Smart home cooperative control method based on Internet of Things

Abstract

The invention relates to the field of intelligent equipment control, and provides a smart home cooperative control method based on the Internet of Things. The method comprises the following steps: analyzing historical state data of a smart home environment through machine learning to obtain an equipment association map, and encoding collected real-time state data to obtain a global state vector; clustering by combining the global state vector and the equipment association map to obtain a life scene mode set, extracting a trigger condition to form a library, and inputting the global state vector into a neural network to predict scene occurrence probability distribution; determining standard equipment state configuration corresponding to a target scene in the trigger condition formation library according to scene occurrence probability distribution, and performing multi-target optimization on standard equipment state control to obtain a cooperative control strategy list; and executing instructions in groups on the cooperative control strategy list according to the equipment network partitions, and performing monitoring compensation in the execution process. According to the invention, the scene intelligent control accuracy of the intelligent equipment is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent device control technology, and in particular to an intelligent home collaborative control method based on the Internet of Things. Background Technology

[0002] With the rapid development of IoT technology, smart home systems have gradually entered thousands of households. Smart home systems, through the deployment of various sensor nodes and smart devices, can perceive the status of the home environment in real time and execute corresponding control operations. Traditional smart home control methods mainly rely on preset rules or manual user operation. However, with the increase in the number of home devices and the diversification of user needs, how to achieve intelligent collaborative control between devices has become an important research direction in the smart home field. In recent years, machine learning and artificial intelligence technologies have provided new technical means to improve the intelligence level of smart homes.

[0003] Existing smart home control methods typically employ independent control of single devices or simple scene mode switching, lacking in-depth exploration and utilization of the relationships between devices. In practical applications, complex spatial adjacency, functional complementarity, and usage relationships exist between different devices. Existing methods fail to fully consider the synergistic effects between these devices, leading to problems such as unreasonable device activation order, poor energy consumption optimization, and difficulty in balancing user comfort and system efficiency when executing scene control. Furthermore, existing smart home scene recognition methods usually directly perform cluster analysis on device state sequences, failing to consider the impact of device relationships on scene recognition accuracy. In practical applications, strongly correlated device combinations often activate simultaneously in specific scenarios, and this correlation information is crucial for accurately identifying users' life scene patterns. Existing methods use a uniform distance metric when clustering state sequences, ignoring the impact of device association weights on sample similarity calculations, resulting in deviations between scene recognition results and actual user behavior patterns, thus affecting the accurate extraction of scene triggering conditions. Summary of the Invention

[0004] This invention provides a smart home collaborative control method based on the Internet of Things to overcome the shortcomings of existing technologies.

[0005] This invention provides a smart home collaborative control method based on the Internet of Things, comprising: S1. Analyze historical state data of the smart home environment through machine learning to obtain a device association map, and encode the collected real-time state data to obtain a global state vector; S2. Combine the global state vector with the device association graph to obtain a set of life scene patterns by clustering and extract the trigger conditions to form a library. Input the global state vector into the neural network to predict the probability distribution of scene occurrence. S3. Determine the standard device state configuration corresponding to the target scenario in the triggering condition formation library according to the scenario occurrence probability distribution, perform multi-objective optimization on the standard device state control, and obtain a list of collaborative control strategies. S4. Execute instructions in groups according to the list of collaborative control strategies based on device network partitions, and perform monitoring and compensation during execution to complete smart home collaborative control based on the Internet of Things.

[0006] According to the IoT-based smart home collaborative control method provided by the present invention, step S1 further includes: S11. Assign device identifiers and spatial location labels to sensor nodes in the smart home environment to obtain a set of sensor nodes; S12. Extract historical status data from the sensor node set, and perform inter-device correlation analysis on the historical status data using machine learning algorithms to obtain a device correlation map; S13. Collect real-time status information of all devices in the sensor node set at fixed time intervals, and convert the status information into a numerical vector format through vectorization encoding to obtain a global status vector.

[0007] According to the IoT-based smart home collaborative control method provided by the present invention, step S12 further includes: S121. Extract historical status data of multiple devices from the sensor node set, sort the historical status data by timestamp, and obtain the device activation time sequence. S122. Perform co-occurrence frequency statistics on the device activation time series through a sliding time window, calculate the number of times any two devices are activated simultaneously within the same time window, and obtain the co-occurrence matrix; S123. Apply correlation coefficient calculation to the co-occurrence matrix to quantify the correlation strength between devices and obtain the correlation strength matrix; S124. Construct a weighted undirected graph based on the association strength matrix, with device identifiers as nodes and association strength values ​​as edge weights, to obtain the device association graph.

[0008] According to the smart home collaborative control method based on the Internet of Things provided by the present invention, step S2 further includes: S21. The global state vector is segmented according to a time window, and the segmented state sequence is clustered by the DBSCAN clustering algorithm in combination with the association weight in the device association map to obtain a set of life scene patterns. S22. Extract the preconditions of the life scene pattern set through the association rule mining algorithm to obtain the scene trigger condition library; S23. Input the global state vector, the scene semantic labels and time features in the set of life scene patterns into the LSTM neural network for sequence prediction processing to obtain the occurrence probability distribution.

[0009] According to the IoT-based smart home collaborative control method provided by the present invention, step S21 further includes: S211. The global state vector is sliced ​​according to a preset time window to obtain a state vector sample sequence; S212. Extract the association weight between any two devices from the device association map, construct the distance metric function of the state vector sample sequence through the weighted Euclidean distance formula, and obtain the weighted distance matrix between the samples. S213. Density clustering of the state vector sample sequence is performed using the DBSCAN clustering algorithm in conjunction with the weighted distance matrix to obtain multiple state clusters. S214. Perform statistical analysis on the samples in each state cluster to extract the frequency of occurrence and combination patterns of device states, and obtain the typical features of the cluster. Perform semantic interpretation based on the typical features of the cluster, assign scene semantic labels to each cluster, and obtain a set of life scene patterns.

[0010] According to the IoT-based smart home collaborative control method provided by the present invention, step S22 further includes: S221. Extract the time features, sensor status features, and device activity features of each scene from the set of life scene patterns, and obtain the scene pre-feature matrix through feature vectorization encoding. S222. Frequent itemset mining is performed on the scene pre-feature matrix using the Apriori algorithm to obtain the scene triggering rule set; S223. Establish a classification index for the scene triggering rule set according to scene semantic tags, and associate each scene semantic tag with a corresponding triggering condition combination to obtain a scene triggering condition library.

