pattern recognition technology is used to detect unexpected temperature fluctuations in the real-time thermal insulation monitoring data.

[0009] Using graphical modeling analysis, causal relationship mining and pattern co-occurrence correlation processing were performed on the unexpected temperature fluctuation phenomenon to identify the core influencing variables;

[0010] Using the core influencing variables as input parameters, and combining them with reference performance information of the same thermal insulation structure of building roofs, a standard thermal insulation performance reference line is established. By comparing the deviation between the standard thermal insulation performance reference line and the measured thermal insulation performance curve, the thermal insulation performance of building roofs is evaluated by integrating a statistical inference model, and the risk level of thermal insulation weak points is determined.

[0011] Based on the unexpected temperature fluctuations, the core influencing variables, and the risk level of the weak points in insulation, sequence prediction technology is used to estimate the building roof areas that may experience insulation failure within a specified time interval, and a document on insulation optimization and maintenance plan is generated.

[0012] Preferably, the step of using the core influencing variables as input parameters, combining reference performance information of the same thermal insulation structure of building roofs to establish a standard thermal insulation performance reference line, and evaluating the thermal insulation performance of building roofs by comparing the deviation between the standard thermal insulation performance reference line and the measured thermal insulation performance curve, and determining the risk level of weak points in thermal insulation, includes:

[0013] Using the core influencing variables as input parameters, combined with reference performance information of the same thermal insulation structure of the building roof, data filtering technology is applied to remove outliers from information from various historical performance records, and the data after cleaning is output. Curve fitting method is used to model the data after cleaning to form a standard thermal insulation performance reference line. The various historical performance records include: reference performance information of the same thermal insulation structure, measured thermal insulation parameters of the target roof at multiple time points, records of past defect events, and maintenance files.

[0014] Compare the measured thermal insulation performance curve with the standard thermal insulation performance reference line, measure the deviation of the measured thermal insulation performance curve relative to the standard thermal insulation performance reference line, and use the average deviation index to measure the deviation range to obtain a deviation quantification score.

[0015] Based on the aforementioned deviation quantification score, a statistical inference model is used to evaluate the thermal insulation performance of the building roof, generating an initial evaluation report. The initial evaluation report is then optimized using search optimization techniques to obtain a health assessment of the building roof's thermal insulation status.

[0016] Using the aforementioned building roof insulation health status evaluation and combined with the trend prediction model, the evolution path of building roof insulation performance within a specified time interval is predicted to obtain a performance evolution prediction report. The performance evolution prediction report is then analyzed to determine the risk level of insulation weakness points.

[0017] Preferably, the evaluation operation on the thermal insulation performance of the building roof based on the deviation quantification score and the fusion statistical inference model generates an initial evaluation report, and the initial evaluation report is optimized using search optimization techniques to obtain a health evaluation of the building roof thermal insulation status, including:

[0018] A Bayesian inference system is used to perform statistical evaluation processing on the deviation quantification score to generate preliminary state evaluation results;

[0019] Based on the preliminary state evaluation results, the statistical inference model is adjusted in conjunction with the evolutionary algorithm and the gradient descent method to form an optimized statistical inference model.

[0020] The deviation quantification score is input into the optimized statistical inference model, inference calculation is performed, an evaluation result report is output, and the evaluation result report is optimized using swarm intelligence optimization algorithm and heuristic search technology to generate an optimized evaluation result report.

[0021] Based on the ensemble learning framework, the preliminary state evaluation results, the optimized statistical inference model, and the optimized evaluation result report are combined to generate a health assessment of the building roof insulation status.

[0022] Preferably, the step of inputting the deviation quantification score into the optimized statistical inference model, performing inference calculation processing, outputting an evaluation result report, and using swarm intelligence optimization algorithm and heuristic search technology to optimize the evaluation result report to generate an optimized evaluation result report includes:

[0023] Using the optimized statistical inference model, the deviation quantification score is used as the input dataset, and probabilistic inference processing is performed to output a preliminary assessment of the building roof insulation health status.

[0024] A swarm intelligence optimization algorithm is used to perform a global exploration operation on the preliminary building roof thermal insulation health status assessment to find the optimal assessment solution. The preliminary building roof thermal insulation health status assessment is then optimized based on the optimal assessment solution to obtain the swarm intelligence optimization result.

[0025] Based on the swarm intelligence optimization results, the preliminary building roof thermal insulation health status assessment is further optimized using heuristic search techniques to generate an optimized assessment result report.

[0026] Preferably, the graphical model analysis method is used to mine causal relationships and perform pattern co-occurrence association processing on the unexpected temperature fluctuation phenomenon to identify the core influencing variables, including:

[0027] An analysis dataset is constructed using the unexpected temperature fluctuation phenomenon. The dataset is then compressed using feature dimensionality reduction techniques to obtain a simplified dataset. A graph model analysis method is then used to scan the simplified dataset to identify high-frequency associated itemsets.

[0028] The Granger causality test is used to detect the high-frequency association itemset to obtain the causal chain of the real-time thermal insulation monitoring data, and the co-occurrence association of patterns is extracted from the high-frequency association itemset. The core variable set is obtained by combining the causal chain and the co-occurrence association of patterns.

[0029] Using mutual information metrics, similarity coefficients, and uniqueness indicators as evaluation criteria, the causal chains and pattern co-occurrence associations of the core variable set are evaluated, and a comprehensive scoring report is generated.

[0030] Based on the comprehensive scoring report, candidate core influencing variables are selected from the set of high-frequency related items. The decision tree model is then used to rank the candidate core influencing variables by importance, thereby identifying the core influencing variables.

