AI-driven nursing resource distribution system

Abstract

The invention relates to the technical field of data management, in particular to an AI-driven nursing resource allocation system. The system utilizes a degree of deviation from historical behavioral characteristics as a quantitative criterion for assessing patient behavioral abnormalities. Utilizing a graph structure algorithm to construct an inter-patient influence relation graph, and combining the deviation degree to obtain a bearing risk value of each patient. Based on the basic intervention cost, the bearing risk accumulated value is used for adjustment, and the obtained updated intervention cost can effectively represent the risk caused by a nursing decision. And the correct nurse distribution can be determined according to the update intervention cost corresponding to the target patient. According to the invention, based on the behavior characteristics of the patients, the mutual influence relationship between the patients is considered and is synchronously updated to the intervention cost between the nurses and the patients, so that a correct and effective allocation decision can be realized when the nursing resources are allocated.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and more specifically to an AI-driven nursing resource allocation system. Background Technology

[0002] In mental health and long-term care facilities, nighttime monitoring of hospitalized patients is a core element in ensuring their safety. For some patients with cognitive impairment, nighttime is a peak period for emotional and behavioral problems, often manifesting as agitated behavior. Agitation not only poses a safety risk to the patient themselves, but the noises or movements they produce can also disturb other patients, even triggering anxiety and unease in other sensitive patients, creating a behavioral "chain reaction." This phenomenon means that the risk to an individual patient is no longer an isolated event, but can potentially evolve into a systemic risk for the entire care unit.

[0003] Regarding stress risk management for patients, manual rounds by caregivers suffer from the drawbacks of caregiver subjectivity and inherent observation intervals, making it difficult to capture subtle signals in the early stages of agitated behavior, let alone predict and prevent the aforementioned "chain reaction." Existing technologies, such as pressure mattresses and other sensor technologies, monitor patients' status in bed, but their primary function is to monitor out-of-bed events to prevent falls. They have limited ability to perceive postural agitation and physiological restlessness that are precursors to agitated behavior while in bed. These technologies treat each patient as an independent monitoring unit, lacking the ability to assess the interactions between patients within the care unit. They typically only provide a passive response after behavioral problems have occurred or escalated, missing the opportunity for early intervention and systemic prevention at the lowest cost. Summary of the Invention

[0004] To address the limitations of existing technologies in monitoring and sensing patient stress states, and their failure to consider the risks arising from interactions between patients, which leads to untimely and unreasonable allocation of nursing resources, this invention aims to provide an AI-driven nursing resource allocation system. The specific technical solution adopted is as follows: This invention proposes an AI-driven nursing resource allocation system, the system comprising: The behavior analysis module is used to monitor the behavioral characteristics of each patient during the monitoring period. The behavioral characteristics include acoustic characteristics and bed posture characteristics; and to obtain the degree of deviation of the behavioral characteristics from historical behavioral characteristics. The patient impact relationship analysis module is used to construct an impact relationship graph between patients, with each patient as a node. The edge weight between two nodes in the impact relationship graph is obtained from the distance between patients and the Granger causality test results between patients. The risk value of each patient is obtained based on the degree of deviation of each patient and the corresponding edge weight. The nurse intervention cost analysis module is used to retrieve the basic intervention cost between nurses and patients; for any nurse-patient combination, it obtains the cumulative risk value of all patients in the nurse's responsible patient list excluding the patients in the combination; the basic intervention cost is adjusted based on the cumulative risk value to obtain the updated intervention cost; The decision allocation module is used to select the patient with the maximum deviation as the target patient after the patient's risk value and deviation degree meet the preset intervention conditions; and to determine the assigned nurse for the target patient based on the updated intervention cost.