[0011] According to the IoT-based smart home collaborative control method provided by the present invention, step S222 further includes: S2221. Set the minimum support threshold and minimum confidence threshold for the Apriori algorithm; S2222: Perform a single-item frequent set scan on the scene pre-feature matrix, count the occurrence frequency of each feature item, and filter to obtain single-item frequent sets with support greater than or equal to the minimum support threshold. S2223. Perform join and pruning operations on the single-item frequent set to generate candidate multi-item frequent sets, calculate the support of the candidate multi-item frequent sets, and filter to obtain multi-item frequent sets that are greater than or equal to the minimum support threshold. S2224. Generate association rules based on the multiple frequent sets, calculate the confidence level of the association rules, and filter out association rules with a confidence level greater than or equal to the minimum confidence level threshold to obtain a scene triggering rule set.

[0012] According to the IoT-based smart home collaborative control method provided by the present invention, step S3 further includes: S31. Filter out target scenarios according to the occurrence probability distribution, and extract the standard device status configuration corresponding to the target scenario from the scenario triggering condition library; S32. The optimal combination of equipment control parameters is obtained by performing multi-objective optimization on the standard equipment state configuration using the particle swarm optimization algorithm. S33. Based on the association weights in the device association graph, the optimal device control parameter combination is processed for execution timing allocation to obtain a list of collaborative control strategies.

[0013] According to the IoT-based smart home collaborative control method provided by the present invention, in step S32, the expression of the objective function for multi-objective optimization of the standard device state configuration is as follows: ;in, Let be the overall objective function. The objective function is user comfort. These are the weighting coefficients corresponding to the user comfort objective function. Let the energy efficiency objective function be... These are the weighting coefficients corresponding to the energy efficiency objective function. Let the objective function be the device response speed. The weighting coefficient corresponding to the device response speed. The objective function for system security is... These are the weighting coefficients corresponding to the system security objective function; ;in, This is the comfort deviation weighting coefficient. For sensor index values, This represents the total number of comfort-related sensors. For the first Environmental state values ​​sensed by each sensor For the first User preference values ​​for each sensor, For the first The maximum tolerance value of each sensor. For the first Minimum tolerance value for each sensor The comfort balance adjustment coefficient is used. This is the variance calculation function. For comfort-related sensors controlled by parameters The state value under the given state; ;in, The weighting coefficient is the average energy consumption. For electrical equipment index value, The total number of electrical devices participating in the coordinated control. For the first The equipment in control parameters Real-time power under, For the first The continuous working time of the equipment. This is the maximum allowable load power of the household power grid. For scene time period, For the first The rated power of the equipment; ;in, For response speed weighting coefficient, For response delay adjustment coefficient, The average response latency for all devices. The maximum threshold for device response latency. This is the minimum threshold for device response latency; ;in, For safety weighting coefficients, For security-related device index values, The total number of safety-related devices, For the first Risk assessment value of Taiwan's safety equipment For the first The maximum risk value of Taiwan's safety equipment For the first Minimum risk value of Taiwan's safety equipment This is the risk accumulation adjustment coefficient.

[0014] According to the smart home collaborative control method based on the Internet of Things provided by the present invention, step S4 further includes: S41. Group the collaborative control policy list according to the network partition to which the device belongs to obtain the execution instruction set of multiple partitions; S42. Send execution instruction sets to the device according to the execution sequence through the edge control node; S43. Synchronously monitor changes in the global state vector. When the deviation between the actual state and the expected state exceeds a preset threshold, compensate and adjust the execution parameters to obtain the execution log.

[0015] This invention provides an IoT-based smart home collaborative control method. First, by constructing a device association graph and integrating its association weights into the scene recognition process, it can more accurately capture the collaborative activation patterns between devices. This makes the clustered life scene patterns more closely match users' actual behavioral habits, avoiding the scene recognition bias problem caused by traditional methods neglecting device associations. Based on this, the invention uses the Apriori algorithm to perform frequent itemset mining on the scene's pre-features, automatically extracting scene triggering conditions with high support and high confidence. Compared to manually preset rules, this significantly reduces the complexity of system configuration and the cost of manual intervention, while making scene triggering more intelligent and personalized. Furthermore, the invention uses an LSTM neural network to perform temporal prediction of the global state vector, enabling it to predict the scene the user is about to enter, creating a time window for device preparation and collaborative startup, effectively shortening the system response time perceived by the user. In the control strategy generation stage, this invention employs a particle swarm optimization algorithm to optimize a multi-objective function encompassing user comfort, energy efficiency, device response speed, and system safety. This algorithm finds the optimal balance between objectives while satisfying safety and device capability constraints, ensuring user comfort, effective energy consumption control, and balancing system safety and response speed. Finally, this invention further enhances the overall effectiveness of collaborative control and system stability by allocating execution timing based on the association weights in the device association graph. This prioritizes the coordinated startup of highly associated devices, preventing instantaneous power overload or poor control performance caused by improper device startup order. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A schematic diagram of a smart home collaborative control method based on the Internet of Things provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the process for generating a set of life scene patterns according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the process for generating a scenario triggering condition library according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] like Figure 1 As shown, the present invention provides a smart home collaborative control method based on the Internet of Things, comprising: S1. Analyze historical state data of the smart home environment through machine learning to obtain a device association map, and encode the collected real-time state data to obtain a global state vector.

[0020] Step S1 further includes: S11. Assign device identifiers and spatial location labels to the sensor nodes in the smart home environment to obtain a set of sensor nodes.

[0021] In step S11, the present invention generates a unique device identifier for each sensor node in the smart home environment. The device identifier is a string of numbers or characters used to distinguish different sensor nodes. Subsequently, the present invention labels the sensor nodes with spatial location tags based on their actual deployment location. These spatial location tags are text labels, such as "living room," "bedroom," or "kitchen," used to describe the room or area where the sensor node is located. The present invention then associates and binds the device identifiers with the spatial location tags to form a sensor node set. Each record in the final sensor node set contains attribute information such as the device identifier, spatial location tag, and device type.

[0022] S12. Extract historical status data from the sensor node set, and perform inter-device correlation analysis on the historical status data using machine learning algorithms to obtain a device correlation map.