[0031] Preferably, the evaluation criteria for assessing the causal chains and pattern co-occurrence associations of the core variable set using mutual information metrics, similarity coefficients, and uniqueness indicators, and generating a comprehensive scoring report, include:

[0032] Calculate the mutual information measure of causal chains and pattern co-occurrence associations in the core variable set to obtain an association strength score;

[0033] Similarity coefficients are used to perform similarity analysis on causal chains and pattern co-occurrence associations in the core variable set, and the similarity of causal chains and pattern co-occurrence associations in the core variable set is measured to obtain a similarity score;

[0034] The uniqueness index is used to perform a uniqueness assessment on the causal chain and the co-occurrence association of the pattern, and the uniqueness score of the causal chain and the co-occurrence association of the pattern is calculated to obtain a uniqueness score;

[0035] By integrating the association strength score, the similarity score, and the uniqueness score, a comprehensive scoring operation is performed on each causal chain and pattern co-occurrence association in the core variable set to generate a comprehensive scoring report.

[0036] Preferably, the step of detecting unexpected temperature fluctuations in the real-time thermal insulation monitoring data using pattern recognition technology based on the system variation component and the random interference component includes:

[0037] Based on the system variation components and the random interference components, pattern recognition technology is used to classify the real-time thermal insulation monitoring data, detect multiple normal temperature patterns and abnormal temperature patterns, and obtain the classification results based on the normal temperature patterns and the abnormal temperature patterns.

[0038] By comparing the classification results with known standard temperature patterns, real-time thermal insulation monitoring data that do not conform to the standard temperature patterns are detected and marked as unexpected temperature fluctuations.

[0039] Preferably, the step of extracting the system change component and random interference component from the real-time thermal insulation monitoring data includes: acquiring a multi-dimensional temperature sensing data stream of the building roof, and using time series decomposition technology to separate the temperature sensing data stream into long-term trend components, seasonal cycle components and residual components.

[0040] The long-term trend component is processed by wavelet transform to generate low-frequency characteristic coefficients, which are used as system variation components.

[0041] The seasonal cycle component and the residual component are input into the noise filtering model, and the purified high-frequency fluctuation characteristics are output as random interference components through adaptive threshold filtering.

[0042] The low-frequency feature coefficients and the purified high-frequency fluctuation features are synchronously transmitted to the pattern recognition technology processing module.

[0043] Preferably, the deviation of the measured thermal insulation performance curve relative to the standard thermal insulation performance reference line includes: setting a dynamic sampling point set on the standard thermal insulation performance reference line and calculating the rate of change of the tangent slope of the measured thermal insulation performance curve at each sampling point;

[0044] The key inflection points of the standard thermal insulation performance reference line and the measured thermal insulation performance curve are aligned using a dynamic time warping algorithm to generate a set of inflection point offset distances.

[0045] A weighted fusion operation is performed on the set of tangent slope change rate and inflection point offset distance to output a multidimensional deviation feature vector;

[0046] The multidimensional deviation feature vector is input into the average deviation index measurement module to generate a deviation quantification score.

[0047] Preferably, the present invention further includes a building energy-saving roof insulation monitoring system for implementing the above-described building energy-saving roof insulation monitoring method, the system comprising:

[0048] The data acquisition module is used to collect real-time thermal insulation monitoring data of building roofs from multiple sensors, compare and analyze the real-time thermal insulation monitoring data with a pre-stored roof thermal insulation historical database, and extract the system change component and random interference component in the real-time thermal insulation monitoring data.

[0049] Anomaly detection module is used to detect unexpected temperature fluctuations in the real-time thermal insulation monitoring data based on the system variation components and the random interference components, using pattern recognition technology.

[0050] The variable identification module is used to use graphical model analysis methods to mine causal relationships and process co-occurrence patterns of the unexpected temperature fluctuation phenomenon, and to identify the core influencing variables.

[0051] The risk assessment module is used to take the core influencing variables as input parameters, combine them with the reference performance information of the same thermal insulation structure of the building roof, establish a standard thermal insulation performance reference line, and evaluate the thermal insulation performance of the building roof by comparing the deviation of the standard thermal insulation performance reference line with the measured thermal insulation performance curve, and determine the risk level of the thermal insulation weak point by integrating the statistical inference model.

[0052] The prediction output module is used to predict, based on the unexpected temperature fluctuations, the core influencing variables, and the risk level of the weak points in insulation, the projected output module uses sequence prediction technology to estimate the building roof areas that may experience insulation failure within a specified time interval, and generates a document on insulation optimization and maintenance solutions.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] This method for monitoring the thermal insulation of building energy-saving roofs effectively extracts system variation components and random interference components by collecting real-time data from multiple sensors and comparing it with historical databases. This provides a more reliable foundation for subsequent data analysis. This refined data processing helps eliminate irrelevant interference, focusing on key information that truly reflects changes in roof thermal insulation performance, making the detection of unexpected temperature fluctuations more targeted.

[0055] By employing pattern recognition technology to detect unexpected temperature fluctuations, its ability to identify complex patterns allows for the precise capture of abnormal temperature changes that are easily overlooked by traditional methods. This process not only improves the sensitivity to anomalies in the insulation system but also enables the early detection of potential problems, creating conditions for timely intervention.

[0056] By employing graphical modeling analysis to mine causal relationships and process co-occurrence patterns in unexpected temperature fluctuations, we can delve deeper into the core influencing variables behind these fluctuations. This in-depth analysis goes beyond mere observation of surface phenomena, identifying key factors affecting thermal insulation performance and leading to a more scientific and rational evaluation of insulation performance.

[0057] By establishing a standard reference line for thermal insulation performance and combining it with a statistical inference model to evaluate thermal insulation performance, the risk level of weak points in the insulation can be accurately determined by comparing the deviation between the measured curve and the reference line. This data-driven quantitative evaluation method avoids the bias of subjective judgment, allowing relevant personnel to clearly understand the thermal insulation status of each area of ​​the roof and providing direction for subsequent maintenance work.

[0058] Using sequence prediction technology to identify areas prone to insulation failure and generate optimized maintenance plans allows for proactive planning of maintenance work, transforming reactive maintenance into proactive prevention. This not only helps to rationally allocate maintenance resources and reduce unnecessary waste of manpower and materials, but also enables measures to be taken before insulation failure occurs, ensuring the stable operation of the roof insulation system and thus maintaining the building's energy efficiency and indoor thermal comfort.