[0005] Furthermore, the method for obtaining the acoustic features includes: Acquire the patient's audio signal; identify whether the audio signal is an interactive audio signal or a non-speech human voice signal; if it is an interactive audio signal, extract the spectral entropy and rhythm of the interactive audio signal to obtain acoustic features; if it is a non-speech human voice signal, extract the micro-vibrations and micro-amplitudes of the non-speech human voice signal to obtain acoustic features.

[0006] Furthermore, the method for obtaining the bed posture features includes: The pressure signal matrix monitored by the patient's bed is acquired by a matrix of sensors arranged on the bed; the gray-level co-occurrence matrix of the pressure signal matrix is ​​extracted, and the energy and contrast of the gray-level co-occurrence matrix are used as the bed posture features.

[0007] Furthermore, the method for obtaining the degree of deviation includes: Each dimension of the behavioral characteristics is Z-standardized in its respective dimension to obtain normalized behavioral characteristics. If the current monitoring period is in the system startup phase, the mean and standard deviation in the Z-standardization process are calculated from all the historical data collected. If the current monitoring period is in the system steady state phase, the mean and standard deviation in the Z-standardization process are calculated from the historical data within the preset time window before the current monitoring period. The L2 norm of the normalized behavioral features is used as the degree of deviation.

[0008] Furthermore, the edge weight is the weighted sum of the spatial distance weight and the historical behavior following relationship weight according to a preset weight; The spatial distance weight is obtained from the distance between nodes; The weights of the historical behavior follow-up relationships were obtained from the Granger causality test results.

[0009] Furthermore, the method for obtaining the spatial distance weight includes: The spatial distance weight between two patients in the same ward is set to 1, the spatial distance weight between two patients in adjacent wards is set to 0.5, the spatial distance weight between two patients on the same floor is set to 0.2, and the spatial distance weight between two patients in other cases is set to 0.

[0010] Furthermore, the method for obtaining the weights of the historical behavior follow-up relationship includes: During the system startup phase, the weights of all historical behavior follow-up relationships are set to 0. During the steady-state phase of the system, for each pair of nodes, a Granger causality test is performed based on the deviation sequence under the preset time window before the current monitoring cycle to obtain the test coefficient. The test coefficient is subtracted from the positive integer 1. If the result is less than 0, the weight of the historical behavior following relationship is set to 0; if the result is greater than 0, the result is used as the weight of the historical behavior following relationship.

[0011] Furthermore, the method for obtaining the risk value includes: For each patient, the deviation degree of other patients is used as a weight, and the edge weights are summed in a weighted manner to obtain the exogenous risk coefficient. The sum of the exogenous risk coefficient and the deviation degree of the current patient is used as the risk-bearing value.

[0012] Furthermore, the method for obtaining the update intervention cost includes: After normalizing the accumulated risk value, the sum of this value and the basic intervention cost is used as the updated intervention cost.

[0013] Furthermore, the preset intervention conditions include: During the current monitoring period, at least one patient among all patients has a deviation level greater than the preset deviation level threshold, and the maximum risk value among all patients is greater than the preset risk peak threshold.

