A whole-house intelligent switch linkage control system and method based on scene self-learning

Abstract

The application provides a whole-house intelligent switch linkage control system and method based on scene self-learning, comprising: identifying a candidate scene and generating a linkage control instruction according to collected event signals and trigger condition weights and hysteresis time lengths of local scene labels; monitoring whether a user operation signal opposite to the change direction of the on-off state of any controlled device is received within a preset time window after executing the linkage control instruction; if received, generating a rollback instruction to restore the previous state, reducing the trigger condition weight of the candidate scene, increasing the hysteresis time length, and clearing the positive feedback counter; if not received, maintaining the linkage operation and increasing the positive feedback counter by 1; when reaching a preset threshold, increasing the trigger condition weight of the candidate scene, shortening the hysteresis time length, and clearing the positive feedback counter. The application can realize feedback adaptive scene linkage control driven by multi-source event signal uncertainty reasoning, thereby improving the whole-house intelligent switch linkage trigger accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of smart home and Internet of Things control, and more specifically, the present application relates to a whole-house smart switch linkage control system and method based on scenario self-learning. Background Art

[0002] With the rapid development of Internet of Things, edge computing and intelligent sensing technologies, the whole-house smart system collects user behavior and environmental state event signals by deploying various devices such as smart switch panels, human presence sensors, light sensors and door / window magnetic sensors, and realizes multi-device linkage control based on preset scenario rules. As a local computing unit close to the data source end, edge computing nodes can reduce cloud transmission delay and protect user privacy, and have gradually become the core execution carrier for whole-house smart linkage control. Multi-source event signals have various types and different reporting frequencies, which put forward higher requirements for real-time processing and linkage decision-making of edge computing nodes.

[0003] In the prior art, existing whole-house smart switch linkage control methods usually regard various event signals as deterministic inputs, and perform scenario recognition and device linkage based on preset rules or simple threshold matching, without fully considering the data quality uncertainty existing in the collection, transmission and fusion processes of multi-source event signals. For example, human presence sensors are prone to instantaneous false alarms caused by heat source disturbance, door / window magnetic signals may be delayed or lost due to network jitter, and there may also be semantic contradiction or timing conflict when multiple event signals are concurrent within the same time window. The above uncertainty factors lead to a large deviation between the scenario recognition result and the actual user intention, false triggering and missed triggering occur frequently, and users have to manually correct the device status repeatedly. In addition, parameters such as scenario trigger condition weights and hysteresis duration in existing methods lack effective feedback self-learning adjustment, so that such false triggering problems cannot be suppressed with the user's use process, and the automation experience continues to deteriorate. Therefore, how to realize feedback adaptive scenario linkage control driven by uncertainty reasoning of multi-source event signals, so as to improve the triggering accuracy of whole-house smart switch linkage has become a difficult problem faced by the industry. Summary of the Invention

[0004] The present application provides a whole-house smart switch linkage control system and method based on scenario self-learning, which can realize feedback adaptive scenario linkage control driven by uncertainty reasoning of multi-source event signals, thereby improving the triggering accuracy of whole-house smart switch linkage.

[0005] In a first aspect, the present application provides a whole-house smart switch linkage control method based on scenario self-learning, which is executed by an edge computing node deployed in a whole-house smart system, and includes the following steps: identifying candidate scenarios and generating linkage control instructions according to the collected event signals and the trigger condition weights and hysteresis durations of each scenario tag maintained locally by the edge computing node; After executing the linkage control command, monitor whether a user operation signal is received within a preset feedback time window that is opposite to the direction of the on / off state change of any controlled device in the linkage control command. If received, a rollback command is generated to restore the controlled device to its state before execution, reduce the trigger condition weight of the candidate scenario, increase its hysteresis duration, and clear the positive feedback counter. If no response is received, the linkage operation is maintained and the positive feedback counter is incremented by 1. When the positive feedback counter reaches a preset threshold, the trigger condition weight of the candidate scenario is increased, its hysteresis duration is shortened, and the positive feedback counter is cleared to zero.

[0006] In some embodiments, the event signal includes at least one of the following: smart switch action signal, human presence signal, light signal, and door / window magnetic signal.

[0007] In some embodiments, candidate scenes are identified and linkage control commands are generated based on the collected event signals and the trigger condition weights and hysteresis durations of each scene label locally maintained by the edge computing node. Specifically, this includes: The event signals collected within the preset time window are analyzed to obtain the event characteristics of each event signal; Based on the event characteristics and timestamp information of each event signal, a scene fingerprint is generated, and the confidence level of the scene fingerprint is obtained by fusing the data quality characteristics of each event signal. The scene fingerprint and its confidence level are input into a preset uncertainty inference model, and fused with the reference scene fingerprints of each scene tag in the scene tag library maintained locally by the edge computing node to output the scene trigger score and its evaluation confidence level for each scene tag; wherein, the uncertainty inference model is used to handle the data quality uncertainty that exists in the process of event signal acquisition, transmission and fusion. Candidate scene labels are determined based on the scene trigger scores and their evaluation confidence levels for each scene label, combined with the trigger condition weights corresponding to each scene label. Based on the candidate scene tags, obtain the corresponding device linkage strategy from the scene tag library; When the scenario trigger score of the candidate scenario tag is continuously greater than the preset score threshold within the hysteresis duration, and its evaluation confidence is continuously greater than the preset confidence threshold within the hysteresis duration, a linkage control instruction is generated according to the device linkage strategy. The linkage control instruction includes the identifier of the controlled device to be controlled and its target on / off status.

[0008] In some embodiments, a scene fingerprint is generated based on the event characteristics and timestamp information of each event signal, specifically including: The preset time window is divided into multiple sub-time windows, and based on the timestamp information of each event signal, event signals falling within the same sub-time window are identified as co-occurring event signals. The event features of the co-occurrence event signals are extracted to generate a scene fingerprint that characterizes the co-occurrence relationship of device type, operation action and spatial location.

[0009] In some embodiments, candidate scene labels are determined based on the scene trigger score and its evaluation confidence level for each scene label, combined with the trigger condition weights corresponding to each scene label. Specifically, this includes: The scene trigger score of each scene tag is multiplied by the corresponding trigger condition weight to obtain the weighted trigger score of each scene tag. Remove scenario labels whose evaluation confidence level is lower than the preset confidence threshold; From the remaining scene tags, select the scene tag with the highest weighted trigger score as the candidate scene tag; If the remaining scene labels are empty, no candidate scene labels will be generated.

[0010] In some embodiments, the user operation signal includes at least one of a smart switch panel operation signal, a mobile terminal control signal, and a voice control signal.

[0011] In some embodiments, the rollback instruction is only used to restore the controlled device that has been reversed by the user to its state before execution, while other controlled devices remain in their linked state.