[0059] The entire method, from data collection, processing, and analysis to evaluation, prediction, and scheme generation, forms a complete closed loop. Each link works together to improve the comprehensiveness, accuracy, and efficiency of building roof insulation monitoring, and can better meet the needs of building energy conservation for roof insulation monitoring. Attached Figure Description

[0060] Figure 1 This is a schematic diagram illustrating the working principle of the building energy-saving roof insulation monitoring system and insulation monitoring method described in this invention.

[0061] Figure 2 This is a schematic diagram illustrating the working principle of deviation amplitude measurement.

[0062] Figure 3 A working principle diagram for optimizing the evaluation results report;

[0063] Figure 4 This is a schematic diagram illustrating the working principle of detecting unexpected temperature fluctuations. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Please see Figure 1 This invention provides a building energy-saving roof insulation monitoring system and a method for monitoring insulation, the method comprising:

[0066] Real-time thermal insulation monitoring data is generated by collecting parameters such as temperature and heat flux density through a distributed sensor network. This data is input into a time series decomposition unit, which separates the system variation component determined by the building's structural characteristics and the random interference component reflecting environmental noise. A pattern recognition engine receives these two components and, by comparing them with a pre-stored database of typical temperature patterns, identifies unexpected temperature fluctuations that deviate from the normal range. For the identified temperature fluctuations, a graph model is established with sensor nodes as vertices and heat conduction paths as edges. A causal discovery algorithm is applied to extract key influencing factors as core influencing variables. Based on these variables, a standard thermal insulation performance reference line is generated using a reference performance parameter library of roofs with the same structure. The offset between the measured curve and the reference line is input into a Bayesian statistical inference unit. This unit outputs a risk score for weak points in the thermal insulation, and, combined with a sequence prediction model, estimates the coordinates of areas that may fail within a specified future period, automatically generating an optimization plan document containing location coordinates and maintenance measures.

[0067] Example 1: See Figure 2 The IoT gateway continuously receives real-time data streams from a distributed temperature sensor network deployed on the building roof, containing temperature sample values ​​with second-level timestamps. The system performs searches in a pre-stored roof insulation history database, which covers various historical performance records for specific structural types. These records include reference performance parameters for the same insulation materials used in past projects, such as historical measurements of material thermal conductivity, details of defect events recorded in annual maintenance logs, specific maintenance interventions and their effects, and historical environmental parameters collected from weather stations in the same geographical location. The search process filters and matches based on the current roof's structural type, dimensions, and geographic coordinates.

[0068] A Kalman filter algorithm is used to process the reference performance parameters retrieved from the historical database. The filter state variable is set to the thermal resistance value, and the process noise covariance matrix is ​​dynamically initialized based on the fluctuation range of historical data. Through recursive prediction and correction steps, the algorithm calculates the innovation sequence and compares it with a preset statistical confidence interval. Data points corresponding to innovations falling outside this confidence interval are identified as outliers, and the system automatically performs a removal operation. After this processing, the retained data points form a clean dataset, representing the performance benchmark of the insulation structure under normal operating conditions.

[0069] A clean dataset is input into a Gaussian process regression model for curve fitting. The model selects a radial basis function as the covariance kernel function, and its bandwidth parameter is automatically optimized based on the spectral characteristics of the historical data itself. The model fits the data through a Bayesian inference process, generating a smooth and continuous baseline curve, which is defined as the standard insulation performance reference line. Importantly, the model also outputs the predicted confidence interval band of this reference line, the width of which reflects the expected fluctuation range of the reference performance under different conditions.

[0070] The system converts real-time acquired raw temperature sensor data into a thermal resistance value sequence, forming a measured thermal insulation performance curve that varies over time. To accurately quantify the deviation between the measured curve and a standard reference line, the system performs a multi-step analysis: A set of sampling points is dynamically set at fixed intervals on the time axis of the standard thermal insulation performance reference line. At each sampling point, the system calculates the first derivative of the measured thermal insulation performance curve and compares it with the slope of the tangent line to the reference line at that point, recording the difference in slope change. Simultaneously, a dynamic time warping algorithm is used to match key inflection points. The algorithm constructs a cumulative distance matrix and searches for the minimum bending path to align the measured curve with the local maximum and local minimum points (i.e., inflection points) of the reference line. For each successfully aligned inflection point pair, its offset on the time axis and the magnitude difference on the thermal resistance value axis are recorded. A weighted fusion function is designed to integrate the various deviation indicators obtained from the above analysis—including the slope change difference at multiple sampling points and the time offset and magnitude difference of multiple aligned inflection points. The weight coefficients in the function are determined based on the sensitivity analysis results of different deviation types to thermal insulation performance degradation. The fusion operation outputs a multidimensional feature vector, which comprehensively describes the spatial offset characteristics of the measured curve relative to the reference line. This multidimensional feature vector is then input into the normalization module, where a Z-score normalization transformation is performed to make it conform to a standard normal distribution. The Euclidean distance between the normalized vector and the zero vector is calculated, and this distance value is defined as a quantitative characterization of the measured curve's deviation from the standard performance, i.e., the deviation quantification score.

[0071] Using the deviation quantification score as the core observation input, a Hidden Markov Model (HMM) is driven to predict the dynamic evolution of roof insulation performance. The model defines multiple discrete hidden states, such as "normal stable state," "slightly degraded state," "moderately risky state," and "severely failed state." A state transition probability matrix is ​​constructed, with matrix parameters initialized using historical maintenance data and performance degradation paths. The model, based on the current deviation score, infers the probability distribution of each hidden state within a specific time period using either the Viterbi algorithm or the forward-backward algorithm. When the system calculates that the cumulative probability of performance transitioning to the "moderately risky state" or "severely failed state" exceeds a preset decision threshold, the sensor-monitored coordinate point is identified as a high-risk weak point in the insulation.