[0023] In step S12, the present invention reads the status data recorded by each device within a historical time period from the sensor node set. The historical status data includes fields such as timestamp, device identifier, and status value. Subsequently, the present invention inputs the historical status data into a machine learning algorithm for inter-device association analysis. The present invention uses a machine learning algorithm to calculate the association strength between devices by analyzing the temporal relationship and co-occurrence pattern of the status changes of different devices, and finally outputs a device association graph, which represents the association relationship between devices in a graph structure. In the graph, nodes correspond to devices, and the weight of the edges represents the magnitude of the association strength.

[0024] Step S12 further includes: S121. Extract historical status data of multiple devices from the sensor node set, sort the historical status data by timestamp, and obtain the device activation time series.

[0025] Furthermore, in step S121, the present invention extracts the status data records of each device within a historical time period from the sensor node set. Each historical status data record contains three fields: device identifier, timestamp, and status value. Subsequently, the present invention sorts the historical status data in ascending order according to the timestamp field, arranging all records in chronological order. The sorted data forms a device activation time series. In the obtained device activation time series, each record identifies a device that underwent a status change or was activated at a certain moment.

[0026] S122. Perform co-occurrence frequency statistics on the device activation time series through a sliding time window, calculate the number of times any two devices are activated simultaneously within the same time window, and obtain the co-occurrence matrix.

[0027] In step S122, the present invention sets a sliding time window with a fixed length, and then places the sliding time window at the beginning of the device activation time sequence. Subsequently, the present invention counts the number of times any two devices are simultaneously activated within the current window position. Specifically, the present invention first iterates through all activation records within the window. For each pair of devices, if both devices have activation records within the window, the co-occurrence count for this pair of devices is incremented by 1. Then, the present invention moves the sliding window forward by a fixed step, for example, 1 minute, and repeats the above statistical process at the new window position until the sliding window reaches the end of the sequence. Finally, the present invention accumulates the statistical results for all window positions to obtain a co-occurrence matrix between devices. In the co-occurrence matrix, rows and columns correspond to device identifiers, and the elements in the matrix represent the total number of times two devices are simultaneously activated across all time windows.

[0028] S123. Apply correlation coefficient calculation to the co-occurrence matrix to quantify the correlation strength between devices and obtain the correlation strength matrix.

[0029] In step S123, the present invention calculates the correlation coefficient for the co-occurrence frequency of each pair of devices in the co-occurrence matrix. The correlation coefficient is calculated using the Pearson correlation coefficient formula to measure the linear correlation between the activation modes of the two devices. Subsequently, the calculated correlation coefficient is used as the association strength value for that pair of devices. The association strength value ranges from -1 to 1; the closer the value is to 1, the stronger the positive correlation between the activation modes of the two devices. Finally, the present invention organizes the association strength values ​​of all device pairs into an association strength matrix. The association strength matrix is ​​a symmetric matrix, and the elements in the matrix represent the association strength value between the two devices.

[0030] S124. Construct a weighted undirected graph based on the association strength matrix, with device identifiers as nodes and association strength values ​​as edge weights, to obtain the device association graph.

[0031] In step S124, the present invention constructs a weighted undirected graph based on the association strength matrix. Specifically, during graph construction, the present invention creates a graph node for each device identifier. Then, the present invention traverses the association strength matrix. For device pairs in the matrix whose association strength value is greater than a preset threshold, an edge is created between the corresponding two nodes, and the weight of the edge is set to the association strength value of that device pair. Finally, the present invention uses the constructed weighted undirected graph as a device association graph, stored in a graph data structure, including a node set, an edge set, and an edge weight mapping table, supporting queries on the association weight between any two devices.

[0032] S13. Collect real-time status information of all devices in the sensor node set at fixed time intervals, and convert the status information into a numerical vector format through vectorization encoding to obtain a global status vector.

[0033] In step S13, the present invention sets a fixed time interval, such as every second or every 5 seconds. Then, at the end of each time interval, the present invention triggers a data acquisition operation to read the current status information from all devices in the sensor node set. The read real-time status information includes various data such as the device's on / off status, operating parameters, and sensor measurements. Subsequently, the present invention performs vectorized encoding processing on the real-time status information, converting different types of status data into numerical vector format. For discrete variables such as on / off status, one-hot encoding is used to convert them into 0-1 vectors. For continuous variables such as temperature and humidity, normalization processing is performed to map them to the 0-1 interval. Finally, the present invention concatenates the encoded vectors of all devices according to the order of device identifiers to form a global state vector. The global state vector is a one-dimensional numerical array, the length of which is equal to the total number of all device status features. Each element in the array corresponds to the encoded value of a certain status feature of a device.

[0034] S2. Combine the global state vector with the device association graph to obtain a set of life scene patterns and extract the trigger conditions to form a library. Input the global state vector into the neural network to predict the probability distribution of scene occurrence.

[0035] like Figure 2 As shown, step S2 further includes: S21. The global state vector is segmented according to a time window, and the segmented state sequence is clustered by the DBSCAN clustering algorithm in combination with the association weight in the device association map to obtain a set of life scene patterns.

[0036] In step S21, the present invention segments the global state vector according to a preset time window length. Specifically, the present invention sets the time window length to 5 minutes, and extracts a segment every 5 minutes starting from the beginning of the time series of the global state vector. Each segment contains the global state vector data for all moments within that time window. Subsequently, the present invention extracts the association weights between devices from the device association graph and incorporates these association weights into the clustering process. The incorporation method is to construct a weighted distance metric function, assigning greater weight to the device state differences with higher association weights when calculating the distance between two state vector samples. Next, the present invention performs density clustering processing on the segmented state sequences using the DBSCAN clustering algorithm. The present invention uses the DBSCAN algorithm to identify density-connected samples based on the weighted distance metric function, grouping density-reachable samples into the same cluster, and outputting multiple state clusters. Finally, the present invention performs statistical analysis on each state cluster, extracts the typical features of the cluster, and assigns scene semantic labels to the clusters based on the typical features, forming a set of life scene patterns. The obtained set of life scene patterns contains multiple scenes, each scene consisting of scene semantic labels and corresponding device state combination patterns.

[0037] Step S21 further includes: S211. The global state vector is sliced ​​according to a preset time window to obtain a state vector sample sequence.