[0012] In some embodiments, if received, a rollback instruction is generated to restore the controlled device to its pre-execution state, reduce the trigger condition weight of the candidate scenario, increase its hysteresis duration, and reset the positive feedback counter, specifically including: Analyze the reverse user operation signals to determine the controlled device being operated on and the associated candidate scenarios; Generate a rollback command for the controlled device to restore the controlled device to the state before the linkage control command was executed; The trigger condition weight of the candidate scenario is reduced by a first preset step size, but not lower than the preset weight lower limit; the delay duration of the candidate scenario is increased by a second preset step size, but not higher than the preset delay upper limit. Then, the positive feedback counter for the candidate scenario is reset to zero.

[0013] In some embodiments, when the positive feedback counter reaches a preset threshold, the trigger condition weight of the candidate scenario is increased, its hysteresis duration is shortened, and the positive feedback counter is reset to zero, specifically including: In response to the positive feedback counter reaching a preset threshold, the trigger condition weight of the candidate scenario is increased by a third preset step size, but not higher than the preset weight upper limit. The hysteresis duration of the candidate scenario is shortened by the fourth preset step, and is not lower than the preset hysteresis lower limit; The positive feedback counter is reset to zero to re-accumulate the number of positive feedback events.

[0014] Secondly, this application provides a whole-house smart switch linkage control system based on scene self-learning, the system comprising: The scene recognition module is used to identify candidate scenes and generate linkage control commands based on the collected event signals and the trigger condition weights and hysteresis durations of each scene label maintained locally by the edge computing node. The processing module is used to monitor, within a preset feedback time window, whether a user operation signal is received that is opposite to the direction of the on / off state change of any controlled device in the linkage control command after the linkage control command is executed. The execution module is used to generate a rollback instruction if received, restore the controlled device to the state before execution, reduce the trigger condition weight of the candidate scenario, increase its delay time, and clear the positive feedback counter. The execution module is also used to maintain the linkage operation and increment the positive feedback counter by 1 if no response is received; when the positive feedback counter reaches a preset threshold, increase the trigger condition weight of the candidate scenario, shorten its hysteresis duration, and clear the positive feedback counter.

[0015] Thirdly, this application provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described whole-house smart switch linkage control method based on scene self-learning.

[0016] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-mentioned whole-house smart switch linkage control method based on scene self-learning.

[0017] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application improves the accuracy of scene linkage triggering for whole-house smart switches under multi-source event signals by constructing a closed-loop linkage control mechanism that integrates scene fingerprint confidence representation, uncertainty inference fusion, delayed confirmation, and user feedback adaptation. First, it generates scene fingerprint confidence by fusing data quality scores from various event signals, explicitly representing the data quality uncertainty during the acquisition, transmission, and fusion processes, thus providing a robust foundation for subsequent scene recognition. Second, an uncertainty inference model, composed of a feature matching sub-model, a data quality assessment sub-model, and a multi-source event fusion sub-model, outputs a scene trigger score and assessment confidence. When evidence conflicts are detected between different event signals, the matching correction weights are adjusted based on the degree of conflict, thereby suppressing interference from low-quality or abnormal event signals on the scene recognition results. Finally, it eliminates signals with insufficient assessment confidence by combining trigger condition weights. The system identifies scene tags and continuously confirms that the scene trigger scores and evaluation confidence levels of candidate scene tags meet preset thresholds before generating linkage control commands. This effectively avoids false triggers caused by instantaneous signal fluctuations or low-confidence matching. Finally, after executing the linkage control command, the system adjusts the trigger condition weights and lag durations of candidate scene tags punitively or rewardingly based on whether a reverse user operation signal appears within a preset feedback time window. A positive feedback counter is used to control the adjustment rhythm of parameters, enabling the system to dynamically optimize trigger sensitivity based on actual user feedback, gradually suppressing repeated false triggers and improving response speed. In summary, this application achieves adaptive scene linkage control through the synergistic effect of multi-source event signal uncertainty reasoning and feedback self-learning, significantly improving the accuracy and intelligence level of whole-house smart switch linkage triggering. Attached Figure Description

[0018] Figure 1 This is an exemplary flowchart of a whole-house smart switch linkage control method based on scene self-learning, according to some embodiments of this application; Figure 2 This is a schematic diagram of the process for generating scene fingerprints according to some embodiments of this application; Figure 3 This is a schematic diagram of the structure of a whole-house smart switch linkage control system based on scene self-learning, according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a computer device that implements a scene-based self-learning method for the linkage control of smart switches throughout the house, according to some embodiments of this application. Detailed Implementation

[0019] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] refer to Figure 1The figure is an exemplary flowchart of a whole-house smart switch linkage control method based on scene self-learning, according to some embodiments of this application. This whole-house smart switch linkage control method based on scene self-learning is executed by edge computing nodes deployed in the whole-house smart system, and mainly includes the following steps: In step 101, candidate scenarios are identified and linkage control commands are generated based on the collected event signals and the trigger condition weights and hysteresis durations of each scene label locally maintained by the edge computing node.

[0021] It should be noted that the event signals described in this application include at least one of the following: smart switch action signals, human presence signals, light signals, and door / window magnetic signals. In some embodiments, event signals can be collected in the following manner: the edge computing node connects to the smart switch panel, human presence sensor, light sensor, and door / window magnetic sensor in the whole-house smart system via communication methods such as Zigbee, Wi-Fi, or RS485; when any device detects a change in status, it actively reports the event signal to the edge computing node, or the edge computing node requests status data from each device according to a preset polling cycle. Each event signal carries at least device identifier, device type, operation action, spatial location, and timestamp information, wherein the operation action includes at least one of the following: switch on / off, change in human presence status, change in light intensity, and opening / closing of doors / windows. While receiving event signals, edge computing nodes determine the data quality characteristics of the event signals based on parameters such as signal strength, reception latency, and the number of packet loss retransmissions. For example, any situation where the signal strength is lower than a preset strength threshold, the reception latency exceeds a preset latency threshold, or the number of packet loss retransmissions exceeds a preset number threshold is judged as low data quality. The edge computing nodes determine signal strength sub-scores, reception latency sub-scores, and packet loss retransmission sub-scores based on the deviation between the signal strength, reception latency, and number of packet loss retransmissions and the corresponding preset strength threshold, preset latency threshold, and preset number threshold, respectively. Among these, signal strength... The sub-score is determined based on the ratio of signal strength to a preset strength threshold; the higher the ratio, the higher the sub-score, but not exceeding 1. The reception delay sub-score is determined based on the proportion of reception delay exceeding a preset delay threshold; the larger the excess, the lower the sub-score, but not lower than 0. The packet loss retransmission sub-score is determined based on the proportion of packet loss retransmissions exceeding a preset number of retransmissions; the larger the excess, the lower the sub-score, but not lower than 0. The edge computing node performs a weighted average of the signal strength sub-score, reception delay sub-score, and packet loss retransmission sub-score, and uses the weighted average as the data quality score for the event signal. The data quality score ranges from 0 to 1. The weights of the signal strength sub-score, reception delay sub-score, and packet loss retransmission sub-score can be preset according to the degree of influence of each parameter on data quality, for example, set to 0.4, 0.3, and 0.3 respectively. The preset strength threshold, preset delay threshold, and preset number of retransmission threshold can be preset according to the communication method, device type, and historical data statistical characteristics of each device, or dynamically determined based on the average signal strength, average reception delay, and average number of packet loss retransmissions of similar event signals received by the edge computing node within a preset time period.

