An Internet of Things-based intelligent ward control method and system

Abstract

The present invention relates to the technical field of intelligent ward control, and discloses an intelligent ward control method and system based on the Internet of Things, including: collecting characteristic data of the ward, encapsulating the characteristic data in a time series structure to form a multi-dimensional input tensor, and performing channel decoupling processing on the multi-dimensional input tensor; establishing a dual-channel control model, calculating the mutual influence between characteristic data through single-item gating, and performing ward control according to the output of the dual-channel control model; performing asymmetric training on the dual-channel control model according to the calculation results of the single-item gating to realize the adaptive update of the model. The present invention greatly reduces the false alarm rate and resource waste, enhances the interpretability of control decisions and the self-adaptability of the system, and has obvious advantages in terms of safety, practicability and deployment feasibility.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent ward control, and specifically provides an intelligent ward control method and system based on the Internet of Things. Background Art

[0002] In the context of the rapid development of intelligent healthcare, the automation of ward environments and the intelligent monitoring of patients' vital signs have become important directions for the digital upgrade of medical services. Existing intelligent ward systems usually use sensor networks to collect ward environment data and patients' vital sign information, and perform alarm and device linkage control through preset rules or simple models. However, due to the complex sources of vital sign changes, which may be caused by pathological changes or significantly disturbed by environmental changes, traditional methods cannot effectively distinguish the main causes of abnormalities, easily leading to problems such as false alarms, missed alarms, or over-intervention, affecting patient safety and the utilization efficiency of medical resources.

[0003] Currently, common methods based on recurrent neural networks (such as GRU, LSTM) mainly focus on single-channel feature modeling, lacking effective modeling of the causal relationship between vital signs and environmental parameters, and lacking differential optimization for different main causes of abnormalities during the training process, resulting in the model being difficult to balance prediction accuracy and the interpretability of control decisions. Therefore, there is an urgent need for an intelligent ward control method that can simultaneously model the dynamic characteristics of vital signs and environmental parameters, identify the sources of abnormalities, and achieve precise control linkage based on the attribution results. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are:

[0006] To solve the above technical problems, the present invention provides the following technical solution: An intelligent ward control method based on the Internet of Things, including: collecting the characteristic data of the ward, encapsulating the characteristic data in a time series structure to form a multi-dimensional input tensor, and performing channel decoupling processing on the multi-dimensional input tensor;

[0007] Establishing a two-channel control model, calculating the mutual influence between the characteristic data through single-item gating, and performing ward control according to the output of the two-channel control model;

[0008] According to the calculation results of the single-item gating, performing asymmetric training on the two-channel control model to achieve the adaptive update of the model.

[0009] As a preferred solution of the intelligent ward control method based on the Internet of Things according to the present invention, wherein: the characteristic data of the ward includes vital sign data of patients and environmental parameter data in the ward, the vital sign data includes body temperature, heart rate and respiratory rate, and the environmental parameter data includes temperature, humidity and illuminance.

[0010] As a preferred solution of the intelligent ward control method based on the Internet of Things according to the present invention, wherein: the multi-dimensional input tensor includes encapsulating each characteristic data into a sliding window with a length of in a time series structure to form a multi-dimensional input tensor ;

[0011] Perform channel decoupling processing on the multi-dimensional input tensor and divide it into:

[0012] Vital sign input ;

[0013] Environmental perception input ;

[0014] Satisfy ;

[0015] Wherein, Represents the multi-dimensional input tensor; Represents the set of real numbers; Represents the sliding window length; Represents the characteristic data of the ward; Represents the vital sign input; Represents the number of vital sign channels; Represents the environmental perception input; Represents the number of environmental parameter channels.