[0038] Furthermore, in step S211, the present invention sets the length of the preset time window to 5 minutes. Then, starting from the beginning of the time series of the global state vector, the present invention extracts a data segment every 5 minutes, each data segment containing all global state vectors within that 5-minute time window. After extraction, the present invention treats the extracted data segment as a sample and continues the slicing operation until the end of the time series, forming a state vector sample sequence. Each sample in the obtained state vector sample sequence corresponds to the state data of a time window.

[0039] S212. Extract the association weight between any two devices from the device association map, construct the distance metric function of the state vector sample sequence using the weighted Euclidean distance formula, and obtain the weighted distance matrix between the samples.

[0040] In step S212, the present invention first queries the association weight between any two devices from the device association graph. Then, the present invention constructs a weighted Euclidean distance formula as the distance metric function. This weighted Euclidean distance formula calculates the weighted sum of the differences in each dimension when calculating the distance between two state vector samples. Specifically, for two samples x and y, the weighted Euclidean distance is calculated as the square root of the sum of the squares of the differences in each dimension and the products of their corresponding weights. The weights corresponding to the dimensions are derived from the association weights of the devices to which that dimension belongs; devices with higher association weights have a larger proportion in the distance calculation. Subsequently, the present invention iterates through all sample pairs in the state vector sample sequence, applying the weighted Euclidean distance formula to calculate the distance value between each pair of samples, forming a weighted distance matrix between the samples.

[0041] S213. Density clustering of the state vector sample sequence is performed using the DBSCAN clustering algorithm in conjunction with the weighted distance matrix to obtain multiple state clusters.

[0042] In step S213, the present invention first sets two parameters for the DBSCAN clustering algorithm: neighborhood radius and minimum number of samples. The neighborhood radius defines the neighborhood range of a sample, and the minimum number of samples defines the criterion for determining core samples. Then, the present invention traverses each sample in the state vector sample sequence, searching for all neighbor samples within the neighborhood radius of that sample based on the weighted distance matrix. These neighbor samples are used to characterize samples whose weighted distance to the current sample is less than the neighborhood radius. Subsequently, the present invention counts the number of neighbors for each sample, marking samples with a number of neighbors greater than or equal to the minimum number of samples as core samples, and marking samples with a number of neighbors less than the minimum number of samples as boundary samples or noise samples.

[0043] After obtaining the core samples, this invention starts from any unvisited core sample and performs density connectivity expansion processing. During the expansion process, this invention adds all neighbor samples of the current core sample to the same cluster, and recursively processes the core samples among the neighbors, adding their neighbors to the same cluster as well. The expansion process continues until no new density-reachable samples can be added. After completion, the next unvisited core sample is selected, and the above expansion process is repeated to form a new cluster. Finally, all density-connected samples are assigned to their corresponding clusters, forming multiple state clusters.

[0044] S214. Perform statistical analysis on the samples in each state cluster to extract the frequency of occurrence and combination patterns of device states, and obtain the typical features of the cluster. Perform semantic interpretation based on the typical features of the cluster, assign scene semantic labels to each cluster, and obtain a set of life scene patterns.

[0045] Further, in step S214, the present invention performs statistical analysis on the samples in each state cluster. Specifically, in the statistical analysis, the present invention calculates the frequency of occurrence of each device state dimension in the cluster. The frequency of occurrence is obtained by statistically analyzing the percentage of times that dimension takes a specific value in the cluster samples. Subsequently, the present invention identifies the combination patterns of device states in the cluster. The combination pattern is the joint value of multiple device states that frequently appear in the cluster. The present invention then uses the frequency of occurrence and the combination pattern as typical features of the cluster. After obtaining the typical features of the cluster, the present invention infers the corresponding life scene type of the cluster by analyzing the device state combinations in the typical features, and performs semantic interpretation on the typical features of the cluster. For example, if the typical features of the cluster show that the living room lights are off, the TV is on, and the curtains are closed, then it is inferred that the cluster corresponds to a "movie-watching scene". Finally, after assigning a corresponding scene semantic label to each cluster, the present invention organizes all clusters and their scene semantic labels into a life scene pattern set. Each element in the life scene pattern set contains a scene semantic label and the typical features of the cluster corresponding to that scene.

[0046] S22. Extract the preconditions of the life scene pattern set through the association rule mining algorithm to obtain the scene trigger condition library.

[0047] like Figure 3 As shown, step S22 further includes: S221. Extract the time features, sensor status features, and device activity features of each scene from the set of life scene patterns, and obtain the scene pre-feature matrix through feature vectorization encoding.

[0048] Furthermore, in step S221, the present invention first traverses each scene in the set of life scene patterns. For each scene, the present invention searches for the time record of the scene occurrence from historical data. Then, the present invention extracts the data features within a 10-minute time window before the scene occurs. The extracted time features include time attributes such as the number of hours before the scene occurs, the day of the week, and whether it is a weekday. The sensor status features include the measurement values ​​of each sensor within the time window, such as temperature, humidity, and light intensity. The device activity features include the activation status and changes in operating parameters of each device within the time window.

[0049] After extraction, this invention performs vectorized encoding on the extracted time features, sensor status features, and device activity features, converting different types of features into a unified numerical vector format. One-hot encoding is used for categorical features, and normalization is performed on numerical features. Subsequently, this invention concatenates the encoded feature vectors according to feature type to form a scene pre-feature matrix. Rows in the scene pre-feature matrix correspond to different scene occurrence instances, and columns correspond to different pre-feature dimensions.

[0050] S222. Frequent itemset mining is performed on the scene pre-feature matrix using the Apriori algorithm to obtain the scene triggering rule set.

[0051] In step S222, the present invention inputs the scene pre-feature matrix into the Apriori algorithm for frequent itemset mining, aiming to find feature combinations that frequently occur before the scene occurs. Subsequently, the present invention uses the Apriori algorithm to identify feature combinations that meet the minimum support and minimum confidence requirements, and outputs a scene triggering rule set. The obtained scene triggering rule set contains multiple association rules, each rule being in the form of pre-feature combination → scene semantic label, indicating that when the pre-feature combination appears, the corresponding scene has a high probability of occurring.

[0052] Step S222 further includes: S2221. Set the minimum support threshold and minimum confidence threshold for the Apriori algorithm.