[0022] In some embodiments, candidate scenes are identified and linkage control commands are generated based on the collected event signals and the trigger condition weights and hysteresis durations of each scene label locally maintained by the edge computing node. This can be achieved through the following steps: The event signals collected within the preset time window are analyzed to obtain the event characteristics of each event signal; Based on the event characteristics and timestamp information of each event signal, a scene fingerprint is generated, and the confidence level of the scene fingerprint is obtained by fusing the data quality characteristics of each event signal. The scene fingerprint and its confidence level are input into a preset uncertainty inference model, and fused with the reference scene fingerprints of each scene tag in the scene tag library maintained locally by the edge computing node to output the scene trigger score and its evaluation confidence level for each scene tag; wherein, the uncertainty inference model is used to handle the data quality uncertainty that exists in the process of event signal acquisition, transmission and fusion. Candidate scene labels are determined based on the scene trigger scores and their evaluation confidence levels for each scene label, combined with the trigger condition weights corresponding to each scene label. Based on the candidate scene tags, obtain the corresponding device linkage strategy from the scene tag library; When the scenario trigger score of the candidate scenario tag is continuously greater than the preset score threshold within the hysteresis duration, and its evaluation confidence is continuously greater than the preset confidence threshold within the hysteresis duration, a linkage control instruction is generated according to the device linkage strategy. The linkage control instruction includes the identifier of the controlled device to be controlled and its target on / off status.

[0023] It should be noted that the scene fingerprint in this application refers to a data structure used to characterize the temporal co-occurrence relationship of multi-source event signals in terms of device type, operation action and spatial location within a preset time window; the uncertainty reasoning model refers to a reasoning model used to handle the data quality uncertainty in the process of event signal acquisition, transmission and fusion, and to output the scene trigger score and its evaluation confidence of each scene label; the device linkage strategy refers to a set of rules used to specify the controlled device identifier, target on / off status and execution order corresponding to the scene label.

[0024] In practice, the event signals collected within the preset time window are parsed to obtain the event characteristics of each event signal. This can be achieved as follows: the edge computing node reads the event signals falling within the preset time window from its local cache, extracts the device identifier, device type, operation action, spatial location, and timestamp fields according to the message format of the event signal, and uses the device type, operation action, and spatial location as the event characteristics of the event signal. The length of the preset time window can be pre-set based on the event reporting frequency of each device in the smart home system or the average time interval between adjacent event signals, or it can be configured by the user, for example, fixed at 30 seconds or 1 minute. Edge computing nodes locally maintain a table of registered device information. This table contains at least the correspondence between device identifier, device type, and spatial location. If an event signal lacks a device type or spatial location field, the edge computing node queries the table based on the device identifier and completes the corresponding device type and spatial location. The completed device type, operation action, and spatial location are then used as the event characteristics of the event signal. If an event signal lacks an operation action field or the operation action field is empty, the edge computing node determines a default operation action based on the device type. For example, a smart switch operation signal corresponds to an on / off switching action, a human presence signal corresponds to a change in whether someone is present or absent, a light signal corresponds to a change in light intensity, and a door / window magnetic signal corresponds to an opening or closing action. The default operation action is then used as the operation action for the event signal.

[0025] In some embodiments, reference is made to Figure 2 As shown in the figure, this is a schematic flowchart of generating scene fingerprints according to some embodiments of this application. In this embodiment, the generation of scene fingerprints based on the event characteristics and timestamp information of each event signal can be achieved by the following steps: In step 1011, the preset time window is divided into multiple sub-time windows, and event signals falling within the same sub-time window are identified as co-occurring event signals based on the timestamp information of each event signal. In step 1012, the event features of the co-occurrence event signals are extracted to generate a scene fingerprint that characterizes the co-occurrence relationship of device type, operation action and spatial location.

[0026] The sub-time window in this application is a time interval used to divide a preset time window into multiple consecutive time segments to determine the co-occurrence relationship of event signals; the co-occurring event signal is a set of multiple event signals that occur within the same sub-time window and have a temporal co-occurrence relationship.

[0027] In practice, the following steps are taken: First, a preset time window and a preset sub-time window length are obtained. The sub-time window length is determined based on the average time interval between event signal reports from devices in the whole-house smart system, or it can be configured by the user, for example, setting the sub-time window length to 2 seconds. Starting from the beginning of the preset time window, the preset time window is continuously divided into multiple sub-time windows according to the sub-time window length. Each sub-time window is a left-closed, right-open time interval, and event signals with timestamps equal to the start time of the sub-time window are included in that sub-time window. For each sub-time window, the timestamp information of each event signal is read, and all event signals whose timestamps fall within that sub-time window are identified as co-occurring event signals corresponding to that sub-time window; if there are no event signals within a sub-time window, no co-occurring event signal is generated. Next, the event features of each event signal in each co-occurring event signal set are obtained. These event features include device type, operation action, and spatial location. The device type, operation action, and spatial location in the same co-occurring event signal set are arranged according to the chronological order of the event signal's timestamp, forming a device type sequence, an operation action sequence, and a spatial location sequence. These sequences are then used together as a scene fingerprint representing the temporal co-occurrence relationship of the co-occurring event signals across the three dimensions of device type, operation action, and spatial location. When multiple sub-time windows exist within a preset time window and generate scene fingerprints separately, the scene fingerprints corresponding to each sub-time window are concatenated according to their chronological order. The concatenated result is used as the scene fingerprint corresponding to that preset time window. If only one sub-time window generates a scene fingerprint within the preset time window, that scene fingerprint is directly used as the scene fingerprint corresponding to that preset time window.

[0028] In specific implementation, the confidence level of the scene fingerprint obtained by fusing the data quality features of each event signal can be achieved in the following way: After generating the scene fingerprint, the edge computing node obtains the set of co-occurring event signals corresponding to the scene fingerprint and the data quality features of each event signal in the set, where the data quality features are data quality scores, and the higher the data quality score, the higher the data quality of the event signal. Then, the arithmetic mean of the data quality scores of all event signals in the set of co-occurring event signals is calculated, and the calculated arithmetic mean is used as the confidence level of the scene fingerprint. For example, when the set of co-occurring event signals contains three event signals, and the data quality scores of the three event signals are 0.9, 0.8 and 0.7 respectively, the calculated arithmetic mean is 0.8, and 0.8 is used as the confidence level of the scene fingerprint. If the scene fingerprint is formed by splicing scene fingerprints corresponding to multiple sub-time windows, the edge computing node obtains the set of co-occurring event signals corresponding to each sub-time window before splicing, calculates the arithmetic mean of the data quality scores of all co-occurring event signals, and uses the calculated arithmetic mean as the confidence level of the spliced ​​scene fingerprint.

