Intelligent kitchen multi-device linkage management method based on artificial intelligence

Abstract

The application discloses a smart kitchen multi-device linkage management method based on artificial intelligence and belongs to the technical field of artificial intelligence and smart home. The method comprises the following steps: collecting real-time running state parameters and user operation time sequence data of multiple electric devices in a kitchen, and constructing a dynamic graph taking devices as nodes and taking running states as attribute values; extracting a state switching time difference sequence and a user operation sequence chain for each pair of device nodes, and calculating an operation coupling distance through a dynamic time warping algorithm; screening strong linkage device combinations according to the operation coupling distance and a tightness threshold value, and generating a linkage correlation degree weight; monitoring user operation behaviors in real time and matching the combinations, triggering a pre-starting instruction of a non-starting device according to the weight; acquiring an actual response delay time of the device after the pre-starting instruction is sent, and feeding back to the dynamic graph to update the operation coupling distance and the linkage correlation degree weight, so that online adaptive adjustment is realized.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and smart home technology, specifically to a method for multi-device linkage management in a smart kitchen based on artificial intelligence. Background Technology

[0002] In a smart kitchen scenario, various electrical appliances such as range hoods, cooktops, ovens, microwave ovens, and dishwashers often have inherent operational relationships during the user's cooking process. For example, after turning on the cooktop, the user usually turns on the range hood next, and when using the oven, a timer or lighting may be activated simultaneously. Existing technologies typically rely on preset fixed rules or scene patterns to achieve device linkage, such as using IF-THEN logic to predefine that when one device is started, another device will automatically execute a specified action. However, this static rule configuration method requires users to manually set linkage relationships, which cannot adapt to the personalized operating habits of different users, and it is even more difficult to capture the implicit, dynamically changing device coordination patterns in the user's operating rhythm. Existing technical solutions also generally lack the awareness and utilization of device response delays. After the linkage command is issued, the actual start-up delay of each device varies due to factors such as device type, load conditions, and power supply characteristics. This difference directly affects the optimal triggering time of subsequent linkage actions. If the system makes linkage decisions based solely on theoretical command times, ignoring the physical delay between command issuance and actual stable equipment operation, the linkage timing will gradually deviate from the user's actual operating experience over time. This can lead to premature or delayed pre-start times, resulting in energy waste or failure to provide timely assistance. Furthermore, the construction and updating of equipment linkage relationships in existing solutions are fragmented, lacking an adaptive mechanism that uses equipment response delays as feedback signals to continuously optimize linkage model parameters.

[0003] This application addresses two core issues. First, it addresses how to automatically capture, without pre-setting rules, the implicit synchronization or sequential coordination patterns between devices within user operations from a time-series deep mining perspective, ensuring that the linkage relationships accurately reflect the actual operational sequence rhythm of individual users. Second, it addresses how to use the actual physical response delay of the devices after each linkage command execution as closed-loop feedback to drive the dynamic correction of the correlation strength between devices, enabling linkage management parameters to continuously track the evolution of actual operating conditions such as equipment aging and load changes. Summary of the Invention

[0004] The purpose of this invention is to provide a smart kitchen multi-device linkage management method based on artificial intelligence, which realizes automatic discovery of device linkage relationships and online adaptive optimization of parameters without the need for preset rules, thereby improving the personalized matching degree and real-time response accuracy of multi-device collaborative work in the smart kitchen.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a smart kitchen multi-device linkage management method based on artificial intelligence, the method comprising the following steps: Real-time operating status parameters and user operation time-series data of multiple electrical devices in the kitchen are collected, and a dynamic graph is constructed with devices as nodes and operating status as attribute values. Preferably, this dynamic graph adopts a heterogeneous graph neural network architecture, which uses device type as a node attribute and operation coupling distance as an edge feature for joint embedding learning, thereby integrating potential relationships between devices in the node representation and improving the accuracy of subsequent linkage relationship mining.

[0006] As a preferred technical solution of the present invention, the collection of real-time operating status parameters and user operation timing data is specifically achieved in the following manner: A non-intrusive load monitoring terminal deployed on the power supply lines of each kitchen device synchronously captures the voltage transient characteristics and current waveform distortion points of each device; the voltage transient characteristics and current waveform distortion points are input into a pre-trained device type identifier, which outputs the device identifier that caused the current state switch and the power step value before and after the switch; using the system clock as a reference, each state switch event is timestamped, and the device operation actions triggered by the user in ascending order of time are recorded to form user operation timing data. This method eliminates the need for sensors installed inside the devices, relying solely on the electrical quantities at the power supply end to achieve synchronous perception of the status of multiple devices, ensuring the integrity and time consistency of data collection.

[0007] For each pair of device nodes in the dynamic graph, a sliding time window with the current time as the endpoint and a fixed duration as the interval length is extracted from the user operation timing data. Within this sliding time window, for any pair of device nodes, the time difference between the state transition of the second device and the state transition of the first device within a preset time interval is recorded. All time differences are arranged in chronological order to obtain a state transition time difference sequence. Simultaneously, the frequency of the occurrence of operation events of the two devices is counted, and the direction indicator of the user operation sequence chain is determined based on the frequency relationship. The extracted time difference sequence and direction indicator can accurately reflect the time pattern and operation habits of the interaction between devices.

[0008] Based on this, the operational coupling distance between the two devices is calculated using a dynamic time warp algorithm. Specifically, the state switching time difference sequence of the first device is used as the first sequence, and the state switching time difference sequence of the second device is used as the second sequence. A distance matrix is ​​constructed for all point pairs between the two sequences. In this distance matrix, a curved path with the minimum cumulative distance is searched from the starting point to the ending point. This curved path allows one point in one sequence to correspond to multiple consecutive points in the other sequence. The cumulative distance value at the end of the curved path is used as the operational coupling distance between the two devices. The smaller the operational coupling distance, the more synchronous the state switching of the two devices is in time. The dynamic time warp algorithm can effectively tolerate nonlinear fluctuations in device response time and accurately measure the tightness of linkage even when there is scaling or offset on the time axis.

[0009] Based on the operational coupling distance and a preset tightness threshold, device combinations with strong linkage relationships are selected. Preferably, the calculated operational coupling distances between all device pairs are sorted in ascending order of value; from the sorted device pairs, device pairs with operational coupling distances less than the preset tightness threshold are selected and marked as strongly linked device combinations; if the same device appears in multiple strongly linked device combinations, an initial linkage correlation weight is assigned to the device in different combinations based on the value of the operational coupling distance, and the initial linkage correlation weight is inversely proportional to the operational coupling distance. More preferably, the preset tightness threshold is dynamically adjusted based on the total number of devices in the kitchen and the average operational coupling distance, and the threshold value is positively correlated with the total number of devices. In this way, when the number of kitchen devices increases, the threshold is adaptively widened to avoid missing truly linked device combinations; when the number of devices is small, the threshold is tightened to eliminate accidental temporal couplings and ensure the robustness of strong linkage identification.

