Old people home safety decision service system fused with multi-mode perception data

Abstract

The invention relates to the technical field of sensing data processing, and discloses an old people home safety decision service system fusing multi-mode sensing data, which comprises a data interaction analysis unit, a data change analysis unit, a data calculation unit, a safety analysis unit and a safety decision unit. The hidden association of the multi-modal indexes is accurately captured, the early degradation of the collaboration level can be perceived through the graph condensation degree and the graph equilibrium value, the collaboration degradation track can be used for analyzing the collaboration state, and the finally generated safety decision instruction can perform early warning in advance. According to the scheme, single-mode threshold dependence is broken through, hidden dangers hidden in data correlation evolution are recognized in advance, and the lagging of home safety decision of old people is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of perception data processing, in particular to an old people's home safety decision service system fusing multi-modal perception data. BACKGROUND

[0002] At present, the home safety of the old people is a key event of the society, and the current safety decision service system often analyzes and fuses multiple different modal perception data to determine whether to trigger an alarm, so as to ensure the home safety of the old people.

[0003] However, the above data processing method still has the following defects in actual application: at present, when the multi-modal data is processed and fused, it is often determined that it is a dominant risk and an alarm is given only when a single perception data breaks through a threshold value, but this is a confirmation mechanism after a danger occurs; In fact, many safety risks do not manifest as a threshold value of any single modal data in the early stage, but are hidden in the slow degradation or evolution of the correlation between multiple modal data, for example, the long-term decline between heart rate and daily activity, the abnormal increase of the frequency of getting up at night and the length of time in the toilet, and the existing data processing method lacks correlation analysis before the modal perception data reaches the threshold value, and cannot timely identify potential safety hazards, resulting in a certain lag in the service of the old people's home safety decision. SUMMARY

[0004] In view of the defects of the prior art, the present application provides an old people's home safety decision service system fusing multi-modal perception data, which solves the above problems.

[0005] The above technical purpose of the present application is realized by the following technical scheme: An old people's home safety decision service system fusing multi-modal perception data, comprising: A data interaction analysis unit is used to acquire environmental perception data in a room and physiological perception data of a target service object in real time, take the feature vector of each type of perception data after preprocessing as a modal node, calculate the phase lock value between any two modal nodes in a continuous sliding analysis window, take it as an interaction edge weight connecting the two modal nodes, and generate a dynamic interaction graph based on all modal nodes and interaction edge weights, and the target service object is an old person; A data change analysis unit is used to deeply analyze the dynamic change of the coordination level in the dynamic interaction graph, and obtain the graph condensation degree and graph balance value of each time window; A data calculation unit is used to acquire the time sequence value of the graph condensation degree and the graph balance value in a sliding analysis window in real time, form a bivariate time series, calculate the bivariate time series, and generate a coordination degradation trajectory. a safety analysis unit configured to analyze the cooperative degeneration trajectory to generate a trajectory instability index; a safety decision unit configured to calculate the trajectory instability index to generate a risk threshold, compare the trajectory instability index with the risk threshold, and generate a safety decision instruction.

[0006] Further, the feature vector of each type of perception data after preprocessing is taken as a modal node, and within a continuously sliding analysis window, the phase locking value between any two modal nodes is calculated as the interaction edge weight connecting the two modal nodes, and based on all modal nodes and interaction edge weights, a dynamic interaction graph is generated, including: After feature extraction of each type of perception data after preprocessing, a feature vector is formed, and the feature vector is analyzed to generate a feature activity degree; Based on the feature activity degrees of any two modal nodes, the synchronization of any two modal nodes is calculated to obtain a phase locking value.

[0007] Further, the feature vector of each type of perception data after preprocessing is taken as a modal node, and within a continuously sliding analysis window, the phase locking value between any two modal nodes is calculated as the interaction edge weight connecting the two modal nodes, and based on all modal nodes and interaction edge weights, a dynamic interaction graph is generated, further including: An initial weighted undirected graph is constructed with the modal nodes as vertices and the phase locking values as interaction edge weights, the geometric mean of the edge weights of all triangular closed loops in the initial weighted undirected graph is calculated to obtain a structure condensation factor of the current window; The feature activity degree of each modal node is taken as a node attribute, and the structure condensation factor of the current window is taken as a global attribute, and they are embedded into the initial weighted undirected graph to generate a dynamic interaction graph.

[0008] Further, the dynamic changes of the cooperative level in the dynamic interaction graph are analyzed in depth to obtain a graph condensation degree and a graph balance value of each time window, including: For the dynamic interaction graph, based on the phase locking value and the feature activity degree, a coupling resonance ratio is calculated, the coupling resonance ratio is regarded as a field strength distributed on the edges of the dynamic interaction graph, and through analysis of the potential energy diffusion and iteration of the field strength, a resonance field inhomogeneity is obtained; Based on the phase locking value and the coupling resonance ratio, the flowability between all pairs of modal nodes is calculated to generate a network flow bottleneck coefficient; Based on the coupling resonance ratio, the dynamic interaction graph is divided into a high resonance subgraph and a low resonance subgraph, and the interaction edge weights in the high resonance subgraph and the low resonance subgraph are analyzed respectively to obtain a substructure polarization coefficient.

[0009] Further, the dynamic changes of the coordination level in the dynamic interaction map are analyzed in depth to obtain the graph condensation degree and the graph equilibrium value of each time window, and also include: In the continuously sliding analysis window, the node set overlap degree and the edge weight structure similarity between the high resonance subgraph of the current window and the high resonance subgraph of the previous window are calculated to obtain the coordination core transition strength; The structure condensation factor, the resonance field unevenness and the network flow bottleneck coefficient are fused to generate the graph condensation degree of the current time window; The coupling resonance ratio, the substructure polarization coefficient and the coordination core transition strength are fused to generate the graph equilibrium value of the current time window.