[0029] In some embodiments, the scene fingerprint and its confidence level are input into a preset uncertainty inference model, and fused with the reference scene fingerprints of each scene tag in the scene tag library maintained locally by the edge computing node for inference, outputting the scene trigger score and its evaluation confidence level for each scene tag. This can be achieved through the following steps: The scene fingerprint and its confidence level, and the reference scene fingerprint of each scene label are input into the uncertainty reasoning model; wherein, the uncertainty reasoning model includes a feature matching sub-model, a data quality assessment sub-model, and a multi-source event fusion sub-model; The feature matching sub-model calculates the matching degree between the scene fingerprint and each reference scene fingerprint, and uses it as the matching result; The data quality assessment sub-model determines the matching correction weights based on the confidence level of the scene fingerprint and the data quality characteristics of each event signal; The multi-source event fusion sub-model performs weighted fusion of the matching results based on the matching correction weight, and outputs the scene trigger score and its evaluation confidence for each scene label; When different event signals conflict in scene determination, the multi-source event fusion sub-model adjusts the matching correction weight according to the matching correction weight and the degree of conflict to suppress the interference of abnormal event signals on the reasoning results; the scene trigger score is a value used to characterize the quantitative degree to which each scene label is triggered by the current scene fingerprint; the evaluation confidence is an indicator to measure the credibility of the scene trigger score.

[0030] The reference scene fingerprint in this application is a preset scene feature used to match the current scene fingerprint; the feature matching sub-model is a sub-model used to calculate the matching degree between the scene fingerprint and each reference scene fingerprint; the data quality assessment sub-model is a sub-model used to determine the matching correction weight based on the confidence of the scene fingerprint and the data quality features of each event signal; the multi-source event fusion sub-model is a sub-model used to perform weighted fusion of the matching results based on the matching correction weight and output the scene trigger score and assessment confidence.

[0031] In specific implementation, firstly, the edge computing node reads the scene fingerprint and its confidence level corresponding to the current preset time window from the local cache, and reads the reference scene fingerprint corresponding to each scene tag one by one from the locally maintained scene tag library. The scene fingerprint, the confidence level of the scene fingerprint, and the reference scene fingerprint of each scene tag are then input into a preset uncertainty inference model. This uncertainty inference model is composed of a feature matching sub-model, a data quality assessment sub-model, and a multi-source event fusion sub-model connected sequentially. The feature matching sub-model can use either an overlap calculation model or a cosine similarity calculation model. Taking the overlap calculation model as an example, the feature matching sub-model compares the device type sequence, operation action sequence, and spatial location sequence in the scene fingerprint with the device type sequence, operation action sequence, and spatial location sequence in each reference scene fingerprint, position by position. When the lengths of the two sequences are inconsistent, the shorter sequence length is used as the comparison length. Elements exceeding the shorter sequence length are not included in the comparison. The number of identical elements is counted, and the number of identical elements is divided by the total number of elements participating in the comparison. The resulting ratio is used as the matching degree corresponding to the scene tag. The calculated matching degree is then used as the matching result. Secondly, the data quality assessment sub-model extracts the data quality features of each event signal from the co-occurring event signals corresponding to the current scene fingerprint. These data quality features are data quality scores. The data quality assessment sub-model can use a weighted average model to calculate the arithmetic mean of the data quality scores of all co-occurring event signals corresponding to the current scene fingerprint, which is taken as the event data quality mean. The confidence level of the scene fingerprint and the event data quality mean are weighted and summed, where the weight of the confidence level of the scene fingerprint can be set to 0.6 and the weight of the event data quality mean can be set to 0.4. The value obtained by the weighted sum is then used as the matching correction weight corresponding to the scene label. If the value obtained by the weighted sum is greater than 1, the matching correction weight is set to 1; if the value obtained by the weighted sum is less than 0, the matching correction weight is set to 0.Then, the multi-source event fusion sub-model can adopt a weighted fusion model; the multi-source event fusion sub-model obtains the matching result and the matching correction weight, multiplies the matching result by the matching correction weight, and uses the product as the scene trigger score corresponding to the scene label; the multi-source event fusion sub-model averages the matching result and the matching correction weight, and uses the average value as the evaluation confidence level corresponding to the scene label; if there is evidence conflict between different event signals in scene determination, the multi-source event fusion sub-model adjusts the matching correction weight according to the degree of conflict, specifically: the edge computing node pre-associates each event signal with its corresponding candidate scene label, the association method is: the event features of the event signal are initially matched with each reference scene fingerprint in the scene label library, if the matching degree between the event features and a certain reference scene fingerprint exceeds the preset initial matching threshold, then the event feature is... Scene labels serve as candidate scene labels corresponding to the event signal. When candidate scene labels corresponding to multiple event signals within the same sub-time window are inconsistent or the operation actions contradict each other, it is determined that there is evidence conflict. The conflict degree is obtained by dividing the number of contradictory event signals by the total number of co-occurring event signals, and the conflict degree ranges from 0 to 1. The multi-source event fusion sub-model uses the value obtained by multiplying the matching correction weight by 1 and subtracting the conflict degree as the adjusted matching correction weight. Based on the adjusted matching correction weight, the scene trigger score and evaluation confidence corresponding to the scene label are recalculated. The recalculated scene trigger score is used as the scene trigger score of the scene label, and the recalculated evaluation confidence is used as the evaluation confidence of the scene label. The matching correction weights of other scene labels that have not experienced evidence conflict remain unchanged.

[0032] In some embodiments, determining candidate scene labels based on the scene trigger score and its evaluation confidence level for each scene label, combined with the trigger condition weights corresponding to each scene label, can be achieved through the following steps: The scene trigger score of each scene tag is multiplied by the corresponding trigger condition weight to obtain the weighted trigger score of each scene tag. Remove scenario labels whose evaluation confidence level is lower than the preset confidence threshold; From the remaining scene tags, select the scene tag with the highest weighted trigger score as the candidate scene tag; If the remaining scene labels are empty, no candidate scene labels will be generated.

[0033] In practice, the system first reads the scene trigger score and evaluation confidence level corresponding to each scene label from the output of the uncertainty inference model, and then reads the trigger condition weight corresponding to each scene label from the locally maintained scene label parameter table. Next, the scene trigger score of each scene label is multiplied by its corresponding trigger condition weight, and the product is used as the weighted trigger score for that scene label. The preset confidence threshold can be determined based on the lowest evaluation confidence level corresponding to the scene label correctly triggered by the edge computing node within a historical time period, or it can be pre-configured by the user, for example, setting the preset confidence threshold to 0.6. Correct triggering means that after the scene label triggers the linkage control command, no user operation signal with the opposite direction of the on / off state change of any controlled device is detected within the preset feedback time window. Scene labels with an evaluation confidence level lower than a preset confidence threshold are removed from the set of scene labels to be selected. If the set of scene labels remaining after removal is not empty, the edge computing node selects the scene label with the highest weighted trigger score from the remaining scene labels as a candidate scene label. If there are multiple scene labels with the same highest weighted trigger score, the scene label with the highest evaluation confidence level is selected as a candidate scene label. If the set of scene labels remaining after removal is empty, the edge computing node does not generate candidate scene labels, ends the current scene recognition process, and waits for the next preset time window to re-collect event signals. For example, the scene tag library contains a first scene tag and a second scene tag. The first scene tag has a scene trigger score of 0.8, an evaluation confidence level of 0.9, and a trigger condition weight of 0.5. The second scene tag has a scene trigger score of 0.7, an evaluation confidence level of 0.5, and a trigger condition weight of 0.8. The preset confidence threshold is 0.6. The weighted trigger score of the first scene tag is 0.4, and the weighted trigger score of the second scene tag is 0.56. However, the evaluation confidence level of the second scene tag is lower than the preset confidence threshold. Therefore, the second scene tag is removed, and the remaining scene tag is the first scene tag. The first scene tag is then used as a candidate scene tag.