[0016] As a preferred solution of the intelligent ward control method based on the Internet of Things according to the present invention, wherein: the dual-channel control model includes an input layer, a vital sign layer, an environmental perception layer, a gating regulation layer, a splicing layer and a control decision layer;

[0017] The input layer obtains the characteristic data of the ward, forms a multi-dimensional input tensor and performs channel decoupling processing. The input layer is connected to the vital sign channel and the environmental perception channel. The vital sign channel transmits the vital sign input , and the environmental perception channel transmits the environmental perception input ;

[0018] The environmental perception layer is connected to the environmental perception channel and analyzes the environmental perception input through the GRU model, and outputs an environmental hidden state vector ;

[0019] The gating adjustment layer is connected to the environmental perception layer and calculates a one-way gating factor according to Calculate the one-way gating factor;

[0020] The vital sign layer is connected to the vital sign channel and the gating adjustment layer. It analyzes the vital sign input through the GRU model , takes the one-way gating factor as a dynamic perturbation amount to intervene, updates the candidate hidden state, and outputs the vital hidden state vector ;

[0021] The splicing layer is connected to the environmental perception layer and the vital sign layer, and splices and into a joint state representation vector;

[0022] The control decision layer is connected to the splicing layer. It determines whether to trigger ward control according to the joint state representation vector, judges the main cause of state deviation using the one-way gating factor, and executes a differential control strategy.

[0023] As a preferred solution of the intelligent ward control method based on the Internet of Things described in the present invention, wherein: the gating adjustment layer includes calculating an environmental state change trend term according to the environmental hidden state vector:

[0024]

[0025] Calculate the one-way gating factor:

[0026]

[0027] Take the one-way gating factor as a dynamic perturbation amount and intervene when updating the candidate hidden state inside the vital sign layer to construct a causal fusion vital sign representation:

[0028]

[0029]

[0030] Among them, represents the change amount of the environmental hidden state vector; represents the environmental hidden state vector at time ; represents the environmental hidden state vector at time ; represents the one-way gating factor; represents the Sigmoid activation function; represents the weight matrix of the one-way gating factor; represents the bias vector; represents the vital candidate hidden state vector; represents the hyperbolic tangent activation function; The weight matrix representing the candidate hidden state vector of life; Indicating the moment of the vital sign input; Indicating the moment the weight matrix of the life hidden state vector; Indicating the reset gate; Indicating the moment of the life hidden state vector; Indicating the gating intervention weight matrix; Indicating the update gate; Indicating the Hadamard product.

[0031] As a preferred solution of a smart ward control method based on the Internet of Things according to the present invention, wherein: the ward control includes presetting a control threshold of the joint state representation vector , and the control determination layer detects according to the joint state representation vector . If indicates that the characteristic data of the ward is normal, continue to detect and do not perform ward control;

[0032] If indicates that the characteristic data of the ward is abnormal, trigger the main cause attribution determination;

[0033] The main cause attribution determination includes setting thresholds and of the one-way gating factor, and . When , it is determined as the environmental main cause, and the equipment in the ward is adjusted according to the environmental parameter data. After adjustment, continue to detect within cycles. If indicates that the risk is lifted, if then trigger the hybrid anomaly process;

[0034] When , it is determined as the pathological main cause, trigger a risk alarm and push it to the nurse station;

[0035] When , it is determined as a mixed factor, adjust the equipment in the ward according to the environmental parameter data, trigger a risk alarm at the same time, continue to detect and trigger the main cause attribution determination again in the next cycle.

[0036] As a preferred solution of a smart ward control method based on the Internet of Things according to the present invention, wherein: the asymmetric training includes performing asymmetric training according to the result of the main cause attribution determination, and the loss function is expressed as:

[0037]

[0038] ​Among them, represents the overall loss; represents the local auxiliary loss output from the vital sign channel; represents the local auxiliary loss output from the environmental channel; represents the main classification loss of the control decision output after splicing; 、 represents the weighting factor that dynamically changes according to gated attribution;

[0039] The local auxiliary loss includes adding a small classification head separately at the end of the outputs of the environmental perception layer and the vital sign layer, and calculating the local risk classification probability only using the outputs of this layer;

[0040] For samples of different main cause categories, dynamically set 、 , calculate the local auxiliary loss and the main classification loss respectively, and combine them into the overall loss according to the weighting; ;

[0041] When training samples dominated by environmental main causes, preferentially update the parameters of the environmental perception layer branch;

[0042] When training samples dominated by pathological main causes, preferentially update the parameters of the vital sign layer branch;

[0043] For the gated intervention weight matrix , jointly optimize on all samples, and give higher weights to samples dominated by vital signs.