[0010] Further, the time series values of the graph condensation degree and the graph equilibrium value are obtained in real time within the sliding analysis window to form a bivariate time series, and the bivariate time series is calculated to generate a coordination degradation trajectory, including: The coordination state corresponding to the graph condensation degree and the graph equilibrium value in the bivariate time series is identified, and at each time point, the included angle between the current graph condensation degree and the graph equilibrium value and the coordination state is calculated to obtain the coordination state offset angle; In the two-dimensional phase plane formed by the graph condensation degree and the graph equilibrium value, the data points are connected to form a trajectory, the change of the trajectory at each point is analyzed, and the local trajectory curvature inertia ratio is obtained.

[0011] Further, the time series values of the graph condensation degree and the graph equilibrium value are obtained in real time within the sliding analysis window to form a bivariate time series, and the bivariate time series is calculated to generate a coordination degradation trajectory, also including: The interaction between the graph condensation degree sequence and the graph equilibrium value sequence is analyzed to generate a bivariate lead-lag quantity; Within the analysis window, a plurality of short trajectory segments of fixed length are randomly intercepted, each short trajectory segment is linearly fitted, and the residual of all data points to the corresponding fitting line is calculated to generate a trajectory form stiffness; The coordination state offset angle, the local trajectory curvature inertia ratio, the bivariate lead-lag quantity and the trajectory form stiffness are fused to obtain the coordination degradation trajectory corresponding to the current analysis window.

[0012] Further, the coordination degradation trajectory is analyzed to generate a trajectory instability index, including: The time series change quantity of the coordination degradation trajectory is symbolized, and the trajectory sequence entropy is generated by analyzing the arrangement complexity of the symbol sequence on multiple time scales; The coordination degradation trajectory is decomposed to generate an abnormal fluctuation energy.

[0013] Further, the coordination degradation trajectory is analyzed to generate a trajectory instability index, also including: Identify the long-term trend turning point in the collaborative degradation trajectory and analyze it to generate trend inertia strength; Fusion of trajectory sequence variable entropy, abnormal fluctuation energy and trend inertia strength to obtain trajectory instability index.

[0014] Further, the trajectory instability index is calculated to generate a risk threshold, and the trajectory instability index is compared with the risk threshold to generate a safety decision instruction, including: Analyzing the trajectory instability index of multiple different windows to generate risk intrinsic potential energy; Synergistically calculating the risk intrinsic potential energy, abnormal fluctuation energy and trend inertia strength at the current time to obtain the risk threshold; Comparing the trajectory instability index with the risk threshold to generate a safety decision instruction.

[0015] In summary, the present application mainly has the following beneficial effects: The feature activity degree determines the dynamic change of the single mode index, and the phase lock value is combined to capture the synchronicity of the multi-modal node, and then the structure condensation factor is combined to generate a dynamic interaction atlas, realizing deep mining of the correlation characteristics of multi-modal data, and accurately capturing the implicit correlation between heart rate and step frequency, temperature and humidity, and blood oxygen and other indicators. The data change analysis unit fuses parameters such as coupling resonance ratio, network flow bottleneck coefficient and substructure polarization coefficient to generate graph condensation degree and graph balance value, which can determine the multi-modal network synergy strength, distribution uniformity and structure stability, and timely perceive early signs of index correlation degradation.

[0016] The bivariate time series calculated by the data calculation unit is fused by the cooperative state offset angle, local orbit curvature inertia ratio, bivariate traction lag and trajectory shape stiffness to generate a collaborative degradation trajectory, accurately understanding the degradation process of the cooperative state of the elderly and the environment, and the safety analysis unit generates a trajectory instability index by trajectory sequence variable entropy, abnormal fluctuation energy and trend inertia strength, realizing multi-dimensional judgment of trajectory change complexity, sudden abnormality and long-term trend stability. The safety decision instruction generated by the safety decision unit can identify potential hazards hidden in the evolution of multi-modal data correlation in advance, effectively solving the lag problem of safety decision of the elderly at home. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a schematic diagram of a multi-modal sensing data fusion-based safety decision service system for the elderly at home. DETAILED DESCRIPTION

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

[0019] refer to Figure 1 A home safety decision-making service system for the elderly that integrates multimodal sensing data includes: The data interaction analysis unit is used to acquire indoor environmental perception data and physiological perception data of the target service object in real time. The feature vector of each type of perception data after preprocessing is used as a modal node. Within a continuously sliding analysis window, the phase lock value between any two modal nodes is calculated and used as the interaction edge weight connecting the two modal nodes. Based on all modal nodes and interaction edge weights, a dynamic interaction map is generated. The target service object is the elderly. Environmental sensing data includes: temperature and humidity, oxygen content, smoke concentration, and gas concentration; Physiological sensory data include: heart rate, blood pressure, respiratory rate, body temperature, blood oxygen saturation, electromyographic signals, cadence, etc. in the elderly; The data change analysis unit is used to conduct in-depth analysis of the dynamic changes in the level of collaboration in the dynamic interaction graph, and to obtain the graph cohesion and graph equilibrium values ​​for each time window. The data calculation unit is used to acquire time series values ​​of graph cohesion and graph equilibrium in real time within a sliding analysis window, form a bivariate time series, calculate the bivariate time series, and generate a co-degradation trajectory. The safety analysis unit is used to analyze the cooperative degradation trajectory and generate a trajectory instability index; The safety decision unit is used to calculate the trajectory instability index, generate a risk threshold, compare the trajectory instability index with the risk threshold, and generate a safety decision instruction.

[0020] In one embodiment, the feature vectors of each type of preprocessed sensing data are used as modal nodes. Within a continuously sliding analysis window, the phase lock value between any two modal nodes is calculated and used as the interaction edge weight connecting the two modal nodes. Based on all modal nodes and interaction edge weights, a dynamic interaction graph is generated, including: After feature extraction for each type of preprocessed sensing data, a feature vector is formed. The feature vector is then analyzed to generate feature activity. Specifically, this includes: normalizing each sensing parameter in the environmental and physiological sensing data to the 0-1 interval to form a feature vector; within a short time window, calculating the root mean square value of the difference between consecutive data points in the feature vector and using it as the fluctuation intensity; performing linear fitting on the data points in the feature vector within the window, calculating the absolute value of its slope and normalizing it to the 0-1 interval to obtain the trend stability. The fluctuation intensity is multiplied by the trend stability to generate the characteristic activity. The characteristic activity is mainly used to quantify the dynamic change intensity and trend persistence of single-modal indicators, such as heart rate and gas concentration, to avoid ignoring the precursors of abnormal fluctuations due to a single indicator not reaching the threshold.