[0010] Real-time monitoring of user actions performed in the kitchen at any given moment. Preferably, a motion sensor array and image acquisition unit are used to capture the three-dimensional motion trajectory of the user's hand on the kitchen countertop and their touch switch actions. The three-dimensional motion trajectory is compared with pre-stored gesture templates for each device to identify the type of device the user intends to operate. When the user's intention to operate a certain device is identified, the device identifier and the start time of the operation are recorded as the user action currently being performed. By fusing motion trajectories and gesture templates, the operation intention can be accurately identified the moment the user touches or approaches a device, providing a time window for pre-activating other devices.

[0011] The user operation is matched with the generated strongly linked device combinations. Specifically: the first device identifier corresponding to the current user operation is obtained; all strongly linked device combinations are traversed to find combinations containing the first device identifier; if found, the device identifiers other than the first device in the combination are extracted as a set of devices to be linked; the current running status of each device in the set of devices to be linked is checked, and devices in the "not started" state are selected as target pre-start devices. When multiple strongly linked device combinations containing the first device identifier are found, the devices to be linked in all combinations are merged, and then pre-start instructions are triggered sequentially according to the linkage correlation weight from high to low, thereby prioritizing the activation of the device with the highest linkage confidence.

[0012] Based on the aforementioned linkage correlation weight, a pre-start command is triggered for the non-started device in the corresponding device group. Preferably, the linkage correlation weight of the target pre-start device in its strongly linked device group is obtained and compared with a preset pre-start confidence threshold. When the linkage correlation weight is greater than or equal to the pre-start confidence threshold, a pre-start command is immediately sent to the target pre-start device, the command content including the device's target operating mode and power parameters. When the linkage correlation weight is less than the pre-start confidence threshold, the pre-start command is sent after a short waiting interval inversely proportional to the linkage correlation weight. This hierarchical triggering mechanism ensures rapid start-up of high-probability linked devices while setting an observation buffer period for low-probability linked devices, effectively avoiding energy waste and equipment damage caused by false triggering.

[0013] Obtain the actual response delay time of each device after the pre-start command is issued. Simultaneously with issuing the pre-start command to the target pre-start device, record the system timestamp of the command issuance time; continuously monitor the current waveform changes of the target pre-start device; when the effective current value of the device rises to the set proportion of the power parameter corresponding to the target operating mode in the pre-start command, record the confirmation time of the device's actual start-up completion; subtract the command issuance time from the confirmation time to obtain the actual response delay time of the device in this instance.

[0014] The actual response delay time is fed back into the dynamic graph to update the operation coupling distance and linkage correlation weights. For a device pair that generates an actual response delay time, the actual response delay time is added as a new element to the end of the original state switching time difference sequence of that device pair, while the element with the earliest timestamp in the sequence is removed, keeping the sequence length fixed. The calculation steps of the dynamic time warping algorithm are re-executed using the updated state switching time difference sequence to obtain the updated operation coupling distance. Based on the updated operation coupling distance, the determination of strongly linked device combinations and the linkage correlation weights are recalculated to complete the online adaptive adjustment of linkage management parameters. Through the above closed-loop feedback mechanism, the system can continuously correct linkage parameters as user operating habits drift and the rate of equipment aging changes, ensuring that the linkage triggering timing is always synchronized with the actual operating rhythm, improving response timeliness while reducing the number of invalid pre-starts, achieving a kitchen management effect that balances energy saving and intelligence.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By constructing a dynamic graph with devices as nodes and operating status as attribute values ​​using real-time operating status parameters of various kitchen devices and user operation timing data, and extracting the state switching time difference sequence between each pair of device nodes, the operation coupling distance is calculated using a dynamic time warping algorithm. This distance reflects the similarity or following behavior of the state switching between two devices in the time dimension. The dynamic time warping algorithm allows for local scaling of the sequence on the time axis. Even if the user's operation rhythm changes, such as a longer interval due to answering a phone call, the algorithm can still identify the inherent deformation similarity between two state switching sequences, rather than relying solely on strict alignment within a fixed time window. The resulting operation coupling distance is directly derived from the temporal patterns contained in the user's historical operation data, abandoning the dependence on fixed triggering rules. This allows the discovered strongly linked device combinations to adapt to different users' operating habits, capturing subtle and personalized linkage patterns such as "turning on the range hood at low speed a little longer after turning on the stove" or "starting the countdown timer just before the oven finishes preheating," thus improving the accuracy of linkage triggering timing. The actual response delay time of each device after each pre-start command is issued is fed back to the dynamic graph. This delay data is added as a new element to the state switching time difference sequence of the corresponding device pair. The earliest element in the sequence is removed to maintain a fixed sequence length. Then, the operation coupling distance and linkage correlation weight are recalculated based on the updated sequence, making the entire linkage management parameters form a continuously running online adaptive closed loop. When the response delay of a device in the kitchen gradually increases due to the aging of its internal relays caused by long-term use, this change will be gradually absorbed into the updated operation coupling distance through multiple feedbacks, causing the coupling distance value to tend to increase. Consequently, the linkage correlation weight will be weakened, and the subsequent pre-start timing will be delayed according to the new weight, avoiding invalid waiting or repeated starts caused by issuing commands too early according to the original parameters. This mechanism makes linkage management no longer dependent on a static threshold calibrated once, but continuously tracks the drift of actual physical characteristics throughout the entire life cycle of the device, maintaining a dynamic match between the linkage strategy and the actual device response capability. It can maintain the coordination and reliability of multi-device collaboration in the smart kitchen without manual intervention. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart of a smart kitchen multi-device linkage management method based on artificial intelligence; Figure 2This is a flowchart for calculating the coupling distance of kitchen electrical equipment operations; Figure 3 This is a flowchart of the screening of strongly linked equipment combinations and the calculation of initial linkage correlation weights; Figure 4 This is a flowchart of the device pre-startup process based on user operation behavior recognition; Figure 5 This is a graph showing the distribution and fitting analysis of the operational coupling distance and cosine similarity. Figure 6 This is a schematic diagram showing the changes in the operation coupling distance sorting curve and the preset tightness threshold; Figure 7 This is a schematic diagram showing the matching of real-time 3D motion trajectory and device operation gesture template; Figure 8 It is a curve showing the change in response delay due to the coupling distance of multiple devices in operation. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 This invention provides a smart kitchen multi-device linkage management method based on artificial intelligence, comprising the following steps: collecting real-time operating status parameters and user operation timing data of multiple electrical devices in the kitchen, and constructing a dynamic graph with devices as nodes and operating status as attribute values; for each pair of device nodes in the dynamic graph, extracting the state switching time difference sequence and user operation sequence chain within a continuous time window, and calculating the operation coupling distance between the two devices using a dynamic time warping algorithm; based on the operation coupling distance and a preset tightness threshold, selecting device combinations with strong linkage relationships, and generating initial linkage correlation weights for each strongly linked device combination; monitoring the user operation behavior currently being performed in the kitchen, matching the user operation behavior with the generated strongly linked device combinations, and triggering a pre-start command for the non-started device in the corresponding device combination based on the linkage correlation weight when a match is successful; obtaining the actual response delay time of each device after the pre-start command is issued, and feeding the actual response delay time back to the dynamic graph to update the operation coupling distance and linkage correlation weights.