[0044] An Internet of Things-based intelligent ward control system adopting any of the methods described in the present invention, wherein: a collection module, which collects the characteristic data of the ward, encapsulates the characteristic data in a time series structure to form a multi-dimensional input tensor, and performs channel decoupling processing on the multi-dimensional input tensor;

[0045] A processing module, which establishes a dual-channel control model, calculates the mutual influence between characteristic data through single-item gating, and controls the ward according to the output of the dual-channel control model;

[0046] A training module, which performs asymmetric training on the dual-channel control model according to the calculation results of single-item gating to realize the adaptive update of the model.

[0047] A computer device, comprising: a memory and a processor; the memory stores a computer program, including: when the processor executes the computer program, it implements the steps of any of the methods described in the present invention.

[0048] A computer-readable storage medium, on which a computer program is stored, including: when the computer program is executed by a processor, it implements the steps of any of the methods described in the present invention.

[0049] Advantages of the present invention: By designing a GRU modeling structure with dual-channel decoupling and combining a unidirectional environmental gating mechanism, the present invention effectively realizes the causal separation modeling of changes in vital signs and changes in environmental parameters, and can accurately distinguish the main causes of abnormal patient signs. By introducing an attribution determination and asymmetric training mechanism based on environmental gating factors, the present invention not only improves the accuracy of anomaly detection, but also can select different control strategies according to the main cause of the anomaly, realizing the intelligent linkage of device intervention and medical alarm. Compared with traditional single-channel modeling or unified training methods, the present invention significantly reduces the false alarm rate and resource waste, enhances the interpretability of control decisions and the self-adaptability of the system, and has obvious advantages in safety, practicability and deployment feasibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0051] Figure 1 The overall flowchart of a method for controlling a smart ward based on the Internet of Things provided by an embodiment of the present invention;

[0052] Figure 2 The schematic diagram of a dual-channel control model of a method for controlling a smart ward based on the Internet of Things provided by the second embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be made in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention shall fall within the protection scope of the present invention.

[0054] Embodiment 1, referring to Figure 1 , which is an embodiment of the present invention, provides a method for controlling a smart ward based on the Internet of Things, including:

[0055] S1: Collect the characteristic data of the ward, encapsulate the characteristic data according to the time series structure to form a multi-dimensional input tensor, and perform channel decoupling processing on the multi-dimensional input tensor.

[0056] In the intelligent ward control system of the present invention, firstly, the characteristic data in the ward is systematically collected and standardized to form a structured input for state modeling. The collection of characteristic data mainly focuses on two aspects: one is the vital sign parameters of patients, and the other is the perception parameters of the ward environment. The vital sign parameters include but are not limited to body temperature, heart rate, blood oxygen saturation , respiratory rate, blood pressure, etc., and the environmental perception parameters include but are not limited to indoor temperature, humidity, light intensity, carbon dioxide concentration, noise level, etc. These characteristic data are sourced from the sensor network deployed in the ward and are continuously collected at a fixed sampling frequency to form an original data stream.

[0057] To ensure the temporal continuity of subsequent state modeling and the extraction of dynamic change characteristics, the present invention performs time serialization encapsulation on the collected original characteristic data. Specifically, the data at each time point is combined in chronological order to form a sliding window with a fixed length of . Each window contains consecutive steps of characteristic data, and each characteristic data covers all characteristic channels. Thus, a unified multi-dimensional input tensor is formed, where is the sliding window length, is the overall number of characteristic channels, and each row in the tensor corresponds to a data snapshot at a sampling time step, and each column corresponds to a specific characteristic (such as heart rate, room temperature, etc.).