[0021] Based on the characteristic activity of any two modal nodes, the synchronization of any two modal nodes is calculated to obtain the phase lock value. Specifically, for the characteristic activity time series of any two modal nodes, within the analysis window, the proportion of the number of points with the same characteristic activity change at each time point of the two characteristic activity time series to the total number of points is calculated to obtain the direction synchronization rate. Among them, the characteristic activity change is of three types: increase, decrease or no change. Calculate the ratio sequence of the absolute values ​​of the change in activity in the two feature activity time series within the window. Take the logarithm of each ratio in the ratio sequence to the base 10, and calculate the arithmetic mean of these logarithms to obtain the strength coupling degree. The time of maximum activity of two feature activity time series is examined within an independent window. The absolute value of the difference between the two time points is calculated and converted into a negative exponential function with the natural constant e as the base, to obtain the time decay factor. The directional synchronization rate, intensity coupling degree, and time decay factor are multiplied together to obtain the phase lock value. The phase lock value mainly focuses on the correlation characteristics of multimodal sensing data in the home environment of the elderly, such as understanding the synchronization of heart rate and cadence, and the coupling strength of temperature and humidity and blood oxygen, because changes in such sensing data can facilitate the identification of potential risks to the home safety of the elderly.

[0022] In one embodiment, the feature vectors of each type of preprocessed sensing data are used as modal nodes. Within a continuously sliding analysis window, the phase lock value between any two modal nodes is calculated and used as the interaction edge weight connecting the two modal nodes. Based on all modal nodes and interaction edge weights, a dynamic interaction graph is generated. The method also includes: Using modal nodes as vertices and phase lock values ​​as interaction edge weights, an initial weighted undirected graph is constructed. The geometric mean of the edge weights of all triangular closed loops in the initial weighted undirected graph is calculated to obtain the structural cohesion factor of the current window. Specifically, this includes: constructing an initial weighted undirected graph based on all modal nodes and their pairwise phase lock values, where each modal node is a vertex of the initial weighted undirected graph, and each pair of vertices is connected by an edge, which is the interaction edge, and the weight of the interaction edge is the phase lock value; Identify all triangular substructures in the initial weighted undirected graph that consist of three interconnected vertices. For each triangular substructure, multiply the weights of its three interacting edges, and then take the cube root of the product to obtain the geometric mean of the edge weights of that triangular substructure. Calculate the arithmetic mean of the geometric mean of the edge weights of all triangular substructures to obtain the structural cohesion factor. The structural cohesion factor mainly reflects the local clustering tightness of the multimodal indicator network. For example, when the physical functions of the elderly are stable, the clustering tightness of heart rate, blood pressure, and blood oxygen is relatively high. If the structural cohesion factor continues to decline, it may mean that the synergistic regulatory ability between multiple physiological indicators has deteriorated, which is an early signal of accumulated health risks.

[0023] The feature activity of each modal node is used as a node attribute, and the structural cohesion factor of the current window is used as a global attribute. These are embedded together into the initial weighted undirected graph to generate a dynamic interaction graph. Specifically, this involves: using the feature activity of each modal node as its node weight and the structural cohesion factor of the current window as a global scaling factor; performing embedding calculations on each interaction edge in the initial weighted undirected graph: calculating the mean of the feature activity of the modal nodes at both ends of the interaction edge, multiplying this mean by the phase lock value of the interaction edge, and then multiplying by the global scaling factor to obtain the adjusted embedding edge weight; and constructing a dynamic interaction graph based on all modal nodes and the adjusted embedding edge weights. The dynamic interaction graph abstracts the physiological state of the elderly and the state of their home environment into a relational network. Through the dynamic updating of node weights and edge weights, the coordinated state of the elderly's body and environment at home can be confirmed in real time.

[0024] By constructing a dynamic interaction map, the dynamic change intensity and trend persistence of single-modal indicators are determined based on feature activity, avoiding the neglect of abnormal fluctuation precursors due to a single indicator not reaching the threshold. The phase lock value is calculated based on directional synchronization rate, intensity coupling degree, and temporal decay factor to accurately capture the synchronization and coupling strength between multimodal data. By identifying early signals of the degradation of the synergistic regulation ability of physiological indicators through structural cohesion factor, the existing ex-post confirmation mechanism that relies on single-modal threshold is broken through, realizing the timely identification of potential safety hazards hidden in the correlation evolution of multimodal data, and solving the lag in home safety decision-making services for the elderly.

[0025] In one embodiment, a detailed analysis of the dynamic changes in the synergy level in the dynamic interaction graph is performed to obtain the graph cohesion and graph equilibrium values ​​for each time window, including: For dynamic interaction graphs, the coupling resonance ratio is calculated based on the phase lock value and characteristic activity. The coupling resonance ratio is regarded as the field strength distributed on the edge of the dynamic interaction graph. By analyzing the potential energy diffusion and iteration of the field strength, the inhomogeneity of the resonance field is obtained. Specifically, for each interaction edge, the product of its phase lock value and the characteristic activity of the two modal nodes is divided by the sum of the characteristic activity of the two modal nodes to obtain the coupling resonance ratio of the interaction edge. The coupling resonance ratio is used as the initial field strength. In the dynamic interaction graph, the average of the initial field strengths of all adjacent interaction edges of each vertex is collected as its potential energy. The potential energy diffusion process is as follows: the new field strength of each interaction edge is updated to the average potential energy of the modal nodes at its two ends. This diffusion process is iterated twice. The standard deviation of the final field strength of all interaction edges after diffusion is calculated, and the standard deviation is divided by the average of all field strengths to obtain the resonance field non-uniformity. Among them, the increase of resonance field non-uniformity means that the correlation strength of some indicators is abnormally prominent. For example, the correlation between the frequency of getting up at night and the duration of toileting is sharply enhanced, or the correlation between heart rate and blood oxygen is suddenly weakened. This mainly reflects the sudden change in the physical condition of the elderly or the abnormality of the environment.