[0020] Example 1:

[0021] See Figure 5In the figure, the horizontal axis represents the operational coupling distance between device pairs, and the vertical axis represents the cosine similarity of the corresponding device pairs. In the legend, hollow circles represent data points of strongly linked device combinations, and hollow triangles represent data points of non-strongly linked device combinations; solid lines are the fitting curves of strongly linked device combinations, and dashed lines are the fitting curves of non-strongly linked device combinations.

[0022] As can be seen from the figure, the operational coupling distance of strongly linked device combinations is mainly concentrated in the range of 0.1 to 1.2, with corresponding high cosine similarity, mostly between 0.5 and 0.8. Moreover, the cosine similarity shows a significant decreasing trend as the operational coupling distance decreases, and the slope of the fitted line is negative, indicating that the smaller the operational coupling distance, the more similar the final representations of the device nodes are, reflecting strong operational timing synchronization and feature representation consistency.

[0023] In contrast, the operational coupling distance of non-strongly linked device combinations is significantly larger, mainly ranging from 0.6 to 3.3, with correspondingly lower cosine similarities, concentrated between 0.05 and 0.45. Furthermore, the cosine similarity shows a slight upward trend with increasing operational coupling distance, and the slope of the fitted line is positive. This trend indicates poor time synchronization in non-strongly linked device combinations, resulting in lower similarity in device node feature representations, which aligns with the threshold settings for operational coupling distance and cosine similarity used in determining strong linkage relationships.

[0024] Overall, the two categories of device combinations in the figure show a clear separation in the distribution of operational coupling distance and cosine similarity. Device combinations with strong linkages cluster in the region of low coupling distance and high similarity, while device combinations with weak linkages cluster in the region of high coupling distance and low similarity. This result verifies the effectiveness of the method in Example 1, which calculates operational coupling distance based on the dynamic time warping algorithm and combines it with the cosine similarity of the final node representation to determine strong linkages. This provides data support for subsequent screening of strongly linked device combinations and calculation of linkage correlation weights.

[0025] Example 2:

[0026] In specific implementation, please refer to Figure 2The system collects real-time operating status parameters and user operation timing data of multiple electrical devices in the kitchen through non-intrusive load monitoring terminals deployed on the power supply lines of each device. These terminals are installed at the main incoming line of the kitchen distribution box and at the outgoing lines of each branch circuit, with each branch circuit corresponding to one electrical device. The non-intrusive load monitoring terminal includes a voltage sampling module, a current sampling module, an analog-to-digital converter (ADC), and a signal processing module. The voltage sampling module converts the AC voltage signal into a low-voltage analog signal using a resistor divider network. The current sampling module converts the AC current signal into a low-voltage analog signal using a Rogowski coil or Hall effect sensor. The ADC synchronously acquires the low-voltage analog signals output from the voltage and current sampling modules at a sampling frequency of 64 kHz and converts them into digital signals. The signal processing module receives the digital signal output from the ADC and extracts voltage transient characteristics and current waveform distortion points.

[0027] The voltage transient feature extraction process is as follows: The signal processing module continuously monitors the effective value of the voltage waveform. When it detects that the effective voltage value drops by more than 10% of the rated voltage within three consecutive cycles, a voltage transient event is determined to have occurred. The signal processing module records the amplitude drop depth, recovery time, and oscillation frequency during the voltage transient event as voltage transient features. The amplitude drop depth is the ratio of the minimum effective voltage value to the rated voltage during the voltage transient event; the recovery time is the time it takes for the effective voltage value to recover from the transient minimum point to 90% of the rated voltage; and the oscillation frequency is the dominant frequency of the high-frequency oscillation component superimposed on the voltage waveform during the voltage transient event. The current waveform distortion point extraction process is as follows: The signal processing module calculates the waveform slope of the current signal using a sliding window method. The sliding window length is five sampling points, and adjacent sliding windows overlap by four sampling points. The linear regression slope of the current signal within each sliding window is calculated. When the absolute value of the slope change between adjacent sliding windows exceeds a preset slope change threshold, the starting sampling point of the corresponding sliding window is determined to be the current waveform distortion point. The preset slope abrupt change threshold is set to one-tenth of the rated peak value of the current signal. The signal processing module simultaneously extracts the peak amplitude and harmonic content change rate corresponding to the current waveform distortion point. The peak amplitude is the ratio of the instantaneous current value at the current waveform distortion point to the rated peak value of the current, and the harmonic content change rate is the difference between the total harmonic distortion rate of the two cycles before and after the current waveform distortion point.

[0028] The voltage transient features and current waveform distortion points are input into a pre-trained device type identifier, which outputs the device identifier that caused the state switch at the current moment, as well as the power step values ​​before and after the switch. The device type identifier adopts a one-dimensional convolutional neural network architecture, which includes an input layer, three convolutional layers, two pooling layers, one global average pooling layer, one fully connected layer, and one output layer. The input layer receives a fixed-length feature vector, which is composed of voltage transient features and current waveform distortion points. The voltage transient features include three scalar values: amplitude drop depth, recovery time, and oscillation frequency. The current waveform distortion point features include two scalar values: peak amplitude and harmonic content change rate. These five scalar values ​​are arranged in order to form a feature vector of length 5. The input layer copies and expands the feature vector into a multi-channel format to accommodate the convolution operation. The first convolutional layer contains 32 one-dimensional convolutional kernels, each with a size of 2 and a stride of 1, using the ReLU activation function. The first pooling layer uses max pooling with a pooling window size of 1 and a stride of 1. The second convolutional layer contains 64 one-dimensional convolutional kernels, each with a size of 1 and a stride of 1, using the ReLU activation function. The second pooling layer employs max pooling with a pooling window size of 1 and a stride of 1. The third convolutional layer contains 128 one-dimensional convolutional kernels, each with a size of 1 and a stride of 1, using the ReLU activation function. The global average pooling layer averages the output feature map of the third convolutional layer over time, resulting in a 128-dimensional feature vector. The fully connected layer contains 128 neurons, using the ReLU activation function, and is fully connected to the global average pooling layer. The output layer has two branches: the first branch is for device identification classification, with an output dimension equal to the total number of electrical devices in the kitchen, using the Softmax activation function, and outputting the predicted probability for each device category; the second branch is for power step value regression, with an output dimension of 2, corresponding to the power step value before and after the switch, respectively, using a linear activation function.