[0072] The system locates these high-risk coordinate points within an integrated geographic information system platform and identifies adjacent vulnerable point clusters using spatial clustering algorithms. For each cluster, the system retrieves its corresponding building structure diagram, material types used, and recent environmental exposure history. Based on this information, the system uses a pre-defined rule engine to generate a detailed insulation optimization and maintenance plan document. This document specifies the location coordinates, recommended intervention measures, recommended material specifications, and implementation time windows. Intervention measures may include repairing localized damaged areas, replacing specific material layers, or recommending the replacement of the entire roof unit if the overall degradation assessment is severe. The document is automatically output in a standard operating report format. The entire analysis process is repeated at a set frequency or in an event-driven manner, and the time-series data of the deviation quantification score is updated and stored in a historical database in real time, providing a basis for subsequent analysis iterations.

[0073] Example 2: See Figure 3 The deviation quantification score obtained from Example 1 is used as initial input data to enter the building roof insulation health assessment process. The system initializes a Bayesian inference engine, which loads a preset prior probability distribution model. The parameters of this prior distribution are based on a building maintenance history case library built into the storage system, which contains previously recorded roof insulation defect types, occurrence frequencies, and corresponding performance deviation observations. The engine uses the input real-time deviation quantification score to update its probabilistic belief about the current roof state based on Bayes' theorem, outputting a preliminary state assessment result. This result is expressed as a probability vector of the roof being in different degradation states.

[0074] The system initiates a parallel, two-threaded optimization process to improve the performance of the inference model. In the first optimization thread, a genetic algorithm is implemented. This algorithm defines the parameter space of the inference model as a "chromosome" to be optimized. For each generation of the chromosome population, the system uses the information entropy of the state evaluation results as the core evaluation metric for the fitness function. The fitness function is designed based on the information entropy metric of the feature weights, while also considering the discriminative parameter of the state distribution. The algorithm performs evolutionary operations including selection, crossover, and mutation, aiming to maximize the output value of the fitness function. The entire iteration process is limited to the number of evolutionary cycles, with each iteration adjusting the chromosome, i.e., the model parameter set. In the second optimization thread, the conjugate gradient descent method is applied to specifically adjust the likelihood function expression structure of the statistical inference model. This method searches for a better model structure representation along the conjugate direction of the gradient. The inheritance relationship and conflict resolution mechanism set between the two threads ensure that the optimization results can be coordinated, ultimately producing a statistical inference model that is improved in both parameter space and structural expression.

[0075] This optimized statistical inference model receives a new input data stream. This data stream consists of quantified scores of temporal deviations within specific time windows. The model processes these score sequences, performing probability-based inference calculations. For each discrete time slice in the input sequence, the model outputs a probability estimate of the roof's condition at each preset health level. These probability estimates, sequenced by time slice, constitute a preliminary assessment report. This report describes the dynamic trends in the roof's insulation condition during the observation period.

[0076] The system initiates a particle swarm optimization process based on the preliminary evaluation results report. The initialization process sets a specific number of particles, each coded to represent a possible interpretation or processing scheme for the evaluation report. The selection of the evaluation scheme is based on attributes, such as differentiated interpretations of the evaluation report or different combinations of decision thresholds. For each particle's corresponding scheme, the system calculates the performance index of the evaluation report under that scheme, such as the area under the receiver operating characteristic curve in a classification evaluation task, which is a comprehensive indicator of the model's discriminative ability. The particle swarm algorithm runs within a specified number of iterations, with particles updating their velocity and position information based on their individual and swarm historical best positions. After the algorithm completes, the scheme identifier corresponding to the particle whose performance index reaches its maximum value is selected; this scheme represents the current optimal evaluation interpretation method. Based on this optimal scheme, the preliminary evaluation results report is revised or reinterpreted, generating the swarm intelligence optimization results.

[0077] To further optimize the swarm intelligence results, a rule-based and exploration-based heuristic search technique is implemented. A set of potential sources of systemic bias that could affect the accuracy of the evaluation results is established. A set of heuristic rules is formulated to guide the search direction, such as prioritizing evaluation results near recent state abrupt changes or focusing on results in low-confidence probability intervals. This heuristic rule set defines the priority and constraints for selecting search paths. Guided by the rule set, the system enumerates possible local adjustment operations, such as recalculating results for specific low-confidence time slices or fine-tuning the probability weights of different states. A set of adjustment operations that improve the logical consistency and coherence of the final evaluation results is selected. These selected operations are applied to modify the swarm intelligence optimization results, generating a finely tuned final evaluation result report.

[0078] The system constructs an integrated learning framework to synthesize the outputs of the aforementioned stages and generate a final evaluation of the building roof insulation health status. This framework defines clear input sources and processing rules: the preliminary state evaluation results generated by initial Bayesian inference serve as the foundation; the optimized statistical inference model and its parameter configuration serve as the model structure reference; and the finely tuned final evaluation result report serves as the main information input. Within the framework, the basic evaluation results serve as the baseline prediction value. The optimized report output represents correction factors or supplementary information for the baseline. Feature importance analysis results are used to assign weight coefficients to each information source. The framework integrates these multi-dimensional inputs according to a pre-defined calculation formula, such as weighted averaging or a model-based fusion strategy, to calculate and generate a single numerical quantitative score. This score is the final insulation health status evaluation index, which in engineering terms represents the ratio between the effective thermal resistance performance of the current building roof unit and the expected thermal resistance performance based on design theory. This ratio directly reflects the actual maintenance level of the roof insulation performance.

[0079] Example 3: The system constructs an analysis dataset for identified unexpected temperature fluctuations. This dataset contains raw temperature readings and their derived features from a distributed sensor network. Each record in the dataset consists of a 128-dimensional feature vector, covering time-domain statistics, frequency-domain features, and spatial correlation indicators. To reduce computational complexity and highlight key influencing factors, the system implements principal component analysis (PCA) to compress the original feature space. This process calculates the eigenvalues ​​and eigenvectors of the feature covariance matrix, retaining the top 15 principal components according to the cumulative contribution rate criterion. These principal components can explain more than 90% of the variability in the original data. Through this processing, the original high-dimensional data is transformed into low-dimensional feature vectors, forming a simplified dataset.