[0053] First, this invention sets the minimum support threshold of the Apriori algorithm to 0.3 and the minimum confidence threshold to 0.6. The minimum support threshold defines the minimum frequency requirement for a feature combination to appear in all scenario instances, while the minimum confidence threshold defines the minimum confidence requirement for an association rule.

[0054] S2222: Perform a single-item frequent set scan on the scene pre-feature matrix, count the occurrence frequency of each feature item, and filter to obtain single-item frequent sets with support greater than or equal to the minimum support threshold.

[0055] Furthermore, in step S2222, the present invention performs a frequent-item scan on the scene pre-feature matrix. During the scan, the present invention traverses each feature dimension of the matrix, counts the number of times the feature dimension takes a specific value in all scene instances, and calculates the support of the feature item. The support is equal to the number of times the feature item appears divided by the total number of scene instances. Subsequently, the present invention filters out feature items with a support greater than or equal to 0.3, and treats these feature items as frequent-item sets, where each element in the set is a single feature item.

[0056] S2223. Perform join and pruning operations on the single-item frequent set to generate candidate multi-item frequent sets, calculate the support of the candidate multi-item frequent sets, and filter to obtain multi-item frequent sets that are greater than or equal to the minimum support threshold.

[0057] In step S2223, the present invention performs a join operation on the single-item frequent sets, combining the feature terms of two single-item frequent sets to generate a candidate two-item set containing two feature terms. Then, the present invention performs a pruning operation on the candidate two-item set, removing candidate items whose subsets are not frequent sets. Subsequently, the present invention scans the scene's preceding feature matrix, counts the occurrence frequency of the candidate two-item set in the matrix, and calculates its support. After calculating the support, the present invention selects candidate two-item sets with a support greater than or equal to 0.3, forming a two-item frequent set. The present invention repeats the join and pruning operations, sequentially generating three-item frequent sets, four-item frequent sets, etc., until no longer frequent sets can be generated. Finally, all the frequent sets obtained constitute a multinomial frequent set.

[0058] S2224. Generate association rules based on the multiple frequent sets, calculate the confidence level of the association rules, and filter out association rules with a confidence level greater than or equal to the minimum confidence level threshold to obtain a scene triggering rule set.

[0059] In step S2224, the present invention traverses multiple frequent itemsets. For each frequent itemset, all possible association rules are generated. Specifically, the association rules are generated by dividing the frequent itemset into antecedent and consequent. The antecedent is a subset of the feature items, and the consequent is the remaining feature items. The present invention then calculates the confidence score of each association rule. The confidence score is equal to the number of times the antecedent and consequent occur simultaneously divided by the number of times the antecedent occurs. After calculation, the present invention selects association rules with a confidence score greater than or equal to 0.6, and uses these rules as a scenario trigger rule set. Each rule in the set contains four fields: antecedent, consequent, support, and confidence score.

[0060] As a preferred implementation, in order to further improve the predictive accuracy and practicality of the mined scenario triggering rules and avoid generating a large number of redundant rules that are statistically significant but have low predictive value, the present invention can further use a scenario triggering rule comprehensive utility evaluation model for fine screening after obtaining the generated association rules.

[0061] Specifically, the aforementioned scenario triggering rule comprehensive utility evaluation model selects each candidate rule R (in the form of: combination of prior features). →Scene semantic tags When considering the rule, a comprehensive utility score U(R) is introduced. This score not only considers the traditional statistical strength (confidence) of the rule, but also quantifies its time predictability and rule simplicity, thereby selecting the triggering rule with the most operational value.

[0062] The expression for the comprehensive utility score is as follows: ;in, The overall utility score for rule R is... Let R be the confidence level of rule R. As a moderating factor for time utility, As a complexity penalty factor, The average guidance time for rule R is the time from the combination of antecedent features. First complete appearance, in the scene The mean of the time intervals between actual occurrences. This represents the maximum average bootstrapping time of all rules in the candidate association rule set. Let the antecedent complexity of rule R be denoted as , i.e., the antecedent The number of features contained in it.

[0063] By introducing the comprehensive utility evaluation model, this invention can automatically identify high-quality rules from a large number of candidate rules that have high confidence, can provide early warnings before the appropriate time window, and have simple and clear triggering conditions. This significantly improves the accuracy of subsequent scenario prediction and the timeliness of system response, upgrading the quality of the scenario triggering condition library from statistical correlation to predictive usability, and enhancing the intelligence level of the entire smart home system's scenario perception and proactive services.

[0064] S223. Establish a classification index for the scene triggering rule set according to scene semantic tags, and associate each scene semantic tag with a corresponding triggering condition combination to obtain a scene triggering condition library.

[0065] In step S223, the present invention iterates through each rule in the scene triggering rule set and extracts the scene semantic tags of the rule consequents. Then, the present invention groups the rules according to the scene semantic tags, grouping rules with the same consequents together. After grouping, the present invention creates an index for each scene semantic tag, and the index is associated with the antecedents of all rules corresponding to that scene; the antecedents are the triggering condition combinations for that scene. Subsequently, the present invention stores the scene semantic tags and their associated triggering condition combinations as a scene triggering condition library. The final stored scene triggering condition library uses a hash table or dictionary structure, with the key being the scene semantic tag and the value being a list of triggering condition combinations for that scene.

[0066] S23. Input the global state vector, the scene semantic labels and time features in the set of life scene patterns into the LSTM neural network for sequence prediction processing to obtain the occurrence probability distribution.

[0067] In step S23, the present invention performs time alignment between the global state vector and the scene semantic labels in the set of life scene patterns. During the alignment process, the present invention labels the global state vector at each moment with the scene semantic label corresponding to that moment. Subsequently, the present invention extracts the current time features, which include information such as hour, day of the week, and whether it is a weekday. Following this, the present invention concatenates the global state vector, scene semantic labels, and time features to form an input feature sequence. Each moment in the obtained input feature sequence contains three parts of data: state vector, scene label, and time feature.

[0068] After forming the feature sequence, this invention inputs the feature sequence into an LSTM neural network. The LSTM neural network includes an input layer, an LSTM layer, and an output layer. The input layer receives the input feature sequence, the LSTM layer extracts the temporal dependencies of the sequence through its memory units and gating mechanism, and the output layer maps the hidden states of the LSTM layer into a probability distribution through a fully connected layer and a Softmax activation function. The final output probability distribution contains the probability value of each scene occurring in the future time period, forming an occurrence probability distribution.