[0058] Furthermore, to adapt to the dual-channel decoupling modeling mechanism proposed by the present invention, it is necessary to implement a structural division of the characteristic channels inside the multi-dimensional input tensor, that is, channel decoupling processing. Specifically, the input tensor is divided according to the characteristic source to obtain the vital sign input and the environmental perception input , where represents the vital sign input; represents the number of vital sign channels; represents the environmental perception input; represents the number of environmental parameter channels. It satisfies . During the division process, all time steps are kept synchronized and aligned to ensure that at any moment, the vital sign and environmental perception data correspond to the ward state at the same moment.

[0059] In the present invention, the reason for first performing unified encapsulation and then channel decoupling instead of directly processing from the original data is based on the characteristics of multi-source heterogeneous data in the intelligent ward environment and the requirements of subsequent joint modeling. Unified encapsulation can ensure the time synchronization consistency of different data sources and avoid the misalignment distortion of state features caused by factors such as different sampling frequencies and data delays. At the same time, logical channel decoupling after unified encapsulation can separately extract the characteristics of vital sign changes and environmental changes in the subsequent model, and model the dynamic causal relationship between the two through a one-way gating mechanism in the subsequent process, ensuring the accuracy and interpretability of control decision-making inference. If unified encapsulation is not performed, there will be drift between vital signs and environmental characteristics on the time axis, seriously affecting the accuracy of state prediction and main cause attribution judgment based on time series models (such as GRU).

[0060] Therefore, data acquisition, time encapsulation, and channel decoupling are the first basic links of the method of the present invention and are the prerequisite conditions for ensuring the quality of overall state modeling and the reliability of subsequent intelligent control decisions. When designing the present invention, the acquisition characteristics and real-time processing requirements of multi-source heterogeneous data in the intelligent ward are fully considered, and sliding window encapsulation and channel structure decoupling are adopted to ensure the standardization, controllability, and adaptability of data input.

[0061] S2: Establish a dual-channel control model, calculate the mutual influence between feature data through one-way gating, and perform ward control according to the output of the dual-channel control model.

[0062] Furthermore, in order to achieve dynamic perception and accurate attribution of vital sign changes and environmental state changes in the ward, the present invention designs a dual-channel control model for parallel modeling of two types of feature data, calculates the mutual influence between feature data through a one-way gating mechanism, and makes intelligent ward control decisions according to the modeling output.

[0063] Specifically, the dual-channel control model includes an input layer, a vital sign layer, an environmental perception layer, a gating adjustment layer, a splicing layer, and a control determination layer, and there is a clear data flow and function division between each layer.

[0064] The input layer obtains the feature data of the ward, forms a multi-dimensional input tensor and performs channel decoupling processing. The input layer is connected to the vital sign channel and the environmental perception channel, and the vital sign channel transmits the vital sign input , and the environmental perception channel transmits the environmental perception input .

[0065] The environmental perception layer is connected to the environmental perception channel and analyzes the environmental perception input through a GRU model , and outputs the environmental hidden state vector .

[0066] To further capture the impact of environmental changes on vital sign changes, the gating regulation layer is connected to the environmental perception layer, and according to a one-way gating factor is calculated. The one-way gating factor is compressed to the range of 0 to 1 through the Sigmoid function to quantify the potential interference intensity of environmental state changes on the evolution of vital signs.

[0067] The vital sign layer is connected to the vital sign channels and the gating regulation layer. The input of vital signs is analyzed through the GRU model , and the one-way gating factor is introduced as a dynamic perturbation amount to update the candidate hidden state and output the vital hidden state vector . In the process of vital sign modeling, the present invention innovatively introduces the one-way gating factor as a dynamic perturbation amount into the update calculation of the candidate hidden state inside the GRU, so that the modeling of vital sign changes is not only affected by its own historical information, but also can dynamically adjust the state memory path according to the environmental change trend, construct a causal fusion sign representation, and output the updated vital hidden state vector.

[0068] The splicing layer is connected to the environmental perception layer and the vital sign layer, and and are spliced into a joint state representation vector;

[0069] The control decision layer is connected to the splicing layer. It determines whether to trigger ward control according to the joint state representation vector, uses the one-way gating factor to judge the main cause of state deviation, and executes a differentiated control strategy.