[0014] The present invention has the following beneficial effects: To enhance the sensing capabilities of monitoring equipment, this invention considers both the patient's acoustic and bed posture characteristics, using these two features as patient behavioral features for analysis. The degree of deviation from historical behavioral characteristics is used as a quantitative standard for assessing abnormal patient behavior. To analyze the mutual influence relationships between patients, this invention utilizes a graph structure algorithm to construct a patient influence relationship graph. In this graph, the edge weights between nodes are obtained through the distance between patients and the Granger causality check results, ensuring that the edge weights simultaneously encompass both spatial relationships and historical behavioral interactions, allowing them to serve as an exogenous risk indicator for each patient. Furthermore, combining the degree of deviation yields the risk-bearing value for each patient. Before allocating nursing resources, this invention adjusts the basic intervention cost using the accumulated risk-bearing value, resulting in an updated intervention cost that effectively characterizes the risk caused by a nursing decision. The correct nurse can then be assigned based on the updated intervention cost corresponding to the target patient. This invention, based on patient behavioral characteristics and considering the mutual influence relationships between patients, synchronously updates the intervention cost between nurses and patients, enabling accurate and effective allocation decisions during nursing resource allocation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a block diagram of an AI-driven nursing resource allocation system provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an AI-driven nursing resource allocation system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The embodiment of the present invention is directed to the allocation of nursing resources in the psychiatric department. Since the embodiment of the present invention involves machine learning algorithms such as graph structure construction, Granger causality check, and global optimization, it needs to be driven by existing AI technologies. The various algorithms in the system involved in the embodiment of the present invention can be executed by existing artificial intelligence chips. The embodiment of the present invention relies on the existing nursing resource platform. In the existing basic nursing resource platform, the original work schedules of each nurse are stored, and the basic intervention costs between nurses and patients can be obtained through existing technologies. The basic intervention costs can be obtained from three dimensions: ability matching, historical familiarity, and current load. For example, in a medical monitoring method based on multimodal prompt interaction proposed in the patent with the publication number CN120319425A, the method for obtaining the basic intervention cost is a well-known technical means for those skilled in the art. The embodiment of the present invention takes it as known prior data, and this data is pre-stored in the nursing resource platform.

[0020] The following specifically describes the specific solution of an AI-driven nursing resource allocation system provided by the present invention with reference to the accompanying drawings.

[0021] Please refer to Figure 1 , which shows a block diagram of an AI-driven nursing resource allocation system provided by an embodiment of the present invention. The system includes a behavior analysis module 101, a patient influence relationship analysis module 102, a nurse intervention cost analysis module 103, and a decision-making allocation module 104.

[0022] Among them, the behavior analysis module 101 is used to monitor the behavior characteristics of each patient in the monitoring period. The behavior characteristics include acoustic characteristics and in-bed posture characteristics. In the embodiment of the present invention, the acoustic characteristics are obtained by collecting acoustic signals through an acoustic acquisition device in the ward and analyzing them; the in-bed posture characteristics are obtained by collecting pressure signals through a pressure sensor installed on the patient's hospital bed and analyzing them.

[0023] In the embodiment of the present invention, the monitoring period is set to 1 minute, that is, the analysis and warning of the embodiment of the present invention are performed every time one minute of data is collected, ensuring the timely allocation of nursing resources.

[0024] In the embodiment of the present invention, an omnidirectional MEMS microphone is deployed inside each independent ward. The microphone continuously collects audio signals at a sampling rate of 16 kHz and a precision of 16 bits. To protect patient privacy, the original audio waveform data is limited to be processed within the edge computing unit supporting the sensor. The unit segments the audio signals with a non-overlapping time window of 5 seconds. That is, there are 12 segments of audio signals in one monitoring period.

[0025] In this embodiment of the invention, a flexible piezoresistive sensor array consisting of 32 rows and 32 columns of sensing units is laid under the mattress of each hospital bed. This array synchronously collects the pressure value of each sensing unit at a frequency of 1Hz, forming a 32x32 pressure matrix. That is, within the monitoring period, each element in the pressure matrix corresponds to one pressure signal.

[0026] Preferably, in this embodiment of the invention, considering that patients' agitated behavior is often accompanied by specific types of sound events, such as aimless rubbing sounds or suppressed groans, and that there is a large amount of irrelevant background noise in the ward environment, it is necessary to first qualitatively identify the sound type when acquiring acoustic features. The object interaction audio signal represents the signal generated when the patient interacts with objects such as the bed and clothing. This type of signal is usually generated by the patient's aimless, repetitive limb movements due to restlessness, such as tossing and turning in bed. Non-vocal signals typically represent suppressed groans or murmurs made by the patient due to anxiety or pain. Therefore, this embodiment of the invention extracts these two types of signals from the audio signal, which can effectively characterize the abnormal behavior of the current patient caused by their illness.