[0069] S3. Determine the standard device state configuration corresponding to the target scenario in the triggering condition formation library based on the scenario occurrence probability distribution, perform multi-objective optimization on the standard device state control, and obtain a list of collaborative control strategies.

[0070] Step S3 further includes: S31. Filter the target scenario according to the occurrence probability distribution, and extract the standard device status configuration corresponding to the target scenario from the scenario triggering condition library.

[0071] Furthermore, in step S31, the present invention sorts the probability values ​​in the probability distribution from largest to smallest. The scenario with the highest probability value is placed first in the sorted sequence, and this first scenario is the target scenario. Simultaneously, the semantic tag of the target scenario is obtained. Subsequently, the present invention queries the scene triggering condition library for the record corresponding to the semantic tag of the scene, retrieving the standard device state configuration corresponding to the target scenario. The obtained standard device state configuration is a data structure containing the target state parameters that each device should achieve in this scenario, such as the brightness value of the lighting device, the temperature setting value of the air conditioning device, and the opening angle of the curtain device.

[0072] S32. The standard equipment state configuration is optimized using a particle swarm optimization algorithm to obtain the optimal combination of equipment control parameters.

[0073] In step S32, the present invention defines multiple objective functions, including a user comfort objective function, an energy efficiency objective function, a device response speed objective function, and a system safety objective function. Each objective function quantifies the performance of the control strategy in the corresponding dimension, and assigns a weight coefficient to each objective function. The weight coefficient reflects the importance of different objectives.

[0074] After setting the objective function, this invention uses a particle swarm optimization (PSO) algorithm to optimize the standard equipment state configuration. Specifically, the PSO algorithm first initializes a set of particles, with each particle's position representing a set of equipment control parameters. The algorithm then iteratively updates the particle positions and velocities. In each iteration, this invention calculates the multi-objective function value for each particle and obtains the comprehensive objective function value through weighted summation. Subsequently, this invention updates the individual optimal position and the global optimal position of each particle. The individual optimal position is the best position historically reached by that particle, and the global optimal position is the best position historically reached by all particles. After the update, this invention updates the particle's velocity and position based on the individual optimal position and the global optimal position. The iterative process continues until the maximum number of iterations is reached or the objective function converges. Finally, the parameter combination corresponding to the global optimal position is the optimal equipment control parameter combination.

[0075] In step S32, the expression for the objective function used to perform multi-objective optimization of the standard device state configuration is: ;in, Let be the overall objective function. The objective function is user comfort. These are the weighting coefficients corresponding to the user comfort objective function. Let the energy efficiency objective function be... These are the weighting coefficients corresponding to the energy efficiency objective function. Let the objective function be the device response speed. The weighting coefficient corresponding to the device response speed. The objective function for system security is... These are the weighting coefficients corresponding to the system security objective function; ;in, This is the comfort deviation weighting coefficient. For sensor index values, This represents the total number of comfort-related sensors. For the first Environmental state values ​​sensed by each sensor For the first User preference values ​​for each sensor, For the first The maximum tolerance value of each sensor. For the first Minimum tolerance value for each sensor The comfort balance adjustment coefficient is used. This is the variance calculation function. For comfort-related sensors controlled by parameters The state value under the given state; ;in, The weighting coefficient is the average energy consumption. For electrical equipment index value, The total number of electrical devices participating in the coordinated control. For the first The equipment in control parameters Real-time power under, For the first The continuous working time of the equipment. This is the maximum allowable load power of the household power grid. For scene time period, For the first The rated power of the equipment; ;in, For response speed weighting coefficient, For response delay adjustment coefficient, The average response latency for all devices. The maximum threshold for device response latency. This is the minimum threshold for device response latency; ;in, For safety weighting coefficients, For security-related device index values, The total number of safety-related devices, For the first Risk assessment value of Taiwan's safety equipment For the first The maximum risk value of Taiwan's safety equipment For the first Minimum risk value of Taiwan's safety equipment This is the risk accumulation adjustment coefficient.

[0076] Furthermore, the four weighting coefficients in the above formula all take values ​​in the range of (0, 1). In implementation... Take 0.4, Take 0.3, and Each is set to 0.15, and the comfort deviation weighting coefficient is set to the default value. To balance individual comfort deviations with group comfort equilibrium, the comfort equilibrium adjustment coefficient is taken as a positive number, and in implementation... This is used to amplify the impact of group comfort differences (variance) on the target. The greater the difference, the faster the exponential function decays. The energy consumption mean weighting coefficient is set to a value of [value missing]. , The power consumption is determined by the equipment's operating status. For example, when an air conditioner is set to 20℃, its power consumption is 1500W. The maximum allowable load power of the household power grid is determined by the capacity of the incoming circuit. This is to avoid the total energy consumption exceeding the grid's carrying capacity. The rated power of the equipment is an inherent parameter of the equipment, and the response speed weighting coefficient is [value missing]. The response delay adjustment coefficient takes a positive value and is implemented in practice. This is used to amplify the impact of response delay on the target. The exponential function decays rapidly as delay increases. The average response delay of all devices is determined by control parameters and device relationships. In practice, for example, the response delay of high-priority devices is 0.2s, medium-priority devices are 0.5s, and low-priority devices are 0.35s. The safety weighting coefficient is set to... Prioritize the safety of critical equipment. , , The safety device itself determines the outcome; for example, if the gas sensor shows no leakage... =0.1, during leakage Gas sensor =1, The value is 0, and the risk accumulation adjustment coefficient is a positive number during implementation. This is used to punish situations where multiple devices simultaneously pose security risks; the greater the accumulated risk value, the faster this exponential function decays.

[0077] S33. Based on the association weights in the device association graph, the optimal device control parameter combination is processed for execution timing allocation to obtain a list of collaborative control strategies.