[0070] Furthermore, in the specific implementation of the gating regulation layer, first, according to the environmental hidden state vector, the environmental state change trend term of adjacent time steps is calculated:

[0071]

[0072] Then, based on this change trend term, the one-way gating factor is calculated:

[0073]

[0074] Taking the one-way gating factor as a dynamic perturbation amount, it is introduced when updating the candidate hidden state inside the vital sign layer to construct a causal fusion sign representation:

[0075]

[0076] The final vital hidden state vector is fused and updated through the update gate ztz_tzt, and the calculation formula is:

[0077]

[0078] Among them, represents the change amount of the environmental hidden state vector; The environmental hidden state vector representing the moment ; The environmental hidden state vector representing the moment ; Represents the unidirectional gating factor; Represents the Sigmoid activation function; The weight matrix representing the unidirectional gating factor; Represents the bias vector; The candidate life hidden state vector; Represents the hyperbolic tangent activation function; The weight matrix representing the candidate life hidden state vector; Represents the moment Vital sign input; Represents the moment The weight matrix of the life hidden state vector; Represents the reset gate; Represents the moment The life hidden state vector; Represents the gating intervention weight matrix; Represents the update gate; Represents the Hadamard product.

[0079] Compared with the existing single-channel time series modeling (such as ordinary GRU, LSTM), the present invention first embeds the unidirectional influence mechanism of environmental changes on vital sign changes into the internal state update of GRU to achieve causal dynamic fusion; in the control decision-making stage, not only triggers control according to the overall anomaly judgment, but also selects different control paths according to the attribution results, improving the accuracy and interpretability of control decisions; designs a trainable or semi-adaptive main cause attribution threshold mechanism, enabling the system to have the ability to dynamically adjust the attribution judgment criteria with the evolution of data.

[0080] Furthermore, the ward control process includes setting the control threshold of the joint state representation vector. The preset joint state representation vector Control threshold , the control decision-making layer detects according to the joint state representation vector . If Indicates that the characteristic data of the ward is normal, continue to detect and do not perform ward control;

[0081] If Indicates that the characteristic data of the ward is abnormal, trigger the main cause attribution judgment;

[0082] The main cause attribution judgment includes setting the threshold of the unidirectional gating factor and , and , when When it is determined to be the environmental main cause, the equipment in the ward is adjusted according to the environmental parameter data. After the adjustment, continue to detect within cycles. If indicates that the risk is lifted. If then trigger the hybrid anomaly process;

[0083] When it is determined to be the pathological main cause, trigger a risk alarm and push it to the nurse station;

[0084] When it is determined to be a mixed factor, adjust the equipment in the ward according to the environmental parameter data, trigger a risk alarm at the same time, continue to detect and trigger the main cause attribution determination again in the next cycle.

[0085] The present invention can significantly improve the accuracy of patient status anomaly detection and the reliability of main cause determination in the intelligent ward system, reduce the false alarm rate and missed alarm rate, support the execution of the most appropriate control strategy according to different abnormal main causes, thereby improving the comfort of the ward environment and the safety of patients. At the same time, the present invention has good model interpretability and auditability, meeting the actual application requirements of intelligent control systems in the medical industry.

[0086] S3: According to the calculation results of single-channel gating, perform asymmetric training on the dual-channel control model to achieve adaptive update of the model.

[0087] After completing the modeling of ward feature data and main cause attribution determination, an asymmetric training mechanism based on the calculation results of single-channel gating is further proposed, aiming to achieve adaptive optimization and update of the dual-channel control model. By dynamically adjusting the training attention and parameter update ratio of each channel branch, the adaptability and discrimination accuracy of the model to different abnormal main cause samples are improved.

[0088] Specifically, the asymmetric training includes adjusting the training strategy according to the results of main cause attribution determination, and optimizing the model in the form of a specific weighted loss function. The loss function is expressed as:

[0089]

[0090] Among them, represents the overall loss; represents the local auxiliary loss of the output of the sign channel; represents the local auxiliary loss of the output of the environment channel; represents the main classification loss of the output of the control determination after splicing; 、 represent the weighted factors that change dynamically according to gating attribution;

[0091] In order to accurately capture the contribution degree of each channel to the final control decision, a small classification head is separately added at the end of the outputs of the environmental perception layer and the vital sign layer. This classification head only uses the hidden state outputs of their respective layers to independently calculate the local risk classification probability, so as to form the vital sign channel auxiliary loss and the environmental channel auxiliary loss. This can directly measure the contribution degree of each channel to the determination of the deviation of the overall state at the local scale.