[0034] In specific implementation, based on the candidate scene label, the corresponding device linkage strategy is obtained from the scene label library. This can be achieved as follows: using the candidate scene label as an index, the device linkage strategy associated with the candidate scene label is searched in the locally maintained scene label library. The scene label library stores the correspondence between each scene label and the device linkage strategy in key-value pairs, where the key is the scene label and the value is the device linkage strategy. The device linkage strategy includes a list of controlled device identifiers, the target on / off status corresponding to each controlled device, and the execution order and execution delay. Each controlled device identifier uniquely corresponds to a target on / off status. The execution order and execution delay can be empty, indicating that they are executed sequentially with a 0-second delay according to the order of the controlled device identifier list. The list of controlled device identifiers in the device linkage strategy is read, and each controlled device identifier and its corresponding target on / off status are taken as the control items to be controlled by the device linkage strategy. If there is no device linkage strategy corresponding to the candidate scene label in the scene label library, the edge computing node ends the current scene recognition process and does not generate linkage control instructions. For example, if the candidate scenario label is "Homecoming Mode," the corresponding device linkage strategy is as follows: the controlled device is identified as the living room light, with the target on / off state being "on"; the controlled device is identified as the living room air conditioner, with the target on / off state being "on"; and the controlled device is identified as the living room curtains, with the target on / off state being "off." The execution order is: living room light, living room air conditioner, and living room curtains are executed sequentially. The execution delay for turning on the living room light is 0 seconds, the execution delay for turning on the living room air conditioner is 0 seconds, and the execution delay for closing the living room curtains is 5 seconds. The list of controlled device identifiers, the target on / off state corresponding to each controlled device identifier, the execution order, and the execution delay in this device linkage strategy are then used as the basis for generating subsequent linkage control commands.

[0035] In specific implementation, when the scene trigger score of the candidate scene tag is continuously greater than the preset score threshold within the hysteresis duration, and its evaluation confidence is continuously greater than the preset confidence threshold within the hysteresis duration, a linkage control command is generated according to the device linkage strategy. The linkage control command includes the identifier of the controlled device to be controlled and its target on / off state, which can be implemented in the following way: the edge computing node reads the hysteresis duration, preset score threshold, and preset confidence threshold corresponding to the candidate scene tag from the scene tag parameter table maintained locally. The preset score threshold and preset confidence threshold can be determined based on the lowest values ​​of the scene trigger score and evaluation confidence corresponding to the correct triggering of the candidate scene tag in the past, or they can be pre-configured by the user. The correct triggering means that after the scene tag triggers the linkage control command, no user operation signal opposite to the direction of change of the on / off state of any controlled device is detected within the preset feedback time window. The hysteresis duration is set to an integer multiple of the preset time window length, and the hysteresis duration is not less than the preset time window length. When a candidate scene label is determined, a hysteresis timer is started, and the scene trigger score and evaluation confidence of the candidate scene label are recalculated at preset time windows. During the hysteresis duration, after each preset time window, the event signal within that preset time window is re-acquired, and scene fingerprint generation and uncertainty inference are re-executed to obtain a new scene trigger score and a new evaluation confidence of the candidate scene label. If the new scene trigger score is greater than a preset score threshold and the new evaluation confidence is greater than a preset confidence threshold, the hysteresis timer continues. If the new scene trigger score is not greater than a preset score threshold or the new evaluation confidence is not greater than a preset confidence threshold, the hysteresis timer is reset and the current triggering process ends, and no linkage control command is generated. When the hysteresis timer reaches the hysteresis duration and the conditions are met in each judgment during the period, it is determined that the scene trigger score and evaluation confidence of the candidate scene label continuously meet the triggering conditions within the hysteresis duration, and a linkage control command is generated according to the device linkage strategy corresponding to the candidate scene label. When generating linkage control instructions, the controlled device identifier list and the target on / off status corresponding to each controlled device identifier are read from the device linkage strategy. Each controlled device identifier and its corresponding target on / off status are written as a controlled item into the linkage control instruction. At the same time, the execution order and execution delay in the device linkage strategy are written into the linkage control instruction. If the execution order and execution delay are empty, they are written sequentially according to the order of the controlled device identifier list with a 0-second delay. The linkage control instruction after writing is used as the instruction to control the controlled device to execute the target on / off status.

[0036] In step 102, after executing the linkage control command, it monitors whether a user operation signal is received within a preset feedback time window that is opposite to the direction of the on / off state change of any controlled device in the linkage control command.

[0037] In some embodiments, after executing the linkage control command, monitoring whether a user operation signal is received within a preset feedback time window that is opposite in direction to the on / off state change of any controlled device in the linkage control command can be achieved as follows: After executing the linkage control command, record the identifier of each controlled device in the linkage control command and its target on / off state after execution, and start timing the preset feedback time window. The length of the preset feedback time window can be determined according to the type of controlled device in the device linkage strategy corresponding to each scene label and the distribution of user historical feedback time, or it can be configured by the user. For example, the preset feedback time window can be set to 60 seconds. The distribution of user historical feedback time is determined by statistically analyzing the time interval distribution of user reverse operations after historical triggering of each scene label, and the length of time covering a preset proportion of historical reverse operation events is used as the length of the preset feedback time window. Within a preset feedback time window, continuously monitor user operation signals from interactive terminals such as smart switch panels, mobile terminals, and voice control devices, and analyze the controlled device identifier and the on / off status after the operation carried in the user operation signal. If the user operation signal does not explicitly carry the on / off status after the operation, determine the on / off status after the operation based on the operation action in the user operation signal and the current actual on / off status of the controlled device. If the user operation signal lacks a controlled device identifier, ignore the user operation signal. For each controlled device identifier in the linkage control command, perform the following monitoring and judgment: match the controlled device identifier in the user operation signal with the controlled device identifier. If the match is successful, compare whether the on / off status after the operation in the user operation signal is opposite to the target on / off status of the controlled device in the linkage control command. If they are opposite, determine the user operation signal as a reverse user operation signal. For example, a linkage control command might include setting the target on / off state of the living room light to "on". After execution, at the 15th second within a preset feedback time window of 60 seconds, a user operation signal is received from the living room smart switch panel. This user operation signal indicates that the on / off state of the living room light should be set to "off". Since "off" is the opposite of "on", this user operation signal is determined to be a reverse user operation signal. If no user operation signal with the opposite direction of the on / off state change of any controlled device is detected within the preset feedback time window, it is determined that the user has not performed a reverse correction on the linkage control command. The detected reverse user operation signal is used as the reverse user operation signal for triggering rollback and parameter adjustment. If multiple reverse user operation signals are detected within the preset feedback time window, the first received reverse user operation signal is used as the reverse user operation signal for triggering rollback and parameter adjustment.