[0078] In step S33, the present invention extracts all relevant device identifiers from the optimal combination of device control parameters. Then, the present invention queries the association weights between these devices in the device association graph. The association weights reflect the strength of the collaborative relationship between devices. After the query, the present invention prioritizes the devices according to their association weights; that is, device combinations with higher association weights should be executed first to ensure the collaborative effect between devices. After calculating the priority, the present invention assigns an execution sequence to each device. The execution sequence includes the device startup time and execution order. Devices with higher association weights are assigned earlier execution times, and devices with lower association weights are assigned later execution times. After allocation, the present invention encapsulates each device's device identifier, corresponding control command, and assigned execution sequence into a triple. The obtained control command contains the specific operation to be performed by the device and the target parameter value. Finally, all triples are sorted according to their execution sequence to form a collaborative control strategy list. The obtained collaborative control strategy list is an ordered list, with elements arranged according to their execution sequence, ensuring that devices start collaboratively in a reasonable order.

[0079] S4. Execute instructions in groups according to the list of collaborative control strategies based on device network partitions, and perform monitoring and compensation during execution to complete smart home collaborative control based on the Internet of Things.

[0080] Step S4 further includes: S41. Group the collaborative control policy list according to the network partition to which the device belongs to obtain the execution instruction set of multiple partitions.

[0081] In step S41, the present invention iterates through each triple in the collaborative control strategy list, extracts the device identifier field from the triple, and then queries the network partition affiliation information of the device in the device topology network based on the device identifier. The network partition affiliation information identifies the physical network area to which the device belongs. After obtaining the affiliation information, the present invention groups the triples according to the network partition. During the grouping process, the present invention creates a grouping mapping with the network partition as the key, aggregating triples belonging to the same network partition into the same group. After the grouping is completed, multiple execution instruction sets for partitions are formed. Each execution instruction set contains the control instructions and execution timing of all devices in that partition.

[0082] S42. Send execution instruction sets to the device according to the execution sequence through the edge control node.

[0083] In step S42, the present invention distributes the execution instruction sets of each partition to the corresponding edge control nodes. During the distribution process, the present invention sends the execution instruction sets to the corresponding nodes according to the mapping relationship between network partitions and edge control nodes. Subsequently, the edge control nodes receive the execution instruction sets and perform instruction scheduling according to the execution timing information in the instruction sets. During the scheduling process, the edge control nodes send control instructions to the corresponding target devices when each execution time arrives. Finally, the control instructions are transmitted to the devices through network communication protocols, and the devices execute the corresponding control operations after receiving the instructions.

[0084] S43. Synchronously monitor changes in the global state vector. When the deviation between the actual state and the expected state exceeds a preset threshold, compensate and adjust the execution parameters to obtain the execution log.

[0085] In step S43, the present invention continuously collects real-time status information of the devices through sensor nodes. The collection process is carried out at fixed time intervals, and the status value of the corresponding device in the global status vector is updated each time. Subsequently, the present invention compares the updated global status vector with the expected status in the collaborative control strategy list. During the comparison process, the present invention calculates the difference between the actual status value and the expected status value for each device to obtain the deviation value. Subsequently, the present invention compares the deviation value with a preset threshold. When the deviation value exceeds the preset threshold, the compensation adjustment process is triggered.

[0086] In the compensation and adjustment process, this invention first performs a cause analysis on the deviation data. By checking the equipment response time, environmental interference factors, and equipment operating status, the type of cause leading to the deviation is diagnosed. The identified cause types include equipment response delay, changes in ambient temperature, and equipment malfunction. Subsequently, this invention retrieves the corresponding compensation scheme from the compensation strategy library based on the cause type. The compensation scheme specifies the adjustment measures to be taken for specific causes, such as increasing equipment power, extending operating time, or activating backup equipment. Following this, this invention adjusts the execution parameters according to the compensation scheme. The adjustments include modifying equipment control parameters and updating the execution timing.

[0087] In the compensation process, this invention also records the deviation value, cause type, compensation plan, and user manual intervention behavior. The recorded content includes fields such as timestamp, device identifier, deviation value, cause type, compensation plan, and intervention behavior. The obtained records are organized in chronological order to form an execution log.

[0088] This invention fundamentally improves the intelligence level, user experience, and energy management efficiency of home systems by constructing a smart home decision-making framework that deeply integrates perception, analysis, and optimization control.

[0089] First, this invention achieves a fundamental shift from discrete device control to global scenario-based collaboration. By automatically mining the dynamic correlation graph between devices and user life scenario patterns, this invention can accurately understand the multidimensional state of the living environment and user behavioral intentions, thereby eliminating the dependence on preset fixed rules or manual grouping. This enables intelligent control to have self-learning and self-adaptive capabilities, solving the core pain point of existing systems' inability to handle complex and dynamic living scenarios.

[0090] Secondly, this invention, through an innovative multi-objective comprehensive optimization function and a prediction-optimization closed-loop mechanism, achieves optimal balance of system performance across multiple dimensions while ensuring user comfort and safety. It not only responds to the current state in real time but also proactively formulates equipment control strategies based on predicted future scenario probabilities, thereby significantly improving energy efficiency, equipment response speed, and system stability. This proactive and preventative management approach effectively avoids the resource waste and fragmented user experience associated with traditional reactive control.

[0091] This invention provides a complete and scalable technical system that closely integrates data-driven artificial intelligence algorithms with specific home control engineering practices. It not only brings users a more considerate, smooth, and energy-efficient smart living experience, but also provides a highly feasible and innovative systematic solution for smart homes, with significant practical value and promising prospects for promotion.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart home collaborative control method based on the Internet of Things, characterized in that, include: S1. Analyze historical state data of the smart home environment through machine learning to obtain a device association map, and encode the collected real-time state data to obtain a global state vector; S2. Combine the global state vector with the device association graph to obtain a set of life scene patterns by clustering and extract the trigger conditions to form a library. Input the global state vector into the neural network to predict the probability distribution of scene occurrence. S3. Determine the standard device state configuration corresponding to the target scenario in the triggering condition formation library according to the scenario occurrence probability distribution, perform multi-objective optimization on the standard device state control, and obtain a list of collaborative control strategies. S4. Execute instructions in groups according to the list of collaborative control strategies based on device network partitions, and perform monitoring and compensation during execution to complete smart home collaborative control based on the Internet of Things.

2. The smart home collaborative control method based on the Internet of Things according to claim 1, characterized in that, Step S1 further includes: S11. Assign device identifiers and spatial location labels to the sensor nodes in the smart home environment to obtain a set of sensor nodes; S12. Extract historical status data from the sensor node set, and perform inter-device correlation analysis on the historical status data using machine learning algorithms to obtain a device correlation map; S13. Collect real-time status information of all devices in the sensor node set at fixed time intervals, and convert the status information into a numerical vector format through vectorization encoding to obtain a global status vector.