[0092] During the training process, according to the main cause category determined by the one-way gating attribution, and are dynamically set. For samples dominated by environmental main causes, is set to give priority to optimizing the branch parameters of the environmental perception layer during the training process; while for samples dominated by pathological main causes, is set to give priority to optimizing the branch parameters of the vital sign layer. Through this dynamic weighting method, the training focus is adaptively adjusted according to the main cause category, improving the sensitivity and recognition accuracy of the model to different deviation types.

[0093] In addition, for the gating intervention weight matrix , an important variable that controls the evolution path of vital signs, the present invention jointly optimizes and updates it on all samples to maintain the stability of the gating adjustment ability. However, when updating the gradient, a higher weight is given to the samples dominated by vital signs to enhance the discriminative ability of the model for pathological abnormal changes, ensuring that even when the environmental change amplitude is small, the potential abnormal trends of the patient's vital signs can be captured in a timely manner.

[0094] By introducing the above asymmetric training mechanism, the present invention can effectively improve the training efficiency, attribution accuracy and abnormal recognition reliability of the dual-channel control model, enabling the intelligent ward system to achieve more accurate, intelligent and interpretable adaptive control in a dynamic and complex environment.

[0095] Example 2, referring to Figure 2 , in an exemplary embodiment, an intelligent ward control system based on the Internet of Things is also provided.

[0096] As Figure 2 shown, it is a schematic diagram of the dual-channel control model structure proposed by the present invention. Among them, the input layer, as the starting point of the model, is responsible for receiving the feature data collected in real time in the ward, including the patient's vital sign parameters (such as body temperature, heart rate, blood oxygen saturation, respiratory rate) and environmental perception parameters (such as indoor temperature, humidity, illuminance, carbon dioxide concentration, etc.). The collected feature data is first encapsulated into a multi-dimensional input tensor according to the structure of the time series sliding window, and channel decoupling is completed at the input layer, and is respectively divided into the vital sign input channel and the environmental perception input channel for subsequent separate modeling and processing.

[0097] The environmental perception channel transmits the decoupled environmental parameter data to the environmental perception layer; the vital sign channel transmits the decoupled vital sign data to the vital sign layer. The channel division ensures the logical independence of the feature data sources and strict synchronization of the time steps, providing a structured basis for subsequent modeling and attribution.

[0098] The environmental perception layer is connected to the environmental perception channel and uses a GRU (Gated Recurrent Unit) network to perform temporal modeling on the environmental perception input data, extract the dynamic features of the environmental state changes, and output the environmental hidden state vector. The environmental hidden state vector not only describes the trend of the environment itself changing over time but also provides basic information for subsequent gated regulation, supporting the modeling of the causal relationship between the potential impact of the environment on the changes in vital signs.

[0099] The gated regulation layer is connected between the environmental perception layer and the vital sign layer. Its main function is to dynamically generate a unidirectional gating factor based on the environmental hidden state vector. The gated regulation layer calculates the change trend term of the environmental hidden state at adjacent time steps, and then applies a weight mapping and a Sigmoid activation function to generate the gating regulation factor. The value range of the unidirectional gating factor is between 0 and 1, which is used to quantify the influence degree of environmental changes on the memory path of vital sign changes. This design enables the model to dynamically adjust the sensitivity and interpretability of vital sign abnormalities according to the environmental change rate.