[0080] A graph model structure is constructed based on a simplified dataset. In the model definition, each temperature sensor node is mapped to a vertex in the graph, and the vertex attributes include the node's temperature change rate, absolute temperature value, and spatial coordinates. Based on the actual physical layout of the building roof, the system sets the generation rules for connecting edges between vertices: when the physical installation distance between two sensor nodes is less than 3 meters, an undirected edge is established between the corresponding vertices. The edge weights are calculated using the following formula:

[0081]

[0082] in This represents the weight coefficient of the edge between vertices u and v. and These represent the temperature changes at the two nodes within the observation window. The parameters are adjusted to control the impact of temperature differences on connection strength. This weighting calculation method results in stronger connections between nodes with similar temperature variation patterns.

[0083] An improved Apriori algorithm is implemented on a graph structure for frequent itemset mining. The algorithm sets a sliding time window of 10 minutes, scanning all nodes for temperature change patterns within this window. A support threshold of 0.8 is defined, meaning that only when a certain temperature change combination appears in more than 80% of the time window is it considered a high-frequency associated itemset. The algorithm employs a layer-by-layer search strategy, starting with single-node patterns and gradually expanding to multi-node combination patterns. To improve computational efficiency, the system implements a prefix tree-based storage structure and applies pruning strategies to eliminate candidate patterns that do not meet the minimum support requirement.

[0084] Granger causality tests were performed on the identified high-frequency association itemsets. A 5-minute time lag window was set in the test to examine whether the temperature change of one node could predict the future temperature of another node. The F-statistic for each node pair was calculated; when this value exceeded a preset critical value of 4.0, a directed causal edge was established between the corresponding nodes. The direction of the edge was determined by the time lag analysis, pointing from the predictor variable to the predicted variable. The edge strength attribute was recorded as the normalized value of the F-statistic. The system also simultaneously calculated the information transfer rate index for each causal path, reflecting the propagation efficiency of temperature fluctuations between nodes.

[0085] Based on the graph model topology, the system calculates the centrality indices of each node to identify key influencing factors. The indices used include degree centrality, proximity centrality, and betweenness centrality. Degree centrality measures the number of direct connections between nodes; proximity centrality reflects the average shortest path length from a node to other nodes in the graph; and betweenness centrality counts the frequency of a node's appearance in all shortest paths. The betweenness centrality index is also calculated for all edges in the graph. The system sets a threshold of 0.7 to filter out nodes and edges with centrality indices exceeding this value, forming an initial set of core variables. This set includes sensor nodes with high connectivity and information relay capabilities, along with their associated environmental parameters.

[0086] An initial set of core variables is input into an extreme gradient boosting decision tree model for importance ranking. The model uses the intensity of temperature anomalies as the target variable, which is obtained by standardizing the deviation between the actual and expected temperatures. The decision tree is set to a maximum depth of 6 layers with a learning rate of 0.3, and an early stopping strategy is employed during training to prevent overfitting. The model outputs a feature importance score for each input variable, which is used to calculate the total gain of the variable as a split point across all decision trees. The system selects variables with importance scores greater than 0.5 as the final core influencing variables. These variables typically include physical quantities that directly affect the heat conduction process, such as ventilation layer pressure, solar radiation intensity, and changes in material thermal conductivity. Each core variable is accompanied by an indicator of its direction of influence, indicating whether it promotes or inhibits temperature fluctuations.

[0087] The system maintains a dynamic core variable knowledge base, continuously updating the weight coefficients and relationships of each variable. The knowledge base is stored using a graph database structure, supporting efficient relation queries and path analysis. When new monitoring data causes a significant change in variable importance, the system automatically triggers an incremental update mechanism for the knowledge base. This mechanism compares the cosine similarity of the old and new weight vectors; when the difference exceeds a threshold, a retraining process is initiated. The knowledge base's version control function records the composition and evolution trajectory of core variables at different times, providing foundational data for long-term performance analysis.

[0088] The identification results of key influencing variables are integrated with the Building Information Modeling (BIM) system. In the 3D building model, the sensor locations corresponding to highly important variables are specially marked, and real-time monitoring values ​​and historical change curves are displayed. The system supports conditional queries based on key variables, such as filtering out all roof areas significantly affected by solar radiation intensity. This visual representation helps engineers intuitively understand the key drivers of temperature fluctuations and their spatial distribution patterns. The system also generates a key variable analysis report, detailing the physical meaning, influencing mechanism, and quantitative relationship with temperature fluctuations of each variable. The report is organized in a hierarchical structure.

[0089] Example 4: In a building roof insulation monitoring system, when conducting in-depth analysis of the identified core variable set, the system executes a multi-dimensional evaluation process to determine the strength and quality of the relationships between the variables. This process constructs analysis cases based on actual monitoring data, and the implementation details are illustrated below with specific examples.

[0090] A monitoring network consisting of 36 temperature sensors was deployed on the roof of a commercial building. The system identified a set of eight core variables. These variables include: average temperature in the southwest region (T1), air pressure difference in the ventilation layer (P2), moisture content of the insulation material (M3), solar radiation intensity (S4), wind speed (W5), nighttime cooling rate (C6), temperature difference at seams (J7), and material aging index (A8). The system established a relationship matrix between the variables, recording the interaction pattern of each pair of variables within a specific time window.

[0091] Table 1 shows the evaluation index data for the relationships between some variables.

[0092]

[0093] The system first calculates the mutual information entropy value of the relationship between the variables, which reflects the statistical dependence between the two variables. For the variables of average temperature (T1) and solar radiation intensity (S4) in the southwest region, the system statistically analyzes their joint distribution over a continuous 30-day monitoring period. Through discretization, the range of values ​​for each variable is divided into 10 equally wide intervals, and the proportion of data points falling within each two-dimensional interval combination is calculated. Based on this joint distribution matrix, the system derives the mutual information entropy value characterizing the strength of the correlation between the two variables.