3. The smart home collaborative control method based on the Internet of Things according to claim 2, characterized in that, Step S12 further includes: S121. Extract historical status data of multiple devices from the sensor node set, sort the historical status data by timestamp, and obtain the device activation time sequence. S122. Perform co-occurrence frequency statistics on the device activation time series through a sliding time window, calculate the number of times any two devices are activated simultaneously within the same time window, and obtain the co-occurrence matrix; S123. Apply correlation coefficient calculation to the co-occurrence matrix to quantify the correlation strength between devices and obtain the correlation strength matrix; S124. Construct a weighted undirected graph based on the association strength matrix, with device identifiers as nodes and association strength values ​​as edge weights, to obtain the device association graph.

4. The smart home collaborative control method based on the Internet of Things according to claim 1, characterized in that, Step S2 further includes: S21. The global state vector is segmented according to a time window, and the segmented state sequence is clustered by the DBSCAN clustering algorithm in combination with the association weight in the device association map to obtain a set of life scene patterns. S22. Extract the preconditions of the life scene pattern set through the association rule mining algorithm to obtain the scene trigger condition library; S23. Input the global state vector, the scene semantic labels and time features in the set of life scene patterns into the LSTM neural network for sequence prediction processing to obtain the occurrence probability distribution.

5. The smart home collaborative control method based on the Internet of Things according to claim 4, characterized in that, Step S21 further includes: S211. The global state vector is sliced ​​according to a preset time window to obtain a state vector sample sequence; S212. Extract the association weight between any two devices from the device association map, construct the distance metric function of the state vector sample sequence through the weighted Euclidean distance formula, and obtain the weighted distance matrix between the samples. S213. Density clustering of the state vector sample sequence is performed using the DBSCAN clustering algorithm in conjunction with the weighted distance matrix to obtain multiple state clusters. S214. Perform statistical analysis on the samples in each state cluster to extract the frequency of occurrence and combination patterns of device states, and obtain the typical features of the cluster. Perform semantic interpretation based on the typical features of the cluster, assign scene semantic labels to each cluster, and obtain a set of life scene patterns.

6. The smart home collaborative control method based on the Internet of Things according to claim 4, characterized in that, Step S22 further includes: S221. Extract the time features, sensor status features, and device activity features of each scene from the set of life scene patterns, and obtain the scene pre-feature matrix through feature vectorization encoding. S222. Frequent itemset mining is performed on the scene pre-feature matrix using the Apriori algorithm to obtain the scene triggering rule set; S223. Establish a classification index for the scene triggering rule set according to scene semantic tags, and associate each scene semantic tag with a corresponding triggering condition combination to obtain a scene triggering condition library.

7. The smart home collaborative control method based on the Internet of Things according to claim 6, characterized in that, Step S222 further includes: S2221. Set the minimum support threshold and minimum confidence threshold for the Apriori algorithm; S2222: Perform a single-item frequent set scan on the scene pre-feature matrix, count the occurrence frequency of each feature item, and filter to obtain single-item frequent sets with support greater than or equal to the minimum support threshold. S2223. Perform join and pruning operations on the single-item frequent set to generate candidate multi-item frequent sets, calculate the support of the candidate multi-item frequent sets, and filter to obtain multi-item frequent sets that are greater than or equal to the minimum support threshold. S2224. Generate association rules based on the multiple frequent sets, calculate the confidence level of the association rules, and filter out association rules with a confidence level greater than or equal to the minimum confidence level threshold to obtain a scene triggering rule set.

8. The smart home collaborative control method based on the Internet of Things according to claim 1, characterized in that, Step S3 further includes: S31. Filter out target scenarios according to the occurrence probability distribution, and extract the standard device status configuration corresponding to the target scenario from the scenario triggering condition library; S32. The optimal combination of equipment control parameters is obtained by performing multi-objective optimization on the standard equipment state configuration using the particle swarm optimization algorithm. S33. Based on the association weights in the device association graph, the optimal device control parameter combination is processed for execution timing allocation to obtain a list of collaborative control strategies.

9. A smart home collaborative control method based on the Internet of Things according to claim 8, characterized in that, In step S32, the expression for the objective function used to perform multi-objective optimization of the standard device state configuration is: ;in, Let be the overall objective function. The objective function is user comfort. These are the weighting coefficients corresponding to the user comfort objective function. Let the energy efficiency objective function be... These are the weighting coefficients corresponding to the energy efficiency objective function. Let the objective function be the device response speed. The weighting coefficient corresponding to the device response speed. The objective function for system security is... These are the weighting coefficients corresponding to the system security objective function; ;in, This is the comfort deviation weighting coefficient. For sensor index values, This represents the total number of comfort-related sensors. For the first Environmental state values ​​sensed by each sensor For the first User preference values ​​for each sensor, For the first The maximum tolerance value of each sensor. For the first Minimum tolerance value for each sensor The comfort balance adjustment coefficient is used. This is the variance calculation function. For comfort-related sensors controlled by parameters The state value under the given state; ;in, The weighting coefficient is the average energy consumption. For electrical equipment index value, The total number of electrical devices participating in the coordinated control. For the first The equipment in control parameters Real-time power under, For the first The continuous working time of the equipment. This is the maximum allowable load power of the household power grid. For scene time period, For the first The rated power of the equipment; ;in, For response speed weighting coefficient, For response delay adjustment coefficient, The average response latency for all devices. The maximum threshold for device response latency. This is the minimum threshold for device response latency; ;in, For security weighting coefficients, For security-related device index values, The total number of safety-related devices, For the first Risk assessment value of Taiwan's safety equipment For the first The maximum risk value of Taiwan's safety equipment For the first Minimum risk value of Taiwan's safety equipment This is the risk accumulation adjustment coefficient.

10. The smart home collaborative control method based on the Internet of Things according to claim 1, characterized in that, Step S4 further includes: S41. Group the collaborative control policy list according to the network partition to which the device belongs to obtain the execution instruction set of multiple partitions; S42. Send execution instruction sets to the device according to the execution sequence through the edge control node; S43. Synchronously monitor changes in the global state vector. When the deviation between the actual state and the expected state exceeds a preset threshold, compensate and adjust the execution parameters to obtain the execution log.

Images