[0029] The training steps for the device type identifier are as follows: Voltage transient characteristics and current waveform distortion points of each device are collected from non-intrusive load monitoring terminals within a historical time period. Each record is manually labeled with its corresponding actual device identifier and the actual power step value before and after the switch. The labeled dataset is divided into training, validation, and test sets in an 8:1:1 ratio. A joint loss function is used for optimization during training; this function is a weighted sum of classification and regression losses. Cross-entropy loss is used for classification, and mean squared error loss is used for regression. The weight of the classification loss is set to 1.0, and the weight of the regression loss is set to 0.5. The Adam optimizer is selected, with an initial learning rate of 0.001, a batch size of 32, and 200 training epochs. Early stopping is performed when the validation set loss does not decrease for 10 consecutive epochs. After training, the device identifier classification accuracy and the mean absolute error of the power step value regression are evaluated using the test set. When the device identifier classification accuracy exceeds 95% and the mean absolute error is less than 5% of the rated power, the model parameters of the device type identifier are saved. During the online usage phase, the device type identifier receives the voltage transient characteristics and current waveform distortion points extracted in real time by the non-intrusive load monitoring terminal, outputs the device identifier with the highest probability in the device identifier classification branch as the device identifier that will switch states at the current moment, and outputs the power step value before and after the switch in the power step value regression branch.

[0030] Based on the system clock, each state transition event is timestamped. The system clock is provided by the kitchen main controller, and the timestamp uses the Unix timestamp format with millisecond precision. When the device type identifier outputs a device identifier, the main controller immediately reads the current time of the system clock, uses the read time value as a timestamp, and binds it with the device identifier and power step value to form a state transition event record. The main controller stores each state transition event record in ascending order of time into the user operation timing data queue. The user operation timing data queue is a fixed-length circular queue with a queue length of 10,000 records. When a new record arrives, if the queue is full, the record with the earliest timestamp at the head of the queue is removed, ensuring that the user operation timing data queue always stores the most recently occurred state transition events.

[0031] The process of extracting the state transition time difference sequence and user operation sequence chain within a continuous time window is as follows: A sliding time window with the current time as the endpoint and a fixed duration as the interval length is extracted from the user operation time sequence data queue. The fixed duration is set to 30 minutes based on the typical operation interval of kitchen equipment. All state transition event records within the sliding time window are obtained, and the device identifier and timestamp are extracted from each record. Within the sliding time window, for any pair of device nodes, the time difference between the state transition of device A and the state transition of device B within a preset time interval is recorded. The preset time interval is set to 300 seconds. Specifically, the recording method is as follows: traverse each state transition event of device A within the sliding time window, obtain the timestamp of the device A state transition event, search within the sliding time window for all state transition events of device B whose timestamps are later than the timestamp of the device A state transition event and whose time difference does not exceed 300 seconds, calculate the difference between the timestamps of the device B state transition events and the timestamps of the device A state transition events for each pair of found device B events, and arrange all time differences in the order of their generation to obtain the state transition time difference sequence. If a device pair has multiple time difference values ​​that meet the conditions within the sliding time window, all of them are included in the state transition time difference sequence in ascending time order. Each element in the state transition time difference sequence is a non-negative real number, with the unit being seconds.

[0032] Simultaneously, the direction indicator of the user operation sequence chain is counted within the sliding time window. The frequency with which device A's operation event precedes device B's operation event, and the frequency with which device B's operation event precedes device A's operation event, are counted. The counting method is as follows: compare the timestamps of two consecutive operation events of device A and device B within the sliding time window. If the timestamp of device A's operation event is earlier than the timestamp of device B's operation event, and the difference between the two timestamps does not exceed 300 seconds, increment the counter for the frequency of device A preceding device B; if the timestamp of device B's operation event is earlier than the timestamp of device A's operation event, and the difference between the two timestamps does not exceed 300 seconds, increment the counter for the frequency of device B preceding device A. When the frequency of device A preceding device B is greater than the frequency of device B preceding device A, the direction indicator of the user operation sequence chain is set from device A to device B. When the frequency of device B preceding device A is greater than the frequency of device A preceding device B, the direction indicator of the user operation sequence chain is set from device B to device A. When two frequencies are equal, the direction identifier of the user operation sequence chain is set to no direction.

[0033] The method for calculating the operational coupling distance between two devices using the dynamic time warp algorithm is as follows: the state transition time difference sequence of device A is used as the first sequence, and the state transition time difference sequence of device B is used as the second sequence. Let the first sequence be... The second sequence is , Include One element, Include One element, and All are positive integers. Construct the first sequence. With the second sequence Distance matrix between all pairs of points Distance matrix The dimension is , of which Line 1 The elements of the column are calculated using the following formula:

[0034] in, Represents the first sequence The One element, The value is to Integers between; Indicates the second sequence The One element, The value is to Integers between; Distance matrix The Middle Line 1 The Euclidean distance value of the column; This indicates the absolute value operation.

[0035] In the distance matrix In the distance matrix The starting point To the finish line Search for a curved path with the minimum cumulative distance. Curved path It consists of several path points, each path point Corresponding distance matrix A location index, path point number The value ranges from Starting to increase. Curved path. Boundary condition satisfied: First path point The last waypoint , For curved paths Total number of path points included. Curved path. Satisfying continuity and monotonicity constraints: for two adjacent path points and , and The value can only be 0. or And both cannot be simultaneously That is, the curved path in the distance matrix Each movement in the game can only be in one of three directions: horizontal to the right, vertical down, or diagonally to the lower right. (Cumulative distance matrix) The cumulative distance matrix is ​​used to record the minimum cumulative distance to each location. Dimensions and distance matrix Same, cumulative distance matrix The Middle Line 1 Column elements Calculated using dynamic programming recursion:

[0036] Among them, when and hour, .when and hour, .when and hour, According to the recursive formula from Start by calculating the cumulative distance matrix sequentially. The values ​​of all elements in the array are used to obtain the final cumulative distance to the destination. The cumulative distance to the destination This represents the operational coupling distance between device A and device B. The operational coupling distance is a non-negative real number. The smaller the operational coupling distance, the higher the similarity between the state transition time difference sequence of device A and the state transition time difference sequence of device B, that is, the more synchronous the state transitions of device A and device B are in time.