[0094] Similarity analysis employed a modified Jaccard coefficient calculation method. The co-occurrence of the ventilation layer pressure difference (P2) and the moisture content of the insulation material (M3) was systematically examined. Within a defined analysis time window, the proportion of times both variables simultaneously exhibited outliers was counted relative to the total number of outliers for at least one variable. Outlier determination was based on the historical distribution of each variable, using the upper and lower 5% quantiles as thresholds. The calculation process considered time alignment tolerance, allowing outliers of two variables to occur within two consecutive sampling periods still as co-occurrences.

[0095] The path uniqueness assessment focuses on the relationship between the material aging index (A8) and the moisture content (M3) of the insulation material. The system retrieves all causal paths containing these two variables and counts the proportion of paths with the same time delay characteristic out of the total number of paths. The time delay characteristic is defined as the typical time difference between the change in the causal variable and the response of the outcome variable, with a matching window of ±15 minutes set by the system. The uniqueness score is calculated as the reciprocal of the number of paths with the same characteristic, and the final evaluation value is obtained after normalization.

[0096] The three evaluation indicators are combined in the integrated system according to preset weights. The mutual information entropy value is weighted at 50%, reflecting the importance of the basic correlation strength between variables; the Jaccard coefficient is weighted at 30%, emphasizing cooperative behavior under abnormal operating conditions; and the path uniqueness is weighted at 20%, focusing on the specificity of causal relationships. The combined calculation adopts a linear weighted summation method, and each variable relationship pair receives a comprehensive score between 0 and 1.

[0097] The system categorizes variable relationships based on a comprehensive score. Relationships with a score above 0.75 are marked as critical paths and given the highest priority in subsequent analysis. Relationships with scores between 0.6 and 0.75 are considered secondary paths and used to support decision-making. Relationships with scores below 0.6 are listed as reference paths and are only analyzed when additional computational resources are available. This classification method allows the system to focus on the most influential variable interaction patterns.

[0098] In the commercial building case study, the relationship between average temperature and solar radiation intensity in the southwest region scored high at 0.87, classifying it as a critical path. The system recorded detailed characteristics of this relationship: mutual information entropy values ​​showed a strong correlation; the Jaccard coefficient indicated a synchronous temperature increase in the southwest region under 87% strong solar radiation; and path uniqueness showed the typicality of this heat transfer path. In contrast, the relationship between wind speed and solar radiation intensity scored 0.67, classifying it as a less important path, characterized by a moderate statistical correlation and co-variation at certain times.

[0099] The comprehensive scoring report is generated using a standardized template and consists of three main parts. The first part lists all assessed relationship pairs and their respective index values, arranged in descending order of comprehensive score. The second part provides a detailed interpretation of the critical path, including explanations of the physical mechanisms and descriptions of typical operating conditions. The third part includes visualization elements, using a heatmap to display the variable relationship network; node size reflects variable importance, and edge thickness indicates relationship strength.

[0100] The system is equipped with a dynamic update mechanism. When new monitoring data causes a change in a certain indicator exceeding 15%, the overall score of the affected relationship is automatically recalculated. The update process takes into account the decay effect of historical scores, giving higher weight to more recent data. This mechanism ensures that the evaluation results reflect the latest status of the roof insulation system.

[0101] The report output module supports multiple format conversions, including structured database records, interactive web views, and print-optimized document layouts. Users can choose the appropriate presentation method as needed, and the system maintains data consistency across versions. In a commercial building case, the engineering team used the web view to focus on critical path analysis and discovered a strong correlation between temperature anomalies and solar radiation in the southwest region. Following this discovery, shading measures were implemented in that area.

[0102] The assessment results are integrated with the building operations and maintenance system. When variables involved in the critical path show anomalies, the system triggers an early warning mechanism and suggests targeted inspection measures. For example, if the relationship between the air pressure difference in the ventilation layer and the moisture content of the insulation material is marked as a critical path, the system will prioritize checking for potential water seepage points when it detects an anomaly in the air pressure.

[0103] Example 5: See Figure 4 A grid-like temperature sensor array with dimensions of 12 rows × 12 columns is deployed at the building roof monitoring site, with each sensor having a unique spatial location identifier. The system acquires temperature readings from the sensor nodes at fixed time intervals, forming a continuous time-series data stream. The raw temperature data includes three basic pieces of information: node coordinates, sampling timestamp, and Celsius temperature value. This information is transmitted to the data preprocessing module in a structured format.

[0104] The time series decomposition unit performs seasonal decomposition. The algorithm presets the main period length to 24 hours, corresponding to a complete cycle of a natural day. The decomposition process uses a locally weighted regression method to initially smooth the temperature series, fitting a basic trend line. This trend line captures long-term variation characteristics, unaffected by short-term intraday fluctuations, forming a long-term trend component. Subsequently, sub-period characteristics are identified, extracting repeating patterns from different intraday time periods using 8-hour intervals. The identified cyclical fluctuations constitute the seasonal cycle component. The difference between the original series values ​​and the values ​​of the above two components at the same time constitutes the residual component, which contains fluctuation elements that cannot be attributed to the long-term trend or cyclical rhythm.

[0105] The long-term trend component is fed into a multi-level decomposition processor. This processor employs discrete wavelet transform technology, selecting specific types of wavelet basis functions as the analysis tool. The decomposition level is set to 5 levels, progressively extracting sub-band components at different frequency scales from the original sequence. The fifth-level approximation coefficients reflect the slowest changing characteristics, representing the inherent thermodynamic inertia of the system. These coefficients are stored independently as system variation components, their numerical sequences encoded in a dedicated memory area, and accompanied by corresponding time coordinates.

[0106] The seasonal periodic component and the residual component are merged into a composite signal stream input to an adaptive noise filter. This filter employs a recursive estimation algorithm, with a core dynamic adjustment strategy based on a prediction correction mechanism. The filter is configured with a sliding window mechanism, setting the window coverage time span to 30 minutes. Within each time step, the algorithm dynamically updates the parameter matrix of the system state equation based on the statistical characteristics of the data within the window. A key parameter is the estimated process noise covariance, which is automatically calibrated based on the variance values ​​of the most recent 30 sampling points, avoiding insufficient adaptability caused by manual presets. The filter output signal undergoes quantization analysis, maintaining the output signal-to-noise ratio parameter above a specific level. The effective signal components generated in this process are defined as random interference components. These interference components are stored in groups according to sensor nodes, preserving complete spatiotemporal correlation information.