[0027] In this embodiment of the invention, the audio signal can be input into a pre-trained sound event detection model in the edge computing unit. The sound event detection model in this embodiment employs a convolutional recurrent neural network architecture, a technique well-known to those skilled in the art, and its specific construction method will not be elaborated here. This model categorizes the main sound energy within the audio signal into background noise, breathing or snoring, object interaction audio, and non-voice human voice. This invention focuses only on object interaction audio and non-voice human voice; if the audio signal falls into the other two categories, all acoustic feature components are directly set to 0.

[0028] If the audio signal is an object interaction audio signal, considering that the irregular and disordered characteristics of the object interaction audio signal can reflect the irregular movements produced by the patient under stress, this embodiment of the invention extracts the spectral entropy and rhythmicity of the object interaction audio signal as two acoustic features. It should be noted that the method for obtaining the spectral entropy is a well-known technique in the art, and the rhythmicity can be obtained by calculating the autocorrelation function of the signal energy envelope, which is also a well-known technique in the art and will not be elaborated here.

[0029] If the audio signal is a non-speech human voice signal, considering that the vocal instability exhibited by the non-speech human voice signal can reflect the patient's physiological stress, this embodiment of the invention extracts the micro-vibrations and micro-amplitudes of the non-speech human voice signal as two acoustic features. It should be noted that this embodiment of the invention extracts the fundamental frequency sequence and amplitude sequence of the signal, and calculates its period to obtain the relative change in period, thus obtaining the corresponding micro-vibrations and micro-amplitudes. The specific methods for obtaining these features are well-known to those skilled in the art and will not be elaborated upon here.

[0030] In order to ensure data alignment, this embodiment of the invention sets the acoustic features as a four-dimensional sequence. ,in This represents the acoustic feature sequence of the j-th patient in the t-th audio signal. Let be the spectral entropy of the j-th patient when the t-th audio signal is an audio signal generated by an interaction between objects. For the j-th patient, when the t-th audio signal is an interactive audio signal between objects, The microtremor of the j-th patient during the t-th audio signal is a non-speech human voice signal. Let be the amplitude of the j-th patient when the t-th audio signal is a non-speech human voice signal. For each audio signal, the non-existent acoustic feature components are set to 0, that is, each audio signal is ultimately a four-dimensional acoustic feature sequence, and there are two components with 0 in a sequence, and there are 12 acoustic feature segments in one monitoring period.

[0031] Preferably, in this embodiment of the invention, considering that patients often exhibit two different patterns in the pre-agitation phase: "persistent muscle tension" and "high-frequency postural agitation," simple pressure value statistics cannot effectively distinguish between these two patterns. Therefore, this embodiment of the invention uses the analysis of pressure distribution change patterns to quantify the dynamic postural characteristics of the patient during bedtime. The method for obtaining bedtime postural characteristics includes: The pressure signal matrix monitored by sensors arranged in a matrix on the patient's bed is acquired. It should be noted that, because the sensor array in one specific implementation of this invention is in the form of a 32×32 array with a sampling frequency of 1Hz, there will be 60 pressure matrices in one monitoring cycle. In this embodiment, these 60 pressure matrices are sequentially laid out horizontally in chronological order to form a pressure signal matrix of 32 rows and 1920 columns.

[0032] The obtained pressure signal matrix needs to be used to extract quantitative indicators that characterize muscle tension and postural agitation. This matrix can be analyzed as a two-dimensional grayscale image. In this embodiment of the invention, the grayscale co-occurrence matrix of the pressure signal matrix is ​​extracted, and the energy and contrast of the grayscale co-occurrence matrix are used as the in-bed posture features. Energy can quantify the homogeneity of the image. High energy represents the phenomenon of concentrated local pressure in the body due to prolonged maintenance of a certain posture caused by muscle tension. This directly corresponds to the large-area similar local pressure distribution caused by muscle tension. Contrast can quantify the intensity of local changes in pixel grayscale, which directly corresponds to the frequent small-range pressure jumps caused by postural agitation.