[0026] Based on the phase lock value and coupling resonance ratio, the flow between all modal node pairs is calculated to generate the network flow bottleneck coefficient. Specifically, this includes multiplying the phase lock value of each interaction edge by the coupling resonance ratio to obtain the effective flow of that interaction edge. For all modal node pairs in the dynamic interaction graph, assess their flow bottlenecks: calculate the harmonic mean of the effective flux of all interaction edges between the modal node pairs as the path flux baseline; find the minimum effective flux among all interaction edges between the modal node pairs and use it as the direct bottleneck strength. Calculate the ratio of direct bottleneck strength to path flux baseline; then calculate the mean of all modal node comparison values, and multiply the mean by the coefficient of variation of the effective flux values ​​of all interaction edges in the dynamic interaction graph to obtain the network flow bottleneck coefficient. The network flow bottleneck coefficient mainly reflects the degree of obstruction in information transmission between multimodal indicators. For example, for elderly people living at home, smooth information flow between various indicators is a reflection of stable physical condition and good environmental adaptability. If the network flow bottleneck coefficient increases, it may mean that a certain indicator has abnormal fluctuations and cannot be coordinated and regulated by other indicators, such as a sudden rise in blood pressure but the heart rate is not reflected in time. This can be used to identify potential risk signals. In this process, the standard deviation and arithmetic mean of the effective flux values ​​of all interaction edges in the dynamic interaction graph are calculated, and the coefficient of variation is obtained by dividing the standard deviation by the arithmetic mean.

[0027] Based on the coupling resonance ratio, the dynamic interaction graph is divided into high harmonic subgraphs and low harmonic subgraphs. The interaction edge weights in the high harmonic subgraphs and low harmonic subgraphs are analyzed respectively to obtain the substructure polarization coefficient. Specifically, the median of the coupling resonance ratio of all edges in the dynamic interaction graph is calculated as a threshold. Interaction edges with coupling resonance ratios higher than the threshold and their two end modal nodes are used to form high harmonic subgraphs, while the rest are used to form low harmonic subgraphs. The average phase lock value of all interaction edges within the high harmonic resonance oscillator (HHR) and low harmonic resonance oscillator (HNI) is calculated separately. The average phase lock value of the HHR is then divided by the average phase lock value of the HNI to obtain the lock value ratio. Next, the proportion of interaction edges in the HHR to the total number of interaction edges in the dynamic interaction graph is calculated. This proportion is multiplied by the lock value ratio to obtain the substructure polarization coefficient. The substructure polarization coefficient mainly reflects the degree of differentiation among indicators in the multimodal indicator network. For example, when the physical function of the elderly is normal, the correlation strength distribution among physiological indicators is relatively balanced, and the substructure polarization coefficient is low. If the substructure polarization coefficient increases, it may mean that the physical indicators of the elderly are abnormal, which is an early differentiation signal of health risk.

[0028] In one embodiment, a detailed analysis of the dynamic changes in the synergy level in the dynamic interaction graph is performed to obtain the graph cohesion and graph equilibrium values ​​for each time window, and the analysis also includes: In a continuously sliding analysis window, the node set overlap and edge weight structure similarity between the high harmonic oscillator graph of the current window and the high harmonic oscillator graph of the previous window are calculated to obtain the cooperative kernel transition strength. Specifically, this includes: obtaining the modal node sets of the high harmonic oscillator graphs of the current analysis window and the previous analysis window, calculating the number of intersection elements of the two modal node sets, calculating the number of union elements of the two modal node sets, and dividing the number of intersection elements by the number of union elements to obtain the node set overlap. The edge weight structure similarity is calculated as follows: Identify the interactive edges in two high harmonic oscillator graphs that are composed of the same two modal nodes and regard them as common edges; calculate the ratio of the phase lock values ​​of each common edge in the two analysis windows. If the ratio is greater than 1, take its reciprocal to obtain the weight fidelity of the common edge. If the ratio is less than or equal to 1, use the ratio directly as the weight fidelity of the common edge. Then calculate the mean of the weight fidelities of all common edges, which is the edge weight structure similarity. Multiplying the node set overlap by the edge weight structure similarity and then multiplying by the proportion of the number of shared edges to the total number of interactive edges in the hyperharmonic oscillator graph yields the cooperative kernel transition strength. The cooperative kernel transition strength mainly reflects the stability of highly correlated index clusters. In the home safety of the elderly, the stability of highly correlated clusters is a manifestation of the physical condition patterns of the elderly.

[0029] By fusing the structural cohesion factor, the resonant field inhomogeneity, and the network flow bottleneck coefficient, the graphical cohesion of the current time window is generated. Specifically, the structural cohesion factor is multiplied by the inverse of the resonant field inhomogeneity to obtain an intermediate value that measures the synergistic strength and distribution uniformity. The dynamic coefficient is obtained by dividing the median of all interactive edge phase lock values ​​in the current window by the mean of the medians in the past five consecutive windows. The graph cohesion of this time window is obtained by dividing the median by the square root of the network flow bottleneck coefficient and then multiplying it by the dynamic coefficient. Graph cohesion is suitable for comprehensively reflecting the collaborative strength, distribution uniformity, information flow efficiency and dynamic stability of multimodal index networks.