[0037] Example 3:

[0038] In specific implementation, please refer to Figure 3 The process of identifying device pairs with strong interrelationships begins by obtaining the operational coupling distances between all electrical devices in the kitchen. These operational coupling distances are calculated using a dynamic time warping algorithm, and the operational coupling distances of all device pairs constitute a set. All operational coupling distance values ​​in this set are then sorted in ascending order using a quicksort algorithm. The sorted result is an ordered list, where each element contains a device pair identifier and an operational coupling distance value. The head of the ordered list stores the device pair with the smallest operational coupling distance.

[0039] Device pairs are selected sequentially from the sorted list of operational coupling distances, traversing from the head to the tail of the list. For each device pair, the operational coupling distance is compared to a preset tightness threshold. If the operational coupling distance is less than the preset tightness threshold, the device pair is marked as a strongly linked device combination. Strongly linked device combinations are stored as sets, each entry recording the device identifiers of the two devices and their corresponding operational coupling distances. When the operational coupling distance is greater than or equal to the preset tightness threshold, the traversal stops, and subsequent device pairs in the sorted list are no longer selected, because the operational coupling distance list is already sorted in ascending order, and the operational coupling distances of subsequent device pairs are all no less than the operational coupling distance of the current device pair, thus failing to meet the condition of being less than the preset tightness threshold.

[0040] After filtering, the set of marked strongly linked device combinations is processed to identify cases where the same device appears in multiple strongly linked device combinations. Each record in the set of strongly linked device combinations is traversed, and a mapping table is established between device identifiers and their corresponding strongly linked device combinations. The keys in the mapping table are the device identifiers, and the values ​​are lists of strongly linked device combinations containing that device identifier. For device identifiers in the mapping table whose list length is greater than 1, it indicates that the device exists in multiple strongly linked device combinations simultaneously.

[0041] For devices existing in multiple strongly linked device combinations, an initial linkage correlation weight is assigned to the device in different combinations based on the operational coupling distance. The initial linkage correlation weight is inversely proportional to the operational coupling distance; the smaller the operational coupling distance in a strongly linked device combination, the greater the initial linkage correlation weight of the device in that combination. The initial linkage correlation weight is calculated as follows: for each strongly linked device combination containing a given device, the operational coupling distance corresponding to that strongly linked device combination is obtained, and the initial linkage correlation weight is set as the reciprocal of the operational coupling distance, i.e., the weight value is 1 divided by the operational coupling distance. When multiple combinations in the set of strongly linked device combinations contain the same device, the device obtains its own initial linkage correlation weight in different combinations. These initial linkage correlation weights are not normalized and maintain their original numerical relationship with the operational coupling distance.

[0042] For devices that appear only in a single strongly linked device group, the initial linkage correlation weight is calculated in the same way as described above, i.e., the reciprocal of the operational coupling distance. In this way, each device in every strongly linked device group receives an initial linkage correlation weight.

[0043] The preset tightness threshold is dynamically adjusted based on the total number of devices in the kitchen and the average operational coupling distance. The total number of devices in the kitchen is obtained by counting the number of nodes in the dynamic graph, and the total number of devices is a positive integer. The average operational coupling distance is obtained by summing all operational coupling distance values ​​in the operational coupling distance set and dividing by the total number of elements in the operational coupling distance set, and the average operational coupling distance is a non-negative real number. The dynamic adjustment of the preset tightness threshold follows the formula below:

[0044] in, Indicates the preset density threshold; The first adjustment coefficient is defined as 0.01 to 0.5. The first adjustment coefficient is determined through offline calibration during the system initialization phase. The calibration method is to traverse the candidate values ​​of the first adjustment coefficient in a typical kitchen scenario and select the candidate value that makes the number of strongly linked device combinations match the actual user's operating habits the highest as the first adjustment coefficient. This represents the total number of equipment in the kitchen, i.e., the number of equipment nodes in the dynamic graph; This represents the average operation coupling distance, which is the arithmetic mean of all operation coupling distance values ​​in the current set of operation coupling distances. This represents the second adjustment coefficient, which ranges from 1 to 20. The second adjustment coefficient is also determined through offline calibration. The calibration method involves adjusting the value of the second adjustment coefficient under different total number of devices, so that the preset tightness threshold can stably screen out device pairs with obvious synchronous operation characteristics, without excessively relaxing or tightening the screening conditions due to the increase in the total number of devices.

[0045] In average operating coupling distance When there is no change, the preset density threshold According to the total number of kitchen equipment The changes are proportional to the total number of kitchen equipment. Preset density threshold when increasing The total number of kitchen equipment will be adjusted accordingly. Preset density threshold when reducing The corresponding reduction will be made. (This refers to the total number of kitchen appliances.) Keeping the average operating coupling distance constant When the preset tightness threshold changes due to changes in user operation mode, Based on average operating coupling distance With the second adjustment coefficient The ratio is scaled proportionally. Preset density threshold. The update trigger condition is: the total number of devices in the kitchen. Changes or average operating coupling distance The change exceeded 10%. After an update is triggered, the preset tightness threshold is recalculated by substituting the total number of kitchen appliances and the average operational coupling distance at the time of the update into the formula. New preset density threshold Replace the previous preset tightness threshold and re-execute the screening process for strongly linked device combinations, including sorting by operational coupling distance, threshold comparison, device pair marking, and initial linkage correlation weight calculation.

[0046] See Figure 6 In the graph, the horizontal axis represents the serial numbers of all equipment pairs in the kitchen, totaling approximately 500 pairs, and the vertical axis represents the operational coupling distance value of the corresponding equipment pairs. The solid curve is a sorting curve of the operational coupling distances of all equipment pairs in ascending order, reflecting the distribution trend of the operational coupling distances. In the initial stage of the curve, the operational coupling distances are relatively low, approximately between 0.1 and 1.0, showing a slow upward trend; in the middle stage, the curve gradually flattens out, with the operational coupling distances concentrated in the range of 1.0 to 1.5; later, as the serial number of the equipment pair increases, the operational coupling distance rises rapidly, especially after the serial number exceeds 400, the curve sharply increases to over 8.5, indicating that most equipment pairs have large operational coupling distances and weak linkage relationships.

[0047] The dashed line in the diagram represents the current preset tightness threshold, which is approximately 1.2 and serves as the screening threshold for operational coupling distance. The intersection of the solid curve and the dashed line is located at approximately 280 in the device pair sequence number. This indicates that the operational coupling distance of device pairs with sequence numbers less than 280 is less than the current tightness threshold, and they are identified as candidate combinations of devices with strong linkage relationships.