[0033] It should be noted that the methods for obtaining the energy and contrast of the gray-level co-occurrence matrix described above are well-known techniques to those skilled in the art. In other implementations of this invention, other texture analysis methods such as gray-level run-length matrices or local binary patterns can be used to extract similar features to characterize the disorder of elements in the matrix. Ultimately, each patient generates a two-dimensional bed posture feature vector during the monitoring period.

[0034] It should be noted that, due to the different sampling frequencies of audio and pressure signals, the number of acoustic features and bed posture features obtained within a monitoring cycle differs. To ensure data alignment, this embodiment of the invention calculates the average and maximum values ​​of the four components of all acoustic features within a monitoring cycle. These eight features, composed of the average and maximum values, are then combined with a two-dimensional bed posture feature vector to form the final behavioral feature. That is, the final behavioral feature is a ten-dimensional feature vector.

[0035] Because this invention addresses the allocation of nursing resources for sudden abnormalities in patients within mental health or long-term care facilities, these sudden abnormalities, being rare events compared to normal behavior, should result in behavioral characteristics that significantly differ from the average historical behavioral characteristics over a long period. Therefore, after quantifying the patient's behavioral characteristics, the behavior analysis module 101 compares them with historical behavioral characteristics to determine the degree of deviation. A greater deviation indicates a higher probability of the patient exhibiting abnormal behavior. This characteristic represents the risk of abnormality derived from the individual patient's behavior, without considering inter-patient influence.

[0036] Preferably, in this embodiment of the invention, considering that there are multiple components in the behavioral features, each component in its own dimension has its own dimensions and data range, and in order to obtain the degree of deviation, this embodiment of the invention uses Z-standardization to process the feature components of each dimension in the behavioral features. The final result can achieve dimensionless and standardized processing. Furthermore, since the mean and standard deviation in the Z-standardization process are obtained from historical data, they can also reflect the degree of deviation compared to historical behavioral features. This embodiment of the invention divides the entire process into a system startup phase and a system steady-state phase. The system startup phase contains less data; therefore, if the current monitoring period is in the system startup phase, the mean and standard deviation in the Z-standardization process are calculated from all the collected historical data. The system steady-state phase has stored a large amount of historical data; therefore, if the current monitoring period is in the system steady-state phase, the mean and standard deviation in the Z-standardization process are calculated from historical data within a preset time window before the current monitoring period. Finally, after Z-standardization, each component in the normal feature can be considered to be 0; therefore, the L2 norm of the normalized behavioral feature can be used as the degree of deviation.

[0037] It should be noted that the time window in this embodiment of the invention is set to 24 hours. The system startup phase is within 24 hours after system startup, and the system steady-state phase is the phase after the system has been running for more than 24 hours. The specific Z-normalization formula is a technical means well known to those skilled in the art, and will not be elaborated or limited here.

[0038] After analyzing the abnormal risks reflected in the behavioral characteristics of each patient, the system needs to further analyze the influence between patients. In the mental health nursing unit, the acoustic or visual signals generated by a patient's agitated behavior can be perceived by other patients through airborne transmission or direct observation, potentially inducing or exacerbating their anxiety and unease. To quantify this mutual influence effect between patients, the patient influence relationship analysis module 102 constructs a directed weighted graph using a graph structure algorithm, treating each patient as a node to build the influence relationship graph between patients. The edge weights between two nodes in the patient influence relationship graph are obtained from the distance between patients and the Granger causality test results. The closer the distance between patients, the more synchronized their sights and sounds, and the easier it is for them to influence each other. The Granger causality test is specifically used to test whether the past value of one time series has predictive power for the current value of another time series, thus confirming and quantifying this one-way influence relationship from a data perspective. Therefore, the test results can determine the influence relationship between patients. In other words, the edge weights simultaneously consider the spatial relationship between patients and the actual behavioral influence, representing the exogenous influence risk between patients. Ultimately, for each patient, the risk-bearing value can be obtained by combining the degree of deviation with the corresponding edge weights. The risk-bearing value includes both the patient's intrinsic and exogenous risks and is an important reference indicator for the allocation of nursing resources.