[0030] The coupling resonance ratio, substructure polarization coefficient, and cooperative nuclear transition intensity are fused to generate the graph equilibrium value for this time window. Specifically, this includes: calculating the Gini coefficient of the coupling resonance ratio of all edges in the graph to obtain the resonance distribution divergence; calculating the negative exponential function value of the substructure polarization coefficient with the natural constant e as the base to obtain the polarization attenuation factor; and multiplying the cooperative nuclear transition intensity by the resonance distribution divergence and then dividing by the polarization attenuation factor to obtain the equilibrium index. Calculate the moving average of this equilibrium index over the past three consecutive time windows. Calculate the difference between the current equilibrium index and this moving average. If the difference is positive, multiply the difference by 0.8 to obtain the adjusted difference; if the difference is negative, multiply the difference by 1.2 to obtain the adjusted difference; if the difference is 0, the adjusted difference is 0. Then add the moving average and the adjusted difference to obtain the graph equilibrium value for this time window. The graph equilibrium value mainly reflects the structural equilibrium and dynamic stability of the multimodal indicator network. Abnormal fluctuations in the graph equilibrium value indicate an imbalance in the coordination between the elderly and the environment. For example, sudden changes in environmental temperature and humidity can disrupt the balance of physiological indicators in the elderly, which is a key signal that potential risks have transformed from hidden to explicit.

[0031] By accurately capturing abnormal changes and state abrupt changes in the correlation strength of multimodal indicators through parameters such as coupling resonance ratio and resonance field inhomogeneity, and reflecting the degree of information transmission blockage through network flow bottleneck coefficient, early health risk signals of indicator differentiation are identified through substructure polarization coefficient, and the stability of highly correlated indicator clusters is monitored by cooperative kernel transition strength. Finally, the generated graph cohesion and graph equilibrium values ​​can comprehensively evaluate the cooperative strength, distribution uniformity and dynamic stability of multimodal networks, breaking through the existing ex-post confirmation mode that relies on single-modal thresholds, realizing the accurate identification of potential hidden dangers in the evolution of data correlation, and solving the problem of delayed decision-making for home safety of the elderly.

[0032] In one embodiment, within a sliding analysis window, time-series values ​​of graph cohesion and graph equilibrium are acquired in real time to form a bivariate time series. The bivariate time series is then calculated to generate a co-degradation trajectory, including: To identify the co-states corresponding to graph cohesion and graph equilibrium values ​​in a bivariate time series, at each time point, the angle between the current graph cohesion and graph equilibrium values ​​and the co-state is calculated to obtain the co-state offset angle. Specifically, in the bivariate time series, data points where both graph cohesion and graph equilibrium values ​​are at their maximum values ​​are taken as co-states. At each current time point, the percentage increments of the current graph cohesion value and graph equilibrium value relative to the graph cohesion value and graph equilibrium value corresponding to the co-state are calculated. The two percentage increments are treated as two coordinate components of a two-dimensional vector. Calculate the angle between the two-dimensional vector and the optimal growth direction defined by the positive directions of the two coordinate axes, where the optimal growth direction is the direction in which both percentage increments are positive; calculate the cosine of this angle, and normalize its range to between 0 and 90 degrees to obtain the synergistic state offset angle. The synergistic state offset angle reflects the degree of deviation between the current synergistic state of the elderly and the environment, because the ideal state of home safety for the elderly is stable physiological indicators and good environmental adaptability.

[0033] In a two-dimensional phase plane composed of graph cohesion and graph equilibrium values, data points are connected to form a trajectory. The changes in the trajectory at each point are analyzed to obtain the local trajectory curvature inertia ratio. Specifically, for the current point on the trajectory in the two-dimensional phase plane, it, along with its two preceding and two subsequent adjacent points, form a local five-point sequence. The directional changes between consecutive points in the local five-point sequence are calculated. Specifically, the coordinate difference between adjacent points is regarded as a displacement vector. The angle between the vector pointing from the previous point to the current point and then from the current point to the next point is calculated sequentially to obtain three consecutive angle values. The standard deviation of these three angle values ​​is then calculated as the directional fluctuation. The ratio of the Euclidean distance between the two farthest points in the local five-point sequence to the Euclidean distance between the two closest points is calculated as the span scaling ratio. The directional variability and the span scaling ratio are multiplied to obtain the local trajectory curvature inertia ratio. The local trajectory curvature inertia ratio is used to quantify the degree of local variability and inertial characteristics of the trajectory. An increase in its value means that the variability of the cooperative state is aggravated, which may correspond to the instability of the elderly's physical condition or frequent changes in the environment.

[0034] In one embodiment, within a sliding analysis window, the time-series values ​​of graph cohesion and graph equilibrium are acquired in real time to form a bivariate time series. The bivariate time series is then calculated to generate a co-degradation trajectory. The method further includes: The interaction between the graph cohesion sequence and the graph equilibrium value sequence is analyzed to generate a bivariate traction lag. Specifically, for the graph cohesion sequence, within a continuously sliding analysis window, if the graph cohesion value at the current moment is greater than 1.5 standard deviations of the mean of its past five consecutive time windows, then that moment is marked as a cohesion-driven event. For each cohesion-guided event, the behavior of the graph equilibrium value sequence is analyzed within a fixed response time window, where the fixed response time window consists of the last three consecutive sampling time points. Within the response time window, count the number of time points where the direction of change of the graph equilibrium value is consistent with the direction of change of the cohesion that triggered the event. Divide this number by the total length of the response time window (i.e., 3) to obtain the co-directional response rate of the cohesion-guided event; where the direction of change is either increase, decrease, or no change. Within the response time window, starting from the first time point, check to find the time point when the change in the graph equilibrium value first exceeds 0.6 times the average fluctuation of the graph equilibrium value sequence in the first ten time windows, and record the time difference between the time point when the cohesion-guided event occurs and the time difference is used as the effective delay step for the cohesion-guided event; if there is no time point that exceeds the limit, the effective delay step is 3. The mean of the unidirectional response rate for all cohesion-guided events is calculated to obtain the average directional response intensity. The mean of the effective delay steps for all cohesion-guided events is calculated to obtain the typical response delay. The mean directional response intensity is divided by the typical response delay to obtain the bivariate traction lag. The bivariate traction lag reflects the efficiency of the interaction between graph cohesion and graph equilibrium. When the physical condition of the elderly is stable, changes in graph cohesion will quickly lead to synchronous changes in graph equilibrium, that is, the bivariate traction lag will be small. If the bivariate traction lag increases, it means that the physiological indicators of the elderly cannot adapt to environmental changes in a timely manner, indicating the corresponding potential risk.