[0038] In step 103, if received, a rollback instruction is generated to restore the controlled device to its pre-execution state, reduce the trigger condition weight of the candidate scenario, increase its hysteresis duration, and clear the positive feedback counter.

[0039] In some embodiments, if received, a rollback instruction is generated to restore the controlled device to its pre-execution state, reduce the trigger condition weight of the candidate scenario, increase its hysteresis duration, and reset the positive feedback counter. This can be achieved through the following steps: Analyze the reverse user operation signals to determine the controlled device being operated on and the associated candidate scenarios; Generate a rollback command for the controlled device to restore the controlled device to the state before the linkage control command was executed; The trigger condition weight of the candidate scenario is reduced by a first preset step size, but not lower than the preset weight lower limit; the delay duration of the candidate scenario is increased by a second preset step size, but not higher than the preset delay upper limit. Then, the positive feedback counter for the candidate scenario is reset to zero.

[0040] The rollback command described in this application is only used to restore the controlled device that has been reversed by the user to its state before execution, while other controlled devices remain in their linked state.

[0041] It should be noted that this application monitors whether a user operation signal with the opposite direction of the on / off state change of the controlled device appears within a preset feedback time window after the execution of the linkage control command. When such a reverse operation signal appears, a rollback command is generated only for the controlled device that was reversed. This can locally correct the state of the falsely triggered device while keeping the linkage state of other controlled devices unchanged, avoiding the cancellation of the entire scene linkage due to the reverse operation of a single device, and reducing interference with the user's normal automation experience. At the same time, by reducing the trigger condition weight of the candidate scene and increasing its hysteresis duration, a penalty parameter adjustment is applied to the scene label that caused the false trigger, making it more difficult for the scene to be triggered under subsequent similar event conditions, thereby reducing the probability of repeated false triggers. In addition, the positive feedback counter is reset to zero, so that after the scene parameters are penalized and the number of positive feedbacks is re-accumulated, the trigger sensitivity needs to be restored. This prevents the accumulation of historical positive feedback from masking the problem exposed by this false trigger, ensuring that the self-learning process responds to user feedback in a timely manner, thereby improving the accuracy of the linkage control of smart switches throughout the house and user satisfaction.

[0042] In specific implementation, firstly, when executing the linkage control command, the candidate scene label corresponding to the linkage control command is recorded, as well as the association relationship between the identifiers of each controlled device in the linkage control command and the candidate scene label; the reverse user operation signal monitored within the preset feedback time window is obtained, and the controlled device identifier and the on / off state after the operation are parsed from the reverse user operation signal, and the device corresponding to the controlled device identifier is taken as the controlled device to be reverse operated; if the reverse user operation signal does not carry the on / off state after the operation, the edge computing node determines the on / off state after the operation based on the operation action in the reverse user operation signal and the current actual on / off state of the controlled device; at the same time, the association relationship between the controlled device identifier and the candidate scene label recorded when executing the linkage control command is read, the candidate scene label corresponding to the controlled device identifier is found from the association relationship, the candidate scene label is taken as the associated candidate scene label, and the determined controlled device to be reverse operated and the associated candidate scene label are taken as the reverse operation processing object.

[0043] Secondly, before executing the linkage control command, the identifier of each controlled device in the linkage control command and its on / off state before execution are recorded, and the recorded result is used as the pre-execution state; the pre-execution state is read, and the pre-execution on / off state corresponding to the controlled device that was reversed is obtained from the pre-execution state; then a rollback command is generated for the controlled device, which carries the identifier of the controlled device and the pre-execution on / off state, and the rollback command is sent to the controlled device to restore the controlled device to the on / off state before executing the linkage control command; other controlled devices keep their linkage state unchanged, and the generated and sent rollback command is used as the command to control the controlled device that was reversed to restore to the pre-execution state.

[0044] Then, the current trigger condition weight and current hysteresis duration corresponding to the candidate scene label are read from the locally maintained scene label parameter table. Next, the current trigger condition weight is subtracted by a first preset step size. If the subtracted value is less than the preset weight lower limit, the trigger condition weight is set to the preset weight lower limit; otherwise, the subtracted value is used as the new trigger condition weight. The current hysteresis duration is added to a second preset step size. If the added value is greater than the preset hysteresis upper limit, the hysteresis duration is set to the preset hysteresis upper limit; otherwise, the added value is used as the new hysteresis duration. The first preset step size, second preset step size, preset weight lower limit, and preset hysteresis upper limit are all pre-stored in the locally maintained scene label parameter table. The first and second preset step sizes can be determined based on the candidate scene label's historical correct trigger count. For example, when the candidate scene label has 0 historical correct trigger counts, the first preset step size can be set to 0.2 seconds, and the second preset step size can be set to 5 seconds; when the historical correct trigger count is 3 or more, the first preset step size... The first preset step size can be set to 0.1, and the second preset step size can be set to 3 seconds. The reason for this setting is that when the number of correct triggers in the past is small, it means that the candidate scene label has not been fully recognized by the user. It is necessary to reduce the trigger condition weight and increase the lag time with a larger step size to quickly suppress possible false triggers. When the number of correct triggers in the past is large, it means that the candidate scene label has been positively verified many times. The parameter adjustment range should be reduced to avoid excessive adjustment that would slow down the scene response or prevent it from being triggered. The preset weight lower limit can be determined according to the lowest triggerable weight of each scene label in the whole-house smart system. For example, the preset weight lower limit can be set to 0.3, because when the trigger condition weight is lower than 0.3, even if the scene fingerprint matches the reference scene fingerprint perfectly, the weighted trigger score may not meet the trigger requirements, resulting in the scene not being able to be triggered normally. The preset lag upper limit can be determined according to the maximum linkage response time that the user can accept. For example, the preset lag upper limit can be set to 30 seconds, because a lag time of more than 30 seconds will cause the scene trigger to be severely delayed, affecting the user experience. Then, write the updated trigger condition weights and the updated hysteresis duration back to the locally maintained scene label parameter table, and use the adjusted trigger condition weights and the adjusted hysteresis duration as the updated scene parameters for the candidate scene label.

[0045] Finally, the positive feedback counter corresponding to the candidate scene label is cleared to zero, and the cleared positive feedback counter is written back to the locally maintained scene label parameter table to re-accumulate the number of times the candidate scene label has not received a reverse user operation signal after subsequent linkage control execution. The cleared positive feedback counter is used as the counter for re-accumulating the number of positive feedback times for the candidate scene label.