[0107] Systematic variation components and random disturbance components are fed in parallel to the pattern recognition module. This module constructs a composite deep neural network architecture containing 32 cascaded residual computation units. Each residual unit consists of two layers of convolutional operations and shortcut connections, with a modified linear unit variant used as the activation function. The network input layer simultaneously receives multi-channel data of both components: the systematic variation component occupies the first channel, the random disturbance component occupies the second channel, and the third channel inputs the spatial topology feature map. The entire network structure has nine parallel classification output nodes, each corresponding to a preset standard temperature pattern category. These nine preset patterns cover common roof thermal behavior states, such as the linear rising pattern corresponding to the ideal response under stable heating, the step-decreasing pattern representing the heat dissipation process caused by sudden rainfall, and the oscillating steady pattern reflecting the impact of periodic start-stop of the air conditioning system. The network acquires pattern discrimination ability through end-to-end training, with training samples derived from a library of typical temperature curves under historical normal operating conditions.

[0108] During real-time monitoring, when the probability of matching the temperature change trajectory of a certain sensor node with all nine preset patterns is lower than a preset threshold, the system triggers the unexpected temperature fluctuation event identification process. The event annotation stage records four core features: the fluctuation start time is determined based on the sampling time of the first significant deviation from the expected value; the peak size is captured using a local maximum detection algorithm; and the spatial distribution pattern generates a two-dimensional density heat map based on the locations of nodes affected by the event at the same time. All event features are stored in a structured spatiotemporal event log. Log entries include a globally unique event identifier, a list of involved sensor nodes, complete time trajectory data, and a snapshot of the associated original measurement value. The system adds newly identified event feature vectors to an incremental learning queue, periodically updating the classification boundary of the pattern recognition model to adapt to the changing characteristics of the roof thermal environment with seasonal and usage patterns.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring the thermal insulation of building energy-saving roofs, characterized in that... include: Real-time thermal insulation monitoring data of building roofs is collected from multiple sensors. The real-time thermal insulation monitoring data is compared and analyzed with a pre-stored historical database of roof thermal insulation to extract the system variation component and random interference component in the real-time thermal insulation monitoring data. Based on the system variation component and the random interference component, pattern recognition technology is used to detect unexpected temperature fluctuations in the real-time thermal insulation monitoring data. Using graphical modeling analysis, causal relationship mining and pattern co-occurrence correlation processing were performed on the unexpected temperature fluctuation phenomenon to identify the core influencing variables; Using the core influencing variables as input parameters, and combining them with reference performance information of the same thermal insulation structure of building roofs, a standard thermal insulation performance reference line is established. By comparing the deviation between the standard thermal insulation performance reference line and the measured thermal insulation performance curve, the thermal insulation performance of building roofs is evaluated by integrating a statistical inference model, and the risk level of thermal insulation weak points is determined. Based on the unexpected temperature fluctuations, the core influencing variables, and the risk level of the weak points in insulation, sequence prediction technology is used to estimate the building roof areas that may experience insulation failure within a specified time interval, and a document on insulation optimization and maintenance plan is generated.

2. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 1, characterized in that, The process involves using the core influencing variables as input parameters, combining reference performance information from similar insulation structures on building roofs to establish a standard insulation performance reference line, and evaluating the building roof insulation performance by comparing the deviation between the standard insulation performance reference line and the measured insulation performance curve using a statistical inference model. This process determines the risk level of weak points in the insulation, including: Using the core influencing variables as input parameters, combined with reference performance information of the same thermal insulation structure of the building roof, data filtering technology is applied to remove outliers from information from various historical performance records, and the data after cleaning is output. Curve fitting method is used to model the data after cleaning to form a standard thermal insulation performance reference line. The various historical performance records include: reference performance information of the same thermal insulation structure, measured thermal insulation parameters of the target roof at multiple time points, records of past defect events, and maintenance files. Compare the measured thermal insulation performance curve with the standard thermal insulation performance reference line, measure the deviation of the measured thermal insulation performance curve relative to the standard thermal insulation performance reference line, and use the average deviation index to measure the deviation range to obtain a deviation quantification score. Based on the aforementioned deviation quantification score, a statistical inference model is used to evaluate the thermal insulation performance of the building roof, generating an initial evaluation report. The initial evaluation report is then optimized using search optimization techniques to obtain a health assessment of the building roof's thermal insulation status. Using the aforementioned building roof insulation health status evaluation and combined with the trend prediction model, the evolution path of building roof insulation performance within a specified time interval is predicted to obtain a performance evolution prediction report. The performance evolution prediction report is then analyzed to determine the risk level of insulation weakness points.

3. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 2, characterized in that, The evaluation of building roof thermal insulation performance is performed based on the deviation quantification score and a statistical inference model, generating an initial evaluation report. This initial evaluation report is then optimized using search optimization techniques to arrive at a health assessment of the building roof thermal insulation status, including: A Bayesian inference system is used to perform statistical evaluation processing on the deviation quantification score to generate preliminary state evaluation results; Based on the preliminary state evaluation results, the statistical inference model is adjusted in conjunction with the evolutionary algorithm and the gradient descent method to form an optimized statistical inference model. The deviation quantification score is input into the optimized statistical inference model, inference calculation is performed, an evaluation result report is output, and the evaluation result report is optimized using swarm intelligence optimization algorithm and heuristic search technology to generate an optimized evaluation result report. Based on the ensemble learning framework, the preliminary state evaluation results, the optimized statistical inference model, and the optimized evaluation result report are combined to generate a health assessment of the building roof insulation status.

4. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 3, characterized in that, The process involves inputting the quantified deviation score into the optimized statistical inference model, performing inference calculations, outputting an evaluation result report, and then using a swarm intelligence optimization algorithm and heuristic search technology to optimize the evaluation result report, generating an optimized evaluation result report, including: Using the optimized statistical inference model, the deviation quantification score is used as the input dataset, and probabilistic inference processing is performed to output a preliminary assessment of the building roof insulation health status. A swarm intelligence optimization algorithm is used to perform a global exploration operation on the preliminary building roof thermal insulation health status assessment to find the optimal assessment solution. The preliminary building roof thermal insulation health status assessment is then optimized based on the optimal assessment solution to obtain the swarm intelligence optimization result. Based on the swarm intelligence optimization results, the preliminary building roof thermal insulation health status assessment is further optimized using heuristic search techniques to generate an optimized assessment result report.

5. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 1, characterized in that, The graphical model analysis method is used to mine causal relationships and process co-occurrence patterns in the unexpected temperature fluctuation phenomenon, identifying the core influencing variables, including: An analysis dataset is constructed using the unexpected temperature fluctuation phenomenon. The dataset is then compressed using feature dimensionality reduction techniques to obtain a simplified dataset. A graph model analysis method is then used to scan the simplified dataset to identify high-frequency associated itemsets. The Granger causality test is used to detect the high-frequency association itemset to obtain the causal chain of the real-time thermal insulation monitoring data, and the co-occurrence association of patterns is extracted from the high-frequency association itemset. The core variable set is obtained by combining the causal chain and the co-occurrence association of patterns. Using mutual information metrics, similarity coefficients, and uniqueness indicators as evaluation criteria, the causal chains and pattern co-occurrence associations of the core variable set are evaluated, and a comprehensive scoring report is generated. Based on the comprehensive scoring report, candidate core influencing variables are selected from the set of high-frequency related items. The decision tree model is then used to rank the candidate core influencing variables by importance, thereby identifying the core influencing variables.

6. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 5, characterized in that, The evaluation criteria, including mutual information measurement, similarity coefficient, and uniqueness index, are used to assess the causal chains and pattern co-occurrence associations of the core variable set, generating a comprehensive scoring report, including: Calculate the mutual information measure of causal chains and pattern co-occurrence associations in the core variable set to obtain an association strength score; Similarity coefficients are used to perform similarity analysis on causal chains and pattern co-occurrence associations in the core variable set, and the similarity of causal chains and pattern co-occurrence associations in the core variable set is measured to obtain a similarity score; The uniqueness index is used to perform a uniqueness assessment on the causal chain and the co-occurrence association of the pattern, and the uniqueness score of the causal chain and the co-occurrence association of the pattern is calculated to obtain a uniqueness score; By integrating the association strength score, the similarity score, and the uniqueness score, a comprehensive scoring operation is performed on each causal chain and pattern co-occurrence association in the core variable set to generate a comprehensive scoring report.

7. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 1, characterized in that, The method of detecting unexpected temperature fluctuations in the real-time thermal insulation monitoring data using pattern recognition technology based on the system variation components and the random interference components includes: Based on the system variation components and the random interference components, pattern recognition technology is used to classify the real-time thermal insulation monitoring data, detect multiple normal temperature patterns and abnormal temperature patterns, and obtain the classification results based on the normal temperature patterns and the abnormal temperature patterns. By comparing the classification results with known standard temperature patterns, real-time thermal insulation monitoring data that do not conform to the standard temperature patterns are detected and marked as unexpected temperature fluctuations.

8. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 1, characterized in that, The step of extracting the system change component and random interference component from the real-time thermal insulation monitoring data includes: acquiring a multi-dimensional temperature sensing data stream of the building roof, and using time series decomposition technology to separate the temperature sensing data stream into long-term trend components, seasonal cycle components and residual components. The long-term trend component is processed by wavelet transform to generate low-frequency characteristic coefficients, which are used as system variation components. The seasonal cycle component and the residual component are input into the noise filtering model, and the purified high-frequency fluctuation characteristics are output as random interference components through adaptive threshold filtering. The low-frequency feature coefficients and the purified high-frequency fluctuation features are synchronously transmitted to the pattern recognition technology processing module.

9. The method for monitoring the thermal insulation of building energy-saving roofs according to claim 2, characterized in that, The deviation of the measured thermal insulation performance curve relative to the standard thermal insulation performance reference line includes: setting a dynamic sampling point set on the standard thermal insulation performance reference line and calculating the rate of change of the tangent slope of the measured thermal insulation performance curve at each sampling point. The key inflection points of the standard thermal insulation performance reference line and the measured thermal insulation performance curve are aligned using a dynamic time warping algorithm to generate a set of inflection point offset distances. A weighted fusion operation is performed on the set of tangent slope change rate and inflection point offset distance to output a multidimensional deviation feature vector; The multidimensional deviation feature vector is input into the average deviation index measurement module to generate a deviation quantification score.

10. A building energy-saving roof insulation monitoring system, used to implement the building energy-saving roof insulation monitoring method as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module is used to collect real-time thermal insulation monitoring data of building roofs from multiple sensors, compare and analyze the real-time thermal insulation monitoring data with a pre-stored roof thermal insulation historical database, and extract the system change component and random interference component in the real-time thermal insulation monitoring data. Anomaly detection module is used to detect unexpected temperature fluctuations in the real-time thermal insulation monitoring data based on the system variation components and the random interference components, using pattern recognition technology. The variable identification module is used to use graphical model analysis methods to mine causal relationships and process co-occurrence patterns of the unexpected temperature fluctuation phenomenon, and to identify the core influencing variables. The risk assessment module is used to take the core influencing variables as input parameters, combine them with the reference performance information of the same thermal insulation structure of the building roof, establish a standard thermal insulation performance reference line, and evaluate the thermal insulation performance of the building roof by comparing the deviation of the standard thermal insulation performance reference line with the measured thermal insulation performance curve, and determine the risk level of the thermal insulation weak point by integrating the statistical inference model. The prediction output module is used to predict, based on the unexpected temperature fluctuations, the core influencing variables, and the risk level of the weak points in insulation, the projected output module uses sequence prediction technology to estimate the building roof areas that may experience insulation failure within a specified time interval, and generates a document on insulation optimization and maintenance solutions.

Images