[0035] Within the analysis window, multiple short trajectory segments of fixed length are randomly selected. Linear fitting is performed on each short trajectory segment, and the residuals from all data points to their corresponding fitted lines are calculated to generate trajectory shape stiffness. Specifically, this includes: randomly selecting seven short trajectory segments of consecutive time points with a length of 5 within the analysis window; for each short trajectory segment, using the straight line directly determined by its first and last points as a reference baseline; calculating the vertical distances from the three middle data points within the short trajectory segment to the reference baseline, and calculating the mean of the absolute values ​​of the vertical distances to obtain the linear deviation of the short trajectory segment. Calculate the angle between the reference baseline and the positive horizontal axis of the two-dimensional phase plane to obtain seven angle values. Convert these seven angle values ​​into points on the unit circle: each angle corresponds to a vector of length 1. Sum the components of all seven vectors to obtain the composite vector. Calculate the length of the composite vector. Subtract the ratio of the length of the composite vector to the total number of short trajectory segments from 1 to obtain the circular variance. The trajectory morphological stiffness is obtained by dividing the mean of the linear deviation by the square root of the circular deviation and then multiplying it by the ratio of the total number of trajectory segments to the length of the shortest segment (i.e., five-sevenths). The trajectory morphological stiffness is used to reflect the stability and regularity of the cooperative trajectory. The higher the value, the more regular the trajectory is, and the more stable the cooperative state is.

[0036] By fusing the cooperative state offset angle, the local trajectory curvature inertia ratio, the bivariate traction hysteresis, and the trajectory shape stiffness, the cooperative degradation trajectory corresponding to the current analysis window is obtained. Specifically, this includes: multiplying the cooperative state offset angle by the local trajectory curvature inertia ratio to obtain the basic fluctuation amount; calculating the natural logarithm of the bivariate traction hysteresis plus 1, and using the reciprocal of the natural logarithm as the interaction modulation factor. Multiply the basic volatility by the interaction modulation factor, and then divide by the trajectory shape stiffness to obtain the intermediate trajectory value. Calculate the coefficient of variation (i.e., standard deviation divided by mean) of the intermediate trajectory value over the past three consecutive time windows, and multiply the current intermediate trajectory value by the coefficient of variation to obtain the cooperative degradation trajectory value representing the current cooperative state dynamics of the system.

[0037] By constructing a bivariate time series using graph cohesion and graph equilibrium values, and calculating the co-state offset angle to understand the degree of deviation of the co-state, the two local trajectory curvature-inertia ratio can understand trajectory fluctuations and inertia. At the same time, the bivariate traction hysteresis can reflect the efficiency of the interaction between the two. Then, the stability of the trajectory is evaluated based on the trajectory morphological stiffness. Finally, a co-degenerate trajectory is generated, which accurately captures the slow degradation and evolution of correlations between multimodal data, identifies hidden potential risks that have not reached the single-modal threshold in advance, solves the problem of delayed decision-making for home safety of the elderly, and improves the timeliness of risk warning.

[0038] In one embodiment, the cooperative degradation trajectory is analyzed to generate a trajectory instability index, including: The temporal variations of the co-degenerate trajectories are symbolized, and the trajectory sequence entropy is generated by analyzing the complexity of the symbol sequence arrangement across multiple time scales. Specifically, this includes: calculating the first-order difference of the co-degenerate trajectory sequence to obtain the variation sequence; setting three time scales with 1, 2, and 3 sampling intervals; and comparing the variation sequence with the median of the absolute value of the variation sequence over the past ten time windows at each scale. If the current change is greater than 0.5 times the median, mark it with the sign +; if the current change is less than 0.5 times the median, mark it with the sign -; otherwise mark it with the sign 0, resulting in three sign sequences. From each symbol sequence, every four consecutive symbols are extracted as a pattern segment. The number of ternary symbol combinations that appear at three scales and are formed by arranging the first symbols of the three pattern segments at the same position in scale order is counted. Here, a ternary symbol combination refers to a combination formed by arranging the first symbols of the pattern segments at the same position at the three time scales in the order of scale 1, scale 2, and scale 3. The value of each first symbol is independently one of +, -, or 0, such as (+, +, 0), (0, -, +), (-, 0, -), etc. Divide this number of species by the maximum number of species, which is 27 species, and then multiply it by the total frequency of the symbol 0 in all sequences to obtain the trajectory order entropy. The trajectory order entropy reflects the complexity of the change pattern of the co-degenerative trajectory. When the elderly are in a stable physical state, the trajectory change pattern is relatively simple, and the trajectory order entropy is low. If the trajectory order entropy increases, it means that the trajectory change pattern is chaotic, which may indicate that the elderly’s health condition has deteriorated sharply or that the environment has suddenly become abnormal.

[0039] The co-degenerate trajectory is decomposed to generate abnormal fluctuation energy. Specifically, the co-degenerate trajectory is smoothed by a moving average with a window length of 3 to obtain a smoothed trajectory. Subtract the smoothed trajectory from the original co-degenerate trajectory to obtain the residual sequence; calculate the median of the absolute value of the residual sequence over the past five consecutive time windows, and use this median as the dynamic threshold; Within the current analysis window of 10 time points, identify all time points where the absolute value of the residual exceeds the dynamic threshold, and count each point as an impulse event; calculate the absolute value of the residual in each impulse event, which is the intensity; Multiply the sum of the squares of the intensity of all pulse events by the number of pulse events, and then divide by the total number of time points in the current analysis window (10) to generate abnormal fluctuation energy. Abnormal fluctuation energy is used to identify sudden abnormal fluctuations in the trajectory. Changes in its value, such as an increase, indicate a significant deviation from the normal trend, corresponding to sudden risks to the home safety of the elderly or environmental emergencies.

