Intelligent air state control electric anomaly detection method and system based on multi-modal data fusion

Abstract

This invention discloses a method and system for detecting anomalies in intelligent spatial power control based on multimodal data fusion, relating to the fields of intelligent power management and multi-sensor information fusion. It addresses the problems of timing misalignment and false alarms caused by inherent sensor delays in traditional fixed-time-window judgments. The method includes synchronously acquiring sensor signals to generate state sequences, extracting personnel and power state change events, analyzing timing dependencies and inferring the dominant timing direction, estimating inherent sensor time delays, aligning the current event timing and calculating anomaly scores, determining anomalies based on dual thresholds, and generating control commands or warning logs. The system comprises six collaborative functional modules, achieving adaptive delay compensation, reducing false alarm rates, and improving the practicality and control effectiveness of the intelligent spatial power control system.

Claims

1. A method for detecting intelligent spatial-state electrical anomalies based on multimodal data fusion, characterized in that: Includes the following steps: S1. Simultaneously acquire the output signals of the presence sensor and the power sensor, and generate the presence state sequence and the power state sequence respectively; S2. Extract events representing changes in personnel status from the existence state sequence, and extract events representing changes in electricity consumption status from the power state sequence; S3. Based on the events extracted from the historical period, analyze the temporal dependency between the personnel status change event and the power consumption status change event, and infer the dominant causal direction between the two events; S4. Based on the dominant causal direction, estimate the inherent time delay of the power sensor relative to the presence sensor from events within the historical period; S5. Use the inherent time delay to perform time-series alignment of events within the current detection period, and calculate the anomaly assessment score of the current electricity consumption behavior based on the aligned event time relationship and the dominant causal direction. S6. Based on whether the abnormality assessment score exceeds a preset threshold, determine whether the current electricity consumption behavior is abnormal and generate a control command.

2. The intelligent sensing spatial control power anomaly detection method based on multimodal data fusion according to claim 1, characterized in that: S1 further includes the following: A shared timing triggering unit sends synchronous sampling commands to the presence sensor and the power sensor at a fixed sampling period. In response to a synchronous sampling command, a first electrical signal representing the presence or absence of personnel in the monitoring area is acquired from the presence sensor, and a second electrical signal representing the real-time power value of the monitored power circuit is acquired from the power sensor. The first electrical signal is compared with a preset level value, and a first digital signal is output. The high and low levels of the first digital signal correspond to the presence and absence of people in the monitoring area, respectively. The second electrical signal is amplified and converted from analog to digital, and the converted value is compared with a preset power threshold to output a second digital signal; The high and low levels of the second digital signal correspond to the high-power and low-power states of the monitored power circuit, respectively. The level values ​​of the first digital signal are continuously recorded in chronological order, with the sampling period as the time interval, to form the existence state sequence; The power state sequence is formed by continuously recording the level values ​​of the second digital signal in chronological order, with the adoption period as the time interval.

3. The intelligent sensing spatial control power anomaly detection method based on multimodal data fusion according to claim 1, characterized in that: S2 further includes the following: The sequence of existing states is traversed sequentially. When the state values ​​of two adjacent sampling points in the sequence of existing states are different, a state transition is marked at that point. For each state transition of the marker, starting from that transition point, check whether the state values ​​of several consecutive sampling points remain unchanged; If the state values ​​of several consecutive sampling points are consistent with the state value after the transition, the state transition is determined to be valid, and the time and direction of the transition are recorded. The transition direction includes transitioning from a occupied state to an unoccupied state, and transitioning from an unoccupied state to a occupied state; The effective transition from a occupied state to an unoccupied state is defined as a departure event; The effective transition from an unoccupied state to an occupied state is defined as an entry event; The power state sequence is traversed sequentially. When the state values ​​of two adjacent sampling points in the power state sequence are different, a state transition is marked at that point. Similarly, for each state transition of the marker, starting from that transition point, check whether the state values ​​of multiple consecutive sampling points remain unchanged; If the state values ​​of multiple consecutive sampling points are consistent with the state value after the transition, the state transition is determined to be valid, and the time and direction of the transition are recorded. The transition direction includes transitioning from a high-power state to a low-power state, and transitioning from a low-power state to a high-power state. The effective transition from a high-power state to a low-power state is defined as a power-off event. The effective transition from a low-power state to a high-power state is defined as a power-on event.