[0039] Preferably, in this embodiment of the invention, considering that the edge weights simultaneously include two factors, each factor can correspond to a weight. The edge weights can be set as a weighted sum of the spatial distance weight and the historical behavior following relationship weight, based on preset weights. In this embodiment of the invention, the weights of both parameters are set to 0.5. In other implementations of this embodiment, if the system favors spatial influence, the weight of the spatial distance weight can be adaptively increased; similarly, if it favors the causal influence of historical data, the weight of the historical behavior following relationship weight can be adaptively increased. Further details and limitations are not provided.

[0040] The spatial distance weight is obtained from the distance between nodes. In this embodiment of the invention, the spatial distance weight between two patients in the same ward is set to 1, the spatial distance weight between two patients in adjacent wards is set to 0.5, the spatial distance weight between two patients on the same floor is set to 0.2, and the spatial distance weight between two patients in other cases is set to 0.

[0041] The weights of the historical behavior following relationships are obtained from the Granger causality test results. In this embodiment of the invention, considering that the Granger causality test needs to be implemented based on historical data, the time range of the monitoring period is also divided into two stages: During the system startup phase, the weights of all historical behavior follow-up relationships are set to 0. During the steady-state phase of the system, for each pair of nodes, a Granger causality test is performed based on the deviation sequence within a preset time window prior to the current monitoring period to obtain the test coefficient. The test coefficient is the p-value in the Granger causality test; a smaller test coefficient indicates a stronger mutual influence between the two patients. Since the coefficient is a value between 0 and 1, a positive integer 1 is subtracted from the test coefficient. To avoid negative results, if the result is less than 0, the weight of the historical behavior following relationship is set to 0; if the result is greater than 0, the result is used as the weight of the historical behavior following relationship.

[0042] It should be noted that the specific algorithm of the Granger causality test is a well-known technique in the field of causality and will not be described in detail here.

[0043] Preferably, in this embodiment of the invention, the method for obtaining the risk value includes: For each patient, the deviation levels of other patients are used as weights, and the edge weights are summed in a weighted manner to obtain an exogenous risk coefficient. The sum of the exogenous risk coefficient and the deviation level of the current patient is used as the carrying risk value. The carrying risk value is expressed by the formula: ;in Let the risk value be that of the j-th patient in the k-th monitoring period. The degree of deviation for the j-th patient in the k-th monitoring period. For the number of patients, The degree of deviation of patient l, Let be the edge weight of the relationship between patients in the inter-patient influence graph during the k-th monitoring period, pointing from patient l to patient j.

[0044] After the analysis of the above two modules, the risk analysis results for each patient in the current monitoring period are obtained. These results can then be used to adjust the basic intervention costs stored in the nursing platform. The nurse intervention cost analysis module 103 is used to obtain the cumulative risk value of all patients in the nurse's responsible patient list, excluding those in the combination, for any nurse-patient combination; the basic intervention cost is adjusted based on the cumulative risk value to obtain the updated intervention cost.

[0045] It should be noted that the nurse intervention cost analysis module queries the electronic scheduling data of the nursing platform in real time through a standard interface to obtain the list of patients under each nurse's responsibility. Then, it sums up the risk of all patients in the list except those within the same group to obtain the cumulative risk value. The cumulative risk value represents the total uncontrolled risk exposed to other patients under nurse i's care if nurse i is assigned to care for patient j.

[0046] Preferably, in this embodiment of the invention, the method for obtaining the update intervention cost includes: After normalizing the accumulated risk value, the sum of this value and the basic intervention cost is used as the updated intervention cost.