[0040] In one embodiment, analyzing the cooperative degradation trajectory to generate a trajectory instability index further includes: Identify and analyze long-term trend inflection points in the co-degradation trajectory to generate trend inertia strength. Specifically, this involves: performing first-order difference on the co-degradation trajectory and then calculating its cumulative sum sequence; on the cumulative sum sequence, using a sliding window with a length of 12 time points, calculating the slope of the line connecting the first and last points within the window as the trend direction of the window. When the trends of two consecutive windows are opposite, and the range of the cumulative sum series in the current window is greater than the median of the range of the cumulative sum series in the entire analysis window, the intersection of the two windows is marked as a long-term trend reversal point. For each long-term trend turning point, calculate the absolute value of the linear fit slope of the subsequence formed by the six points before and after it, and calculate the harmonic mean of the absolute values ​​of the two subsequences, which is used as the trend maintenance degree of the long-term trend turning point. Calculate the geometric mean of the trend maintenance degree of all long-term trend turning points, multiply it by the number of all long-term trend turning points, and obtain the trend inertia strength. The trend inertia strength mainly reflects the stability of the long-term trend of the co-degenerative trajectory.

[0041] The trajectory instability index is obtained by fusing the trajectory sequence entropy, abnormal fluctuation energy, and trend inertia strength. Specifically, this includes: calculating the geometric mean of the trajectory sequence entropy, abnormal fluctuation energy, and trend inertia strength, and using it as the collaborative baseline; and calculating the ratio of the arithmetic mean of the trajectory sequence entropy, abnormal fluctuation energy, and trend inertia strength to the harmonic mean, and using it as the discrete amplification factor. The intermediate coupling value is obtained by multiplying the trajectory sequence entropy by the abnormal fluctuation energy and then dividing by the trend inertia strength. The trajectory instability index is obtained by multiplying the coordinated baseline by the discrete amplification factor and then by the intermediate coupling value. The trajectory instability index is a quantitative indicator of potential risk and is related to the home safety status of the elderly. The lower the trajectory instability index, the more stable the coordinated state, and the lower the safety risk of the elderly. The higher the trajectory instability index, the more unstable the coordinated state, and the higher the safety risk of the elderly.

[0042] By quantifying the complexity of collaborative degradation trajectory change patterns through trajectory sequence entropy quantification, abnormal and chaotic signals of health status or environment can be accurately captured. Abnormal fluctuations in energy can be used to identify sudden abnormal fluctuations in the trajectory, while trend inertia strength can represent the long-term trend stability of the trajectory and perceive potential degradation trends. The resulting trajectory instability index can accurately quantify potential risks, and can uncover early hidden dangers in the correlation degradation of multimodal data in advance. This can effectively solve the problem of delayed decision-making for home safety of the elderly and improve the timeliness of risk warning.

[0043] In one embodiment, a trajectory instability index is calculated to generate a risk threshold. The trajectory instability index is then compared with the risk threshold to generate a safety decision instruction, including: The trajectory instability index is analyzed across multiple windows to generate intrinsic risk potential. Specifically, this involves analyzing the trajectory instability index sequence at three different time scales, corresponding to windows with 5, 10, and 15 consecutive sampling points, respectively. Within each scale window, the linear trend slope of the trajectory instability index is calculated. If the linear trend slope is positive, the geometric mean of the three linear trend slopes is taken as the multi-scale resonance intensity; otherwise, the multi-scale resonance intensity is zero. Calculate the ratio of the linear trend slope within the shortest window (5 points) to the medium window (10 points) and use it as the acceleration factor. Multiply the multi-scale resonance intensity, the acceleration factor, and the current trajectory instability index value to obtain the intrinsic risk potential energy. The intrinsic risk potential energy is mainly used to judge the cumulative intensity and acceleration trend of risk. An increase in its value means that the risk is not only accumulating, but also accumulating at an accelerating rate, which corresponds to the elderly's physical condition deteriorating and the rate of deterioration accelerating.

[0044] The risk threshold is obtained by jointly calculating the risk intrinsic potential energy, abnormal fluctuation energy, and trend inertia strength at the current moment. Specifically, this includes: dividing the risk intrinsic potential energy by the abnormal fluctuation energy to obtain the potential-to-manifest ratio; and calculating the moving standard deviation of the potential-to-manifest ratio over the past five time windows to obtain the released volatility. Divide the intrinsic risk potential by the trend inertia strength and then multiply by the released volatility to obtain a risk value; Calculate the 70th percentile of the risk value over the past ten time windows, add this percentile to the current risk value, and divide by two to generate the risk threshold.

[0045] The trajectory instability index is compared with a risk threshold to generate safety decision instructions, specifically including: If the trajectory instability index is less than or equal to the risk threshold, the elderly person is deemed to be in good health and a safety instruction is generated. When the trajectory instability index exceeds the risk threshold, it is determined that there is a risk to the safety of elderly people at home, and an early warning instruction is generated.

[0046] By analyzing the trajectory instability index across multiple time scales to generate intrinsic risk potential energy, the intensity and acceleration trend of risk accumulation can be captured. Furthermore, by combining intrinsic risk potential energy, abnormal fluctuation energy, and trend inertia strength, risk thresholds can be dynamically calculated to adapt to risk assessment needs under different conditions. Decision instructions are generated by comparing the trajectory instability index with the risk threshold, breaking through the existing ex-post alarm mechanism that relies on single-modal thresholds. Early risks hidden in the correlation evolution of multimodal data can be identified in advance, avoiding the accumulation of risks for the elderly at home. At the same time, risk assessment and dynamic early warning can be achieved, effectively solving the problem of delayed safety decisions and providing timely decision-making services for the safety of the elderly at home.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A home safety decision-making service system for the elderly that integrates multimodal sensing data, characterized in that, include: The data interaction analysis unit is used to acquire indoor environmental perception data and physiological perception data of the target service object in real time. The feature vector of each type of perception data after preprocessing is used as a modal node. Within a continuously sliding analysis window, the phase lock value between any two modal nodes is calculated and used as the interaction edge weight connecting the two modal nodes. Based on all modal nodes and interaction edge weights, a dynamic interaction map is generated. The target service object is the elderly. The data change analysis unit is used to conduct in-depth analysis of the dynamic changes in the level of collaboration in the dynamic interaction graph, and to obtain the graph cohesion and graph equilibrium values ​​for each time window. The data calculation unit is used to acquire time series values ​​of graph cohesion and graph equilibrium in real time within a sliding analysis window, form a bivariate time series, calculate the bivariate time series, and generate a co-degradation trajectory. The safety analysis unit is used to analyze the cooperative degradation trajectory and generate a trajectory instability index; The safety decision unit is used to calculate the trajectory instability index, generate a risk threshold, compare the trajectory instability index with the risk threshold, and generate a safety decision instruction.

2. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 1, characterized in that, The feature vectors of each type of preprocessed sensing data are used as modal nodes. Within a continuously sliding analysis window, the phase lock value between any two modal nodes is calculated and used as the interaction edge weight connecting the two modal nodes. Based on all modal nodes and interaction edge weights, a dynamic interaction graph is generated, including: After feature extraction for each type of preprocessed sensing data, feature vectors are formed. The feature vectors are then analyzed to generate feature activity. Based on the characteristic activity of any two modal nodes, the synchronization of any two modal nodes is calculated to obtain the phase lock value.

3. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 2, characterized in that, The feature vectors of each type of preprocessed sensing data are used as modal nodes. Within a continuously sliding analysis window, the phase lock value between any two modal nodes is calculated and used as the interaction edge weight connecting the two modal nodes. Based on all modal nodes and interaction edge weights, a dynamic interaction graph is generated, which also includes: Using modal nodes as vertices and phase lock values ​​as interaction edge weights, an initial weighted undirected graph is constructed. The geometric mean of the edge weights of all triangle closed loops in the initial weighted undirected graph is calculated to obtain the structural cohesion factor of the current window. The feature activity of each modal node is used as a node attribute, and the structural cohesion factor of the current window is used as a global attribute. Both are then embedded into the initial weighted undirected graph to generate a dynamic interaction graph.

4. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 3, characterized in that, A thorough analysis of the dynamic changes in synergy levels within the dynamic interaction graph yields the graph cohesion and graph equilibrium values ​​for each time window, including: For dynamic interactive spectra, the coupling resonance ratio is calculated based on the phase lock value and characteristic activity. The coupling resonance ratio is regarded as the field strength distributed on the edge of the dynamic interactive spectra. By analyzing the potential energy diffusion and iteration of the field strength, the inhomogeneity of the resonance field is obtained. Based on the phase lock value and coupling resonance ratio, the flow between all modal node pairs is calculated, and the network flow bottleneck coefficient is generated. Based on the coupling resonance ratio, the dynamic interaction spectrum is divided into high harmonic oscillator graphs and low harmonic oscillator graphs. The interaction edge weights within the high harmonic oscillator graphs and low harmonic oscillator graphs are analyzed separately to obtain the substructure polarization coefficients.

5. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 4, characterized in that, A thorough analysis of the dynamic changes in synergy levels within the dynamic interaction graph yields the graph cohesion and graph equilibrium values ​​for each time window, and also includes: In a continuously sliding analysis window, the node set overlap and edge weight structure similarity between the high harmonic oscillator graph of the current window and the high harmonic oscillator graph of the previous window are calculated to obtain the cooperative kernel transition strength. By integrating the structural cohesion factor, the resonant field inhomogeneity, and the network flow bottleneck coefficient, the graphical cohesion of the cost time window is generated. By fusing the coupling resonance ratio, substructure polarization coefficient, and cooperative nuclear transition intensity, a graphical equilibrium value for the cost time window is generated.

6. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 5, characterized in that, Within the sliding analysis window, time-series values ​​of graph cohesion and graph equilibrium are acquired in real time to form a bivariate time series. The bivariate time series is then used to calculate and generate a co-degradation trajectory, including: Identify the co-states corresponding to graph cohesion and graph equilibrium values ​​in bivariate time series. At each time point, calculate the angle between the current graph cohesion and graph equilibrium value and the co-state to obtain the co-state offset angle. In a two-dimensional phase plane composed of graph cohesion and graph equilibrium values, data points are connected to form a trajectory. The changes in the trajectory at each point are analyzed to obtain the local trajectory curvature inertia ratio.

7. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 6, characterized in that, Within the sliding analysis window, time-series values ​​of graph cohesion and graph equilibrium are acquired in real time to form a bivariate time series. The bivariate time series is then used to calculate and generate a co-degradation trajectory, which also includes: The interaction between the graph cohesion sequence and the graph equilibrium value sequence is analyzed to generate a bivariate traction hysteresis. Within the analysis window, multiple short trajectory segments of fixed length are randomly selected. Linear fitting is performed on each short trajectory segment, and the residuals from all data points to their corresponding fitted lines are calculated to generate trajectory shape stiffness. By fusing the cooperative state offset angle, the local trajectory curvature-inertia ratio, the bivariate traction hysteresis, and the trajectory morphological stiffness, the cooperative degradation trajectory corresponding to the current analysis window is obtained.

8. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 7, characterized in that, The co-degradation trajectory is analyzed to generate a trajectory instability index, including: The temporal variation of the collaborative degradation trajectory is symbolized, and the trajectory sequence entropy is generated by analyzing the complexity of its symbol sequence arrangement across multiple time scales. The co-degenerate trajectory is decomposed to generate abnormal fluctuation energy.

9. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 8, characterized in that, The analysis of the collaborative degradation trajectory generates a trajectory instability index, and also includes: Identify and analyze long-term trend inflection points in the collaborative degradation trajectory to generate trend inertia strength; By fusing trajectory sequence entropy, abnormal fluctuation energy, and trend inertia strength, the trajectory instability index is obtained.

10. The home safety decision-making service system for the elderly that integrates multimodal sensing data according to claim 9, characterized in that, The trajectory instability index is calculated to generate a risk threshold. The trajectory instability index is compared with the risk threshold to generate safety decision instructions, including: The trajectory instability index of multiple windows is analyzed to generate the intrinsic potential energy of risk. The risk threshold is obtained by jointly calculating the intrinsic potential energy of the risk at the current moment, the energy of abnormal fluctuations, and the strength of trend inertia. The trajectory instability index is compared with the risk threshold to generate safety decision instructions.

Images