[0048] The dotted line in the diagram represents the updated preset tightness threshold, which is approximately 1.8, higher than the current tightness threshold. The updated threshold allows more device pairs with operational coupling distances less than the threshold to be included in the strongly linked device group. The device pair number corresponding to the intersection point is approximately 320, an increase of about 40 device pairs compared to the number corresponding to the current threshold.

[0049] Overall, the sorting curve exhibits a three-stage characteristic: the low coupling distance stage corresponds to tightly linked equipment pairs, suitable as a basis for screening strong linkage combinations; the moderate middle stage is a transitional zone for operational coupling distance; and the high coupling distance stage corresponds to loosely linked equipment pairs. The tightness threshold is dynamically adjusted (from the current 1.2 to the updated 1.5) to adapt to changes in the number of equipment in the kitchen and user operation modes, so as to reasonably expand or shrink the scale of strongly linked equipment combinations and improve the flexibility and accuracy of equipment linkage management.

[0050] Example 4:

[0051] In specific implementation, please refer to Figure 4Real-time monitoring of user actions in the kitchen is achieved through the collaborative work of a motion sensor array and an image acquisition unit. The motion sensor array consists of three sets of infrared depth sensors and three sets of ultrasonic sensors. The infrared depth sensors and ultrasonic sensors are installed in pairs on the ceiling above the kitchen countertop, arranged in a triangular pattern to cover the entire countertop operating area. Each infrared depth sensor uses the time-of-flight principle to measure the distance from the sensor to objects on the countertop, while each ultrasonic sensor measures distance by emitting 40 kHz ultrasonic pulses and receiving the echoes. The motion sensor array acquires distance data at a synchronous frequency of 30 frames per second, with the measurements from the three infrared depth sensors and three ultrasonic sensors at the same moment constituting one frame of distance data. The image acquisition unit is a 1920x1080 resolution high-definition camera, mounted directly above the kitchen countertop with its lens pointing vertically downwards to capture color images at a frequency of 30 frames per second.

[0052] Data acquired by the motion sensor array and the image acquisition unit are fused using Kalman filtering to generate a 3D coordinate sequence of hand keypoints. The state vector of the Kalman filter contains the position coordinates and velocity components of the hand keypoints in 3D space, and the state vector has a dimension of 6. The observation vector of the Kalman filter consists of two parts: one part is the 3D position of the hand calculated by triangulation from the distance measurement values ​​of the motion sensor array, and the other part is the 3D position calculated by combining the 2D coordinates of the hand in the image plane identified by the hand detection model from the color image acquired by the image acquisition unit with the intrinsic and extrinsic parameters of the high-definition camera. The prediction step of the Kalman filter predicts the state vector for the next moment based on the velocity components of the hand movement, and the update step corrects the prediction results using the observation vector. After fusion, a smooth 3D coordinate sequence of hand keypoints is output. The 3D coordinate sequence takes the lower left corner of the kitchen countertop as the origin, with the X and Y axes lying in the countertop plane and the Z axis perpendicular to the countertop and upwards.

[0053] The three-dimensional motion trajectory is compared with pre-stored device operation gesture templates to identify the device type the user intends to operate on. The pre-stored gesture templates include an arc-shaped trajectory template for rotating knobs, a vertical downward trajectory template for pressing buttons, and a straight-line trajectory template for swiping touches. The arc-shaped trajectory template for rotating knobs is defined as an arc segment in which the hand moves clockwise or counterclockwise on the table surface. The radius of the arc is set according to the actual knob size, and the trajectory consists of a series of three-dimensional coordinate points, with the Z-axis coordinate remaining essentially constant. The vertical downward trajectory template for pressing buttons is defined as the hand moving a certain distance downward along the Z-axis from a certain height above the table surface and then stopping at a lower height. The X-axis and Y-axis coordinates change by less than 5 mm during the pressing process. The straight-line trajectory template for swiping touches is defined as the hand sliding a straight line a certain distance on the table surface in a certain direction, with the Z-axis coordinate remaining essentially constant. These gesture templates were obtained by offline collection of demonstration operation trajectories from multiple users and averaging the results. Each gesture template is bound to a specific device type.

[0054] The dynamic time warping algorithm is used to calculate the trajectory similarity between the real-time 3D motion trajectory and each gesture template. Let the real-time 3D motion trajectory be a sequence. ,Include There are several trajectory points, each a three-dimensional coordinate vector. Let the gesture template be a sequence. ,Include There are several trajectory points, each of which is also a three-dimensional coordinate vector. Construct a sequence. with sequence The Euclidean distance matrix between all pairs of trajectory points is given by dimension 1. Search the distance matrix for the normalized path with the minimum cumulative distance. The starting point of the normalized path is... The destination is The path movement step size is limited to horizontal, vertical, and diagonal directions. The cumulative distance to the endpoint of the regularized path is used as the trajectory similarity metric. The smaller the trajectory similarity metric, the higher the degree of matching between the real-time 3D motion trajectory and the gesture template. The trajectory similarity metric between the real-time 3D motion trajectory and each gesture template is calculated. The gesture template with the smallest trajectory similarity metric is selected. When the smallest trajectory similarity metric is lower than a preset similarity threshold, it is determined that the device the user intends to operate on is the device type bound to that gesture template. When the smallest trajectory similarity metric is greater than or equal to the preset similarity threshold, it is determined that no clear operation intention has been detected, and no device identifier is output.

[0055] When a user's intent to operate a device is detected, the identified device identifier and the start time of the operation are recorded as the user's current action. The start time is the timestamp corresponding to the first trajectory point in the real-time 3D motion trajectory, which is added synchronously during the Kalman filter output with millisecond precision. The device identifier and the start time of the operation constitute a user action record, which is then sent to the matching module for further processing.

[0056] The user's actions are matched against the generated strongly linked device combinations. The device identifier from the current user action record is used as the first device identifier. The strongly linked device combination set is generated and maintained by the aforementioned strongly linked device combination filtering process. Each record in the strongly linked device combination set contains two device identifiers and their corresponding linkage correlation weights. All records in the strongly linked device combination set are traversed, searching for a record containing the first device identifier. If no record containing the first device identifier is found in the strongly linked device combination set, the matching process ends, and no pre-start command is triggered.