[0047] In this embodiment of the invention, the normalization process can be implemented by range standardization, which involves statistically analyzing the maximum and minimum values ​​in the data dimension. The specific method is a well-known technique to those skilled in the art and will not be elaborated here.

[0048] In a clinical setting, excessively frequent alarms can lead to "alarm fatigue" among nursing staff, reducing the system's reliability and effectiveness. Therefore, the decision allocation module 104, based on preset intervention conditions, only performs nursing decision allocation when the patient's risk level and deviation degree meet these conditions. This ensures that the system only initiates resource allocation at critical moments when both significant individual risk and fragile patient relationships are simultaneously met. During resource allocation, since the deviation degree represents an abnormality in patient behavior, the patient with the highest deviation degree is selected as the target patient. The assigned nurse for this target patient is then determined based on the cost of updating the intervention. In other words, the nurse with the lowest cost of updating the intervention for the target patient is selected. After determining the target patient's allocation decision, the next highest deviation degree patient can be selected as the new target patient for resource allocation until the overall condition no longer meets the intervention conditions, indicating that the patient's overall condition is good.

[0049] In this embodiment of the invention, to make the decision command clearly visible, the system generates a structured text instruction. This instruction includes the following information fields: priority, time, patient information, suggested action, situation assessment, and basis for recommendation. A complete instruction example is as follows: "High priority 301 room - Mr. Wang; Recommendation: Nurse [Zhang San] should immediately conduct a reassuring round; Situation assessment: The patient's current individual behavior intensity is high, and the overall system peak risk of the nursing unit exceeds the threshold; Basis for recommendation: After comprehensively considering the basic intervention cost and the opportunity cost arising from this assignment, the cost of updating the intervention is the lowest." The instruction is ultimately sent to the central display terminal at the nurse station or to the mobile nursing terminal device held by the relevant nursing staff, thus completing a complete technological loop from signal perception and risk analysis to resource allocation decision-making.

[0050] Preferably, in this embodiment of the invention, the preset intervention conditions include: During the current monitoring period, at least one patient among all patients has a deviation level greater than the preset deviation level threshold, and the maximum risk value among all patients is greater than the preset risk peak threshold.

[0051] Among them, the deviation threshold and the risk peak threshold can be obtained by statistically analyzing historical data, and the data at the 95th percentile of the data distribution can be selected as the threshold for their respective dimensions.

[0052] In summary, this invention utilizes the degree of deviation from historical behavioral characteristics as a quantitative standard for assessing abnormal patient behavior. A graph structure algorithm is used to construct an influence relationship graph among patients, and the risk value borne by each patient is obtained by combining the degree of deviation. Based on the basic intervention cost, the updated intervention cost is adjusted using the accumulated risk value, effectively representing the risk caused by a nursing decision. Furthermore, the correct nurse can be assigned based on the updated intervention cost corresponding to the target patient. This invention, based on patient behavioral characteristics, considers the mutual influence relationships between patients and synchronously updates the intervention cost between nurses and patients, enabling accurate and effective allocation decisions in nursing resource allocation.

[0053] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An AI-driven nursing resource allocation system, characterized in that, The system includes: The behavior analysis module is used to monitor the behavioral characteristics of each patient during the monitoring period. The behavioral characteristics include acoustic characteristics and bed posture characteristics; and to obtain the degree of deviation of the behavioral characteristics from historical behavioral characteristics. The patient impact relationship analysis module is used to construct an impact relationship graph between patients, with each patient as a node. The edge weight between two nodes in the impact relationship graph is obtained from the distance between patients and the Granger causality test results between patients. The risk value of each patient is obtained based on the degree of deviation of each patient and the corresponding edge weight. The nurse intervention cost analysis module is used to retrieve the basic intervention cost between nurses and patients; for any nurse-patient combination, it obtains the cumulative risk value of all patients in the nurse's responsible patient list excluding the patients in the combination; the basic intervention cost is adjusted based on the cumulative risk value to obtain the updated intervention cost; The decision allocation module is used to select the patient with the maximum deviation as the target patient after the patient's risk value and deviation degree meet the preset intervention conditions; and to determine the assigned nurse for the target patient based on the updated intervention cost.