[0046] In step 104, if no response is received, the linkage operation is maintained and the positive feedback counter is incremented by 1; when the positive feedback counter reaches a preset threshold, the trigger condition weight of the candidate scenario is increased, its hysteresis duration is shortened, and the positive feedback counter is cleared.

[0047] In some embodiments, when the positive feedback counter reaches a preset threshold, the trigger condition weight of the candidate scenario is increased, its hysteresis duration is shortened, and the positive feedback counter is reset to zero. This can be achieved by the following steps: In response to the positive feedback counter reaching a preset threshold, the trigger condition weight of the candidate scenario is increased by a third preset step size, but not higher than the preset weight upper limit. The hysteresis duration of the candidate scenario is shortened by the fourth preset step, and is not lower than the preset hysteresis lower limit; The positive feedback counter is reset to zero to re-accumulate the number of positive feedback events.

[0048] In specific implementation, firstly, if no user operation signal opposite to the on / off state change direction of any controlled device in the linkage control command is detected within the preset feedback time window, the edge computing node maintains the linkage operation unchanged and increments the positive feedback counter corresponding to the candidate scene label by 1; reads the incremented positive feedback counter value and compares the read value with a preset threshold; if the read value is less than the preset threshold, the edge computing node does not perform positive parameter adjustment, ends the current feedback processing, and waits for feedback monitoring after the next linkage control command is executed; if the read value is equal to the preset threshold, it is determined that the positive feedback counter has reached the preset threshold, triggering the positive parameter adjustment process. The current trigger condition weight and the third preset step size corresponding to the candidate scene label are read from the locally maintained scene label parameter table; the current trigger condition weight is added to the third preset step size. If the sum is greater than the preset weight upper limit, the trigger condition weight is set to the preset weight upper limit; otherwise, the sum is used as the new trigger condition weight. The preset threshold, third preset step size, and preset weight upper limit are all pre-stored in a locally maintained scene label parameter table. The preset threshold can be determined based on the number of consecutive positive feedback confirmations expected by the user; for example, the preset threshold can be set to 3 times. The third preset step size can be determined based on the historical correct trigger count of the candidate scene label; for example, when the historical correct trigger count is less than 5 times, the third preset step size can be set to 0.1, and when the historical correct trigger count is greater than or equal to 5 times, the third preset step size can be set to 0.05. This setting is to appropriately increase the weight when the scene label has not yet obtained sufficient positive verification, and to use a smaller step size after the scene label has been correctly triggered multiple times to prevent the weight from increasing excessively. The preset weight upper limit can be determined based on the highest trigger weight allowed for each scene label in the whole-house smart system; for example, the preset weight upper limit can be set to 0.9 to prevent the trigger condition weight from being too high, causing the scene to be triggered immediately even if there are only a few matches. The historical correct trigger count refers to the number of times that no reverse user operation signal was detected within the preset feedback time window after the scene label historically triggered the linkage control command. Write the adjusted trigger condition weights back to the locally maintained scene label parameter table, and use the adjusted trigger condition weights as the updated trigger condition weights for the candidate scene label.

[0049] Then, the current hysteresis duration and the fourth preset step size corresponding to the candidate scene label are read from the locally maintained scene label parameter table. The current hysteresis duration is subtracted from the fourth preset step size. If the value after subtraction is less than the preset hysteresis lower limit, the hysteresis duration is set as the preset hysteresis lower limit; otherwise, the value after subtraction is used as the new hysteresis duration. Both the fourth preset step size and the preset hysteresis lower limit are pre-stored in the locally maintained scene label parameter table. The fourth preset step size can be determined based on the historical correct trigger count of the candidate scene label. For example, when the historical correct trigger count is less than 5, the fourth preset step size can be set to 2 seconds; when the historical correct trigger count is greater than or equal to 5, the fourth preset step size can be set to 5 seconds. This setting is to accelerate the shortening of the hysteresis duration after the scene label has obtained multiple positive verifications to improve response speed, while using a smaller step size when there are fewer positive verifications to prevent the hysteresis duration from shortening too quickly and causing false triggers. The preset hysteresis lower limit can be determined based on the shortest scene confirmation time allowed for each scene label in the whole-house smart system. For example, the preset hysteresis lower limit can be set to 3 seconds to reserve a necessary time window for confirming scene stability. Write the adjusted hysteresis duration back to the locally maintained scene label parameter table, and use the adjusted hysteresis duration as the updated hysteresis duration for the candidate scene label.

[0050] Finally, the positive feedback counter corresponding to the candidate scene label is cleared to zero, and the cleared positive feedback counter is written back to the locally maintained scene label parameter table, so that the candidate scene label starts to accumulate the number of times that a reverse user operation signal is not received after the execution of subsequent linkage control commands from zero.

[0051] Through steps 101 to 104 above, scene fingerprints and their confidence levels are generated using multi-source event signals. Uncertainty reasoning is used to obtain scene trigger scores and evaluation confidence levels for each scene label. After continuous confirmation within the hysteresis period, a linkage control command is generated. After linkage execution, based on whether a reverse user operation signal appears within a preset feedback time window, the trigger condition weights and hysteresis periods of the corresponding scene labels are adjusted punitively or rewardingly. A positive feedback counter controls the positive adjustment rhythm, forming an adaptive closed-loop adjustment of scene trigger parameters. This control logic enables the whole-house smart switch linkage control to dynamically correct scene trigger sensitivity based on actual user feedback, gradually improving the response speed of correct scenes while reducing false triggers, thereby enhancing the accuracy of whole-house smart switch linkage control and user satisfaction.

[0052] On the other hand, in some embodiments, this application provides a whole-house smart switch linkage control system based on scene self-learning, referencing... Figure 3The figure is a schematic diagram of a whole-house smart switch linkage control system based on scene self-learning according to some embodiments of this application. The whole-house smart switch linkage control system based on scene self-learning includes: a scene recognition module 301, a processing module 302, and an execution module 303, which are described below: Scene recognition module 301, in this application, is mainly used to identify candidate scenes and generate linkage control instructions based on the collected event signals and the trigger condition weights and hysteresis durations of each scene label locally maintained by the edge computing node. Processing module 302, in this application, is used to monitor whether a user operation signal is received within a preset feedback time window after executing the linkage control command, which is opposite to the direction of the on / off state change of any controlled device in the linkage control command. The execution module 303 in this application is mainly used to generate a rollback instruction if received, restore the controlled device to the state before execution, reduce the trigger condition weight of the candidate scenario, increase its delay time, and clear the positive feedback counter. In this application, the execution module 303 is also used to maintain the linkage operation and increment the positive feedback counter by 1 if no response is received; when the positive feedback counter reaches a preset threshold, increase the trigger condition weight of the candidate scenario, shorten its hysteresis duration, and clear the positive feedback counter.

[0053] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described whole-house smart switch linkage control method based on scene self-learning.