[0057] If one or more records of strongly linked device combinations containing the first device identifier are found, the other device identifier besides the first device identifier in the strongly linked device combination record is extracted, and all extracted device identifiers form a set of devices to be linked. The current operating status of each device in the set of devices to be linked is checked. The current operating status of the device is obtained by querying the dynamic graph. Each device node in the dynamic graph stores an operating status attribute value, which includes two states: "started" and "not started". The operating status attribute value is updated according to the power step value monitored in real time by the non-intrusive load monitoring terminal. When the power step value is greater than 5% of the rated power of the device, the operating status attribute value is set to "started"; otherwise, it is set to "not started". Each device identifier in the set of devices to be linked is traversed, and the operating status attribute value of the corresponding device node in the dynamic graph is queried. Devices with an operating status attribute value of "not started" are designated as target pre-start devices, and multiple target pre-start devices constitute a target pre-start device set.

[0058] When multiple strongly linked device combinations containing the first device identifier are found, all device identifiers except the first device identifier in all strongly linked device combinations are first extracted and subjected to union processing. That is, the devices to be linked in multiple combinations are deduplicated and merged. The merged set of device identifiers is the final set of devices to be linked. Then, the final set of devices to be linked is sorted from high to low according to the linkage correlation weight. The linkage correlation weight is obtained as follows: for each device identifier in the set of devices to be linked, backtrack to the strongly linked device combination containing both the device identifier and the first device identifier. If the device identifier and the first device identifier appear in multiple strongly linked device combinations, the maximum value of the linkage correlation weight among the multiple strongly linked device combinations is taken as the sorting criterion for the device identifier. The sorting is in descending order, with the device with the highest linkage correlation weight at the beginning. The pre-start command is triggered sequentially according to the sorted device order.

[0059] When a pre-start command is triggered on an unstarted device in a corresponding device combination based on the linkage correlation weight, the linkage correlation weight of the target pre-start device in its respective strongly linked device combination is obtained. If the target pre-start device and the first device identifier exist in multiple strongly linked device combinations, the one with the highest linkage correlation weight is used as the currently used linkage correlation weight. The linkage correlation weight is compared with a preset pre-start confidence threshold. The preset pre-start confidence threshold is dynamically set based on the historical pre-start success rate. The initial value of the preset pre-start confidence threshold is 0.5, and it is subsequently adjusted based on the ratio of the actual number of successful pre-starts to the total number of pre-starts. The adjustment method is as follows: when the pre-start success rate is higher than 90%, the preset pre-start confidence threshold is decreased by 0.05; when the pre-start success rate is lower than 70%, the preset pre-start confidence threshold is increased by 0.05. The value range of the preset pre-start confidence threshold is limited to between 0.1 and 0.9.

[0060] When the linkage correlation weight is greater than or equal to the preset pre-start confidence threshold, a pre-start command is immediately sent to the target pre-start device. The pre-start command includes the target operating mode and power parameters of the target pre-start device. The target operating mode and power parameters are determined based on the most frequently occurring operating states of the target pre-start device in historical strong linkage scenarios: by statistically analyzing the operating mode and power parameters used by the target pre-start device when it is triggered to start under the condition that the strong linkage relationship between the target pre-start device and the device represented by the first device identifier is established from the user operation timing data, the operating mode and power parameters that occur most frequently are selected as the target operating mode and power parameters.

[0061] When the linkage correlation weight is less than the preset pre-start confidence threshold, the pre-start command is sent after a short waiting interval inversely proportional to the linkage correlation weight. The short waiting interval is calculated as follows:

[0062] in, Indicates a short waiting interval, in seconds; This indicates the preset baseline delay time, which is set to 2 seconds. The preset baseline delay time is based on the statistical median of the minimum tolerable waiting time between when a typical user operates the first device and when the second device actually needs to respond. This represents the linkage correlation weight, which is the reciprocal of the operation coupling distance, a non-negative real number output by the dynamic time warp algorithm. When the linkage correlation weight... When smaller, for example The value is 0.1, representing a short waiting interval. A value of 20 seconds indicates that the startup of the target pre-start device will be delayed by 20 seconds; when the linkage correlation weight When it is large, for example 2.0, short waiting interval The delay time is 1 second. After the delay time is reached, a pre-start command is triggered. The content of the pre-start command is the same as that of the immediate trigger command, which includes the target operating mode and power parameters.

[0063] See Figure 7 The figure illustrates the three-dimensional motion trajectory recognition process based on real-time monitoring of user operations in the kitchen in Example 4. The horizontal axis represents the X-coordinate in millimeters (mm), and the vertical axis represents the Y-coordinate, also in millimeters (mm). The coordinate system has its origin at the lower left corner of the kitchen countertop, and both the X and Y axes lie within the countertop plane. The curves in the figure represent different motion trajectories and gesture template trajectories, specifically including the real-time three-dimensional motion trajectory represented by solid lines, the arc trajectory template of the rotary knob represented by dashed lines, and the straight trajectory template of the sliding touch represented by dotted lines.

[0064] The real-time 3D motion trajectory exhibits a distinct arc shape, with trajectory points fluctuating between approximately 100 and 320 mm on the X-axis and 180 and 420 mm on the Y-axis. The trajectory is generally distributed clockwise, largely conforming to the arc-shaped trajectory template of the rotary knob. The trajectory range covered by the rotary knob's arc-shaped trajectory template highly overlaps with the real-time trajectory, indicating a high degree of match between the user's current operation and the rotary knob's arc-shaped trajectory. In contrast, the linear trajectory template for swipe touch is a straight line connecting coordinate points (approximately X=120, Y=400) to coordinate point (approximately X=270, Y=210), clearly different from the arc-shaped path of the real-time motion trajectory.

[0065] From the perspective of trajectory morphology and spatial distribution characteristics, the arc shape of the real-time three-dimensional motion trajectory significantly deviates from the linear trajectory template for sliding touch control, and has a high degree of coincidence with the arc trajectory template of a rotary knob, which satisfies the condition in Example 4 of calculating trajectory similarity through the dynamic time warping algorithm and determining the user's current intended operation on the device. This graph shows that at the current moment, the system successfully captures the user's operation intention corresponding to the rotary knob, records this operation as the user's current operation behavior, and provides input information for subsequent multi-device linkage management.

[0066] Example 5:

[0067] Referring to Figure 8 , this figure shows the trend curves of the operation coupling distance varying with the number of events for two groups of devices: microwave range hood and induction cooker range hood. The abscissa represents the cumulative number of state switching events, ranging from 0 to 500. The ordinate represents the calculated operation coupling distance under the corresponding events, with the unit of second and the range of approximately 0.5 to 4.5 seconds. The solid line in the figure represents the operation coupling distance between the microwave oven and the range hood, and the dashed line represents the operation coupling distance between the induction cooker and the range hood.