2. The AI-driven nursing resource allocation system according to claim 1, characterized in that, The method for obtaining the acoustic features includes: Acquire the patient's audio signal; identify whether the audio signal is an interactive audio signal or a non-speech human voice signal; if it is an interactive audio signal, extract the spectral entropy and rhythm of the interactive audio signal to obtain acoustic features; if it is a non-speech human voice signal, extract the micro-vibrations and micro-amplitudes of the non-speech human voice signal to obtain acoustic features.

3. The AI-driven nursing resource allocation system according to claim 1, characterized in that, The method for obtaining the bed posture features includes: The pressure signal matrix monitored by the patient's bed is acquired by a matrix of sensors arranged on the bed; the gray-level co-occurrence matrix of the pressure signal matrix is ​​extracted, and the energy and contrast of the gray-level co-occurrence matrix are used as the bed posture features.

4. The AI-driven nursing resource allocation system according to claim 1, characterized in that, The method for obtaining the degree of deviation includes: Each dimension of the behavioral characteristics is Z-standardized in its respective dimension to obtain normalized behavioral characteristics. If the current monitoring period is in the system startup phase, the mean and standard deviation in the Z-standardization process are calculated from all the historical data collected. If the current monitoring period is in the system steady state phase, the mean and standard deviation in the Z-standardization process are calculated from the historical data within the preset time window before the current monitoring period. The L2 norm of the normalized behavioral features is used as the degree of deviation.

5. The AI-driven nursing resource allocation system according to claim 1, characterized in that, The edge weight is the weighted sum of spatial distance weight and historical behavior following relationship weight according to preset weights; The spatial distance weight is obtained from the distance between nodes; The weights of the historical behavior follow-up relationships were obtained from the Granger causality test results.

6. The AI-driven nursing resource allocation system according to claim 5, characterized in that, The method for obtaining the spatial distance weight includes: The spatial distance weight between two patients in the same ward is set to 1, the spatial distance weight between two patients in adjacent wards is set to 0.5, the spatial distance weight between two patients on the same floor is set to 0.2, and the spatial distance weight between two patients in other cases is set to 0.

7. The AI-driven nursing resource allocation system according to claim 5, characterized in that, The methods for obtaining the weights of the historical behavior follow-up relationships include: During the system startup phase, the weights of all historical behavior follow-up relationships are set to 0. During the steady-state phase of the system, for each pair of nodes, a Granger causality test is performed based on the deviation sequence under the preset time window before the current monitoring cycle to obtain the test coefficient. The test coefficient is subtracted from the positive integer 1. If the result is less than 0, the weight of the historical behavior following relationship is set to 0; if the result is greater than 0, the result is used as the weight of the historical behavior following relationship.

8. The AI-driven nursing resource allocation system according to claim 1, characterized in that, The method for obtaining the risk value includes: For each patient, the deviation degree of other patients is used as a weight, and the edge weights are summed in a weighted manner to obtain the exogenous risk coefficient. The sum of the exogenous risk coefficient and the deviation degree of the current patient is used as the risk-bearing value.

9. The AI-driven nursing resource allocation system according to claim 1, characterized in that, The method for obtaining the cost of the update intervention includes: After normalizing the accumulated risk value, the sum of this value and the basic intervention cost is used as the updated intervention cost.

10. The AI-driven nursing resource allocation system according to claim 1, characterized in that, The preset intervention conditions include: During the current monitoring period, at least one patient among all patients has a deviation level greater than the preset deviation level threshold, and the maximum risk value among all patients is greater than the preset risk peak threshold.

Images