4. The intelligent sensing spatial control power anomaly detection method based on multimodal data fusion according to claim 3, characterized in that: S3 further includes the following: A time correlation window is defined; the time correlation window defines the maximum allowable time interval for determining whether two events have a temporal correlation. From the events extracted within the historical period, the total number of departure events and the total number of power-off events are counted. For each departure event, search backward within the time association window to identify the first power-off event that occurs after it. Calculate the first association probability based on the identification results; The first association probability is calculated by dividing the number of departure events that are found to be subsequent power-off events within the time association window by the total number of departure events. For each power-off event, search backward within the time association window to identify the first departure event that occurs after it; Calculate the second association probability based on the identification results; The second association probability is calculated by dividing the number of power-off events that are found to be subsequent departure events within the time association window by the total number of power-off events. The first association probability and the second association probability are numerically compared; When the first association probability is greater than the second association probability, it is inferred that the dominant timing direction is that the departure event takes precedence over the power-off event; When the second association probability is greater than the first association probability, it is inferred that the dominant timing direction is that the power-off event takes precedence over the departure event; When the difference between the first association probability and the second association probability does not reach a preset threshold, the dominant time sequence direction is determined to be undetermined. The dominant temporal direction represents the stable chronological order of the two types of events in the timestamp sequence observed by the system.

5. The intelligent sensing spatial control anomaly detection method based on multimodal data fusion according to claim 4, characterized in that: S4 further includes the following: Based on the dominant temporal direction, filter the valid event pairs within the historical period; However, when the dominant timing direction is that the power-off event takes precedence over the departure event, a valid event pair consists of one power-off event and the first departure event that occurs after it and within the time association window. When the dominant timing direction is that the departure event takes precedence over the power-off event, a valid event pair consists of a departure event and the first power-off event that occurs after it and within the time association window. For a valid event pair, calculate the observation time difference between the power-off event timestamp recorded by the power sensor and the departure event timestamp recorded by the presence sensor; The formula for calculating the observation time difference is: ΔTi=tLi-tCi Where: ΔTi represents the observation time difference of the i-th valid event pair; tCi represents the timestamp of the power-off event recorded by the power sensor in the event pair; tLi represents the timestamp of the departure event recorded by the presence sensor in the event pair; The sign setting in the calculation formula is based on the dominant timing direction; When the dominant timing direction is that the power-off event takes precedence over the departure event, tCi is earlier than tLi in time, and ΔTi is a positive number; The observation time differences of N valid event pairs within the historical period are obtained to form the observation time difference set {ΔT1,ΔT2,...,ΔTN}; Statistical analysis was performed on the set of observation time differences to remove outliers caused by noise; The central tendency statistic of the set of observation time differences after removing outliers is estimated as the inherent time delay Δ of the power sensor relative to the existing sensor. The inherent time delay Δ is calculated as follows: Δ=Median({ΔTk}) Where Median(·) represents the median function; {ΔTk} represents the set of observation time differences after removing outliers.