[0068] From the change trend of the curves, with the increase of the number of events, both curves show an obvious downward trend, indicating that as the system operation progresses, the operation coupling distance between devices gradually decreases, which reflects the enhancement of the operation timing synchronization between devices. Specifically, the operation coupling distance of the microwave oven-range hood combination decreases rapidly from the initial about 3.5 seconds to about 0.7 seconds, with a large decline range and the overall curve is relatively smooth, showing that the linkage relationship between the two devices is gradually strengthened. The operation coupling distance of the induction cooker-range hood combination starts at about 4.2 seconds, the descending speed is slightly slower, and finally approaches about 1 second. The curve fluctuates slightly but tends to be stable on the whole, indicating that the establishment process of the linkage relationship for this device pair is relatively slow and the linkage intensity is slightly lower than that of the microwave oven-range hood combination.

[0069] When the number of events is between about 200 and 400, the descending speed of the two curves slows down significantly and enters a stable stage, indicating that with the continuous update and feedback of the operation coupling distance, the system's modeling of the operation coupling relationship between devices tends to be stable. In addition, the overall operation coupling distance does not drop to zero, which conforms to the fact of the actual fluctuation of the state switching time difference sequence and the device response delay.

[0070] The data shown in this figure reflects the adaptive adjustment process in Embodiment 5, in which the actual response delay time of the device after the pre-start instruction is issued is obtained and fed back to the dynamic graph, thereby updating the operation coupling distance and the linkage correlation weight. With the accumulation of the number of feedbacks, the operation coupling distance between devices is continuously optimized, and the dynamically adjusted coupling distance can more accurately reflect the actual response characteristics of devices, which is conducive to improving the accuracy and real-time performance of system linkage management.

[0071] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A smart kitchen multi-device linkage management method based on artificial intelligence, characterized in that: Includes the following steps: Collect real-time operating status parameters and user operation time sequence data of multiple electrical devices in the kitchen, and construct a dynamic graph with devices as nodes and operating status as attribute values; For each pair of device nodes in the dynamic graph, extract the state switching time difference sequence and user operation sequence chain within a continuous time window, and calculate the operation coupling distance between the two devices using a dynamic time warping algorithm. Based on the operation coupling distance and the preset tightness threshold, device combinations with strong linkage relationships are selected, and an initial linkage correlation weight is generated for each strongly linked device combination. The user operation behavior being performed in the kitchen at the current moment is monitored in real time, and the user operation behavior is matched with the generated strongly linked device combinations. When the match is successful, the pre-start command of the non-started device in the corresponding device combination is triggered according to the linkage correlation weight. The actual response delay time of each device after the pre-start command is issued is obtained, and the actual response delay time is fed back to the dynamic graph to update the operation coupling distance and linkage correlation weight.

2. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 1, characterized in that, The dynamic graph adopts a heterogeneous graph neural network architecture, which uses device type as a node attribute and operation coupling distance as an edge feature for joint embedding learning.

3. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 1, characterized in that, The specific steps for collecting real-time operating status parameters and user operation timing data of multiple electrical devices in the kitchen are as follows: By deploying non-intrusive load monitoring terminals on the power supply lines of various kitchen devices, the voltage transient characteristics and current waveform distortion points of each device are captured synchronously. The voltage transient characteristics and current waveform distortion points are input into a pre-trained device type identifier, which outputs the device identifier that caused the state switch at the current moment and the power step value before and after the switch. Based on the system clock, each state transition event is timestamped, and the device operation actions triggered by the user are recorded in ascending order of time to form user operation timing data.

4. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 3, characterized in that, The extraction of the state transition time difference sequence and the user operation sequence chain within a continuous time window specifically involves: Extract a sliding time window from the user operation time sequence data, with the current time as the endpoint and a fixed duration as the interval length; Within the sliding time window, for any pair of device nodes, the time difference between the state transition of device B and the state transition of device A within a preset time interval after each state transition of device A is recorded. All time differences are arranged in chronological order to obtain the state transition time difference sequence. Meanwhile, within the sliding time window, the frequency of operation events of device A appearing before operation events of device B, and the frequency of operation events of device B appearing before operation events of device A are counted. The direction indicator of the user operation sequence chain is determined based on the relationship between the two frequencies.

5. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 4, characterized in that, The calculation of the operational coupling distance between the two devices using the dynamic time warp algorithm is specifically as follows: Using the state switching time difference sequence of device A as the first sequence and the state switching time difference sequence of device B as the second sequence, construct the distance matrix of all point pairs between the two sequences. In the distance matrix, a curved path with the minimum cumulative distance is searched from the starting point to the ending point. The curved path allows one point in one sequence to correspond to multiple consecutive points in another sequence. The cumulative distance at the end of the curved path is taken as the operational coupling distance between device A and device B. The smaller the operational coupling distance, the more synchronous the state switching of the two devices is in time.

6. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 5, characterized in that, The specific steps for selecting device combinations with strong interrelationships are as follows: Sort all calculated operational coupling distances between device pairs in ascending order of their values; From the sorted device pairs, select device pairs whose operation coupling distance is less than a preset tightness threshold, and mark the device pairs as strongly linked device combinations. When the same device appears in a combination of multiple strongly linked devices, the initial linkage correlation weight of the device in different combinations is assigned according to the value of the operation coupling distance. The initial linkage correlation weight is set inversely proportional to the operation coupling distance.

7. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 6, characterized in that, The preset tightness threshold is dynamically adjusted based on the total number of devices in the kitchen and the average operating coupling distance, and the threshold value is set in a positive correlation with the total number of devices.

8. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 6, characterized in that, The real-time monitoring of user actions being performed in the kitchen at the current moment specifically refers to: Using a motion sensor array and image acquisition unit, the three-dimensional motion trajectory of the user's hand on the kitchen countertop area and the action of touching the switch are captured. The three-dimensional motion trajectory is compared with the pre-stored operation gesture templates of each device to identify the type of device the user intends to operate. When the user's intention to operate a certain device is identified, the device identifier and operation start time of that device are recorded as the user operation behavior being performed at the current moment.

9. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 8, characterized in that, Matching the user's actions with the generated strongly linked devices specifically involves: Obtain the first device identifier corresponding to the current user operation behavior, traverse all strongly linked device combinations, and find the combination containing the first device identifier; If a strongly linked device combination containing the first device identifier is found, the identifiers of other devices in the combination except the first device are extracted as the set of devices to be linked. Check the current operating status of each device in the set of devices to be linked, and designate the devices that are not started as the target pre-start devices.

10. The method for multi-device linkage management of a smart kitchen based on artificial intelligence according to claim 9, characterized in that, When multiple strongly linked device combinations containing the first device identifier are found, the devices to be linked in all combinations are merged and sorted from high to low according to the linkage correlation weight, and the pre-start command is triggered in sequence.

Images