[0054] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer device implementing a scene-based self-learning method for the linkage control of smart switches throughout the house, according to some embodiments of this application. The scene-based self-learning method for the linkage control of smart switches throughout the house in the above embodiments can... Figure 4 The computer device shown is used to implement this, and the computer device 400 includes at least one processor 401, a communication bus 402, a memory 403, and at least one communication interface 404.

[0055] Processor 401 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0056] The communication bus 402 can be used to transmit information between the aforementioned components.

[0057] The memory 403 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 403 may exist independently and be connected to the processor 401 via the communication bus 402. The memory 403 may also be integrated with the processor 401.

[0058] The memory 403 stores program code for executing the solution of this application, and its execution is controlled by the processor 401. The processor 401 executes the program code stored in the memory 403. The program code may include one or more software modules. In the above embodiment, the whole-house smart switch linkage control method based on scene self-learning can be implemented by the processor 401 and one or more software modules in the program code in the memory 403.

[0059] Communication interface 404 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0060] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0061] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0062] In addition, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned whole-house smart switch linkage control method based on scene self-learning.

[0063] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0064] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for controlling the linkage of smart switches in a whole house based on scene self-learning, executed by edge computing nodes deployed in a smart house system, characterized in that, The method includes: Based on the collected event signals and the trigger condition weights and hysteresis durations of each scene label locally maintained by the edge computing node, candidate scenes are identified and linkage control commands are generated. After executing the linkage control command, monitor whether a user operation signal is received within a preset feedback time window that is opposite to the direction of the on / off state change of any controlled device in the linkage control command. If received, a rollback command is generated to restore the controlled device to its state before execution, reduce the trigger condition weight of the candidate scenario, increase its hysteresis duration, and clear the positive feedback counter. If no response is received, the linkage operation is maintained and the positive feedback counter is incremented by 1. When the positive feedback counter reaches a preset threshold, the trigger condition weight of the candidate scenario is increased, its hysteresis duration is shortened, and the positive feedback counter is cleared to zero.

2. The method as described in claim 1, characterized in that, The event signal includes at least one of the following: smart switch action signal, human presence signal, light signal, and door / window magnetic signal.

3. The method as described in claim 1, characterized in that, Based on the collected event signals and the trigger condition weights and latency durations of each scene label locally maintained by the edge computing node, candidate scenes are identified and linkage control commands are generated, specifically including: The event signals collected within the preset time window are analyzed to obtain the event characteristics of each event signal; Based on the event characteristics and timestamp information of each event signal, a scene fingerprint is generated, and the confidence level of the scene fingerprint is obtained by fusing the data quality characteristics of each event signal. The scene fingerprint and its confidence level are input into a preset uncertainty inference model, and fused with the reference scene fingerprints of each scene tag in the scene tag library maintained locally by the edge computing node to output the scene trigger score and its evaluation confidence level for each scene tag; wherein, the uncertainty inference model is used to handle the data quality uncertainty that exists in the process of event signal acquisition, transmission and fusion. Candidate scene labels are determined based on the scene trigger scores and their evaluation confidence levels for each scene label, combined with the trigger condition weights corresponding to each scene label. Based on the candidate scene tags, obtain the corresponding device linkage strategy from the scene tag library; When the scenario trigger score of the candidate scenario tag is continuously greater than the preset score threshold within the hysteresis duration, and its evaluation confidence is continuously greater than the preset confidence threshold within the hysteresis duration, a linkage control instruction is generated according to the device linkage strategy. The linkage control instruction includes the identifier of the controlled device to be controlled and its target on / off status.

4. The method as described in claim 3, characterized in that, Based on the event characteristics and timestamp information of each event signal, a scene fingerprint is generated, specifically including: The preset time window is divided into multiple sub-time windows, and based on the timestamp information of each event signal, event signals falling within the same sub-time window are identified as co-occurring event signals. The event features of the co-occurrence event signals are extracted to generate a scene fingerprint that characterizes the co-occurrence relationship of device type, operation action and spatial location.

5. The method as described in claim 3, characterized in that, Based on the scene trigger score and its evaluation confidence level for each scene tag, and combined with the trigger condition weights corresponding to each scene tag, candidate scene tags are determined, specifically including: The scene trigger score of each scene tag is multiplied by the corresponding trigger condition weight to obtain the weighted trigger score of each scene tag. Remove scenario labels whose evaluation confidence level is lower than the preset confidence threshold; From the remaining scene tags, select the scene tag with the highest weighted trigger score as the candidate scene tag; If the remaining scene labels are empty, no candidate scene labels will be generated.

6. The method as described in claim 1, characterized in that, The user operation signals include at least one of the following: smart switch panel operation signals, mobile terminal control signals, and voice control signals.

7. The method as described in claim 1, characterized in that, The rollback command is only used to restore the controlled device that has been reversed by the user to its state before execution; other controlled devices remain in the same linked state.

8. The method as described in claim 7, characterized in that, If received, a rollback instruction is generated to restore the controlled device to its pre-execution state, reduce the trigger condition weight of the candidate scenario, increase its hysteresis duration, and reset the positive feedback counter. Specifically, this includes: Analyze the reverse user operation signals to determine the controlled device being operated on and the associated candidate scenarios; Generate a rollback command for the controlled device to restore the controlled device to the state before the linkage control command was executed; The trigger condition weight of the candidate scenario is reduced by a first preset step size, but not lower than the preset weight lower limit; the delay duration of the candidate scenario is increased by a second preset step size, but not higher than the preset delay upper limit. Then, the positive feedback counter for the candidate scenario is reset to zero.

9. The method as described in claim 1, characterized in that, When the positive feedback counter reaches a preset threshold, the trigger condition weight of the candidate scenario is increased, its hysteresis duration is shortened, and the positive feedback counter is reset to zero. Specifically, this includes: In response to the positive feedback counter reaching a preset threshold, the trigger condition weight of the candidate scenario is increased by a third preset step size, but not higher than the preset weight upper limit. The hysteresis duration of the candidate scenario is shortened by the fourth preset step, and is not lower than the preset hysteresis lower limit; The positive feedback counter is reset to zero to re-accumulate the number of positive feedback events.

10. A whole-house intelligent switch linkage control system based on scene self-learning, characterized in that, include: The scene recognition module is used to identify candidate scenes and generate linkage control commands based on the collected event signals and the trigger condition weights and hysteresis durations of each scene label maintained locally by the edge computing node. The processing module is used to monitor, within a preset feedback time window, whether a user operation signal is received that is opposite to the direction of the on / off state change of any controlled device in the linkage control command after the linkage control command is executed. The execution module is used to generate a rollback instruction if received, restore the controlled device to the state before execution, reduce the trigger condition weight of the candidate scenario, increase its delay time, and clear the positive feedback counter. The execution module is also used to maintain the linkage operation and increment the positive feedback counter by 1 if no response is received; when the positive feedback counter reaches a preset threshold, increase the trigger condition weight of the candidate scenario, shorten its hysteresis duration, and clear the positive feedback counter.

Images