6. The intelligent sensing spatial control power anomaly detection method based on multimodal data fusion according to claim 1, characterized in that: S5 further includes the following: For a departure event extracted within the current detection period, obtain its timestamp TL recorded by the present sensor; For a power-off event extracted within the current detection period, obtain its timestamp TC recorded by the power sensor; The timestamp of the power-off event is corrected based on the estimated inherent time delay Δ. The correction formula is as follows: TCc = TC + Δ; Where TCc represents the power-off event timestamp after timing alignment correction; Calculate the alignment time interval D between the departure event and the corrected power-off event; The formula for calculating the alignment time interval D is: D = TL - TCc; The anomaly evaluation score S is calculated based on the dominant temporal direction and the sign and magnitude of the alignment time interval D; The calculation logic for the anomaly assessment score S is as follows: When the dominant timing direction is that the power-off event takes precedence over the departure event: If D>0 and the value of D is within the preset delay interval [Lmin,Lmax], it is determined to be in normal operation mode, and S=0 is set. If D≤0, then it is determined to be an abnormal mode, and S=1 is set; If D>0 and D>Lmax, then an abnormal delay is identified, and S is calculated using the following formula: S = min(1, (D-Lmax) / Z); Where Z is a preset proportionality coefficient; When the dominant timing direction is that the departure event takes precedence over the power-off event: If D < 0 and the absolute value of D is within the preset delay interval [Lmin, Lmax], then it is determined to be in normal operation mode, and S = 0 is set. If D≥0, then it is determined to be an abnormal mode, and S=1 is set; If D < 0 and the absolute value of D is greater than Lmax, then an abnormal delay is identified, and S is calculated using the following formula: S=min(1,(|D|-Lmax) / Z); When the dominant timing direction is undetermined, the default scoring logic is used; The default scoring logic is calculated based on the unaligned raw observation time difference ΔA, as shown in the following formula: ΔA = TL - TC; If the absolute value of ΔA is within the preset conservative time threshold Q, then set S=0; Otherwise, set S=1; Wherein, the delay interval [Lmin, Lmax], the scaling factor Z, and the conservative time threshold Q are all preset system parameters.

7. The intelligent sensing spatial control power anomaly detection method based on multimodal data fusion according to claim 6, characterized in that: S6 further includes the following: The anomaly assessment score S is compared with a preset first judgment threshold K1 and a second judgment threshold K2; Where K1 and K2 are preset system parameters, and satisfy 0 <K1<K2<1; Based on the comparison results, perform the corresponding judgment and operation: If S≥K2, the current electricity consumption behavior is determined to be abnormal, and a power outage control command is generated and sent to the power supply circuit of the controlled appliance. If S≤K1, then the current power consumption behavior is determined to be normal, and no control command is generated; If S is greater than K1 and less than K2, the current electricity consumption behavior is determined to be in an uncertain state. At this time, no power outage control command is generated, but an early warning log containing a timestamp, the abnormal evaluation score S, and related event information is generated for subsequent manual verification.

8. A smart sensing spatial-state power control anomaly detection system based on multimodal data fusion, applied to the smart sensing spatial-state power control anomaly detection method based on multimodal data fusion as described in any one of claims 1-7, characterized in that: It includes a synchronous acquisition and sequence generation module, an event extraction module, a time sequence direction inference module, an inherent delay prediction module, an anomaly scoring calculation module, and a decision and instruction generation module; The synchronous acquisition and sequence generation module is used to synchronously acquire the output signals of the presence sensor and the power sensor at a fixed sampling period, and generate the presence state sequence and the power state sequence respectively. The event extraction module is used to extract the departure event and the entry event from the existence state sequence, and to extract the power-off event and the power-on event from the power state sequence; The time-series direction inference module calculates the first correlation probability and the second correlation probability based on historical event data, and infers the dominant time-series direction by comparing the two. The inherent delay prediction module is used to filter valid event pairs according to the dominant time series direction, calculate the observation time difference and predict the inherent time delay Δ; The anomaly scoring calculation module is used to perform time-series alignment of the current event using the inherent time delay Δ, and to calculate the anomaly evaluation score S according to the dominant time-series direction; The decision and instruction generation module is used to compare the anomaly evaluation score S with the first judgment threshold K1 and the second judgment threshold K2, and generate corresponding control instructions or warning logs based on the comparison results.

Images