[0100] The vital sign layer is simultaneously connected to the vital sign channel and the gated regulation layer and uses a GRU structure to extract temporal features of the vital sign input. Based on the traditional GRU update mechanism, the present invention introduces a unidirectional gating factor as a dynamic perturbation quantity in the candidate hidden state update calculation. In this process, the environmental change trend modulates the vital sign hidden state update, thereby realizing the causal dynamic regulation of the vital sign state memory and forming a causal fusion of the vital sign state representation. This mechanism enables the model to distinguish whether the vital sign changes are self-evolving or affected by environmental changes, improving the accuracy of attribution judgment and control decision-making.

[0101] The concatenation layer is connected after the environmental perception layer and the vital sign layer, concatenates the vital sign hidden state vector and the environmental hidden state vector to form a joint state representation vector. The joint state vector combines the dynamic information of the patient's vital signs and the environmental state changes in the ward, serving as an important basis for subsequent control determination.

[0102] The control decision layer is connected to the splicing layer. Using a fully connected neural network and a Sigmoid output unit, it calculates the control probability of the current ward state based on the joint state representation vector, and determines whether there is an abnormal deviation and control needs to be executed. If the control decision output is higher than the set threshold, it enters the main cause attribution process, combines a one-way gating factor to judge the abnormal main cause (environmental main cause, pathological main cause, or mixed main cause), and triggers a differential control strategy accordingly, such as adjusting the ward temperature and humidity, or sending a risk alarm to the nurse station.

[0103] Through the above hierarchical structure and logical flow, the dual-channel control model of the present invention can achieve the closed-loop optimization of anomaly detection, main cause attribution, and intelligent control in the intelligent ward environment, and has high accuracy, strong interpretability, and good system adaptability.

[0104] This embodiment also provides an Internet of Things-based intelligent ward control method, including an acquisition module that acquires the characteristic data of the ward, encapsulates the characteristic data in a time series structure to form a multi-dimensional input tensor, and performs channel decoupling processing on the multi-dimensional input tensor; a processing module that establishes a dual-channel control model, calculates the mutual influence between the characteristic data through one-way gating, and performs ward control according to the output of the dual-channel control model; and a training module that asymmetrically trains the dual-channel control model according to the calculation result of the one-way gating to achieve the adaptive update of the model.

[0105] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0106] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definitional sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or used in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0107] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection portion (electronic device) having one or more wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing as appropriate, and then storing it in a computer memory.

[0108] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An intelligent ward control method based on the Internet of Things, characterized in that, Including: Collect the characteristic data of the ward, encapsulate the characteristic data according to the time series structure to form a multi-dimensional input tensor, and perform channel decoupling processing on the multi-dimensional input tensor; Establish a dual-channel control model, calculate the mutual influence between characteristic data through single-item gating, and perform ward control according to the output of the dual-channel control model; The dual-channel control model includes an input layer, a vital sign layer, an environment perception layer, a gating adjustment layer, a splicing layer, and a control decision layer; The input layer obtains the feature data of the ward, forms a multi-dimensional input tensor and performs channel decoupling processing. The input layer is connected to the vital sign channel and the environmental perception channel. The vital sign channel transmits the vital sign input , and the environmental perception channel transmits the environmental perception input , satisfying ; The environmental perception layer is connected to the environmental perception channel, and analyzes the environmental perception input through the GRU model , and outputs an environmental hidden state vector ; The gating adjustment layer is connected to the environmental perception layer and calculates a unidirectional gating factor according to Calculate the unidirectional gating factor; The vital sign layer is connected to the vital sign channel and the gating regulation layer, and analyzes the vital sign input through the GRU model , intervenes the unidirectional gating factor as a dynamic perturbation quantity, updates the candidate hidden state, and outputs the vital hidden state vector ; The splicing layer is connected to the environment perception layer and the vital sign layer, and splices and into a combined state representation vector; The control decision layer is connected to the splicing layer, determines whether to trigger ward control according to the joint state representation vector, uses the single-direction gating factor to judge the main cause of the state deviation, and executes a differential control strategy; Among them, represents a multi-dimensional input tensor; represents a set of real numbers; represents the sliding window length; represents the characteristic data of the ward; represents the vital sign input; represents the number of vital sign channels; represents the environmental perception input; represents the number of environmental parameter channels; The gating adjustment layer includes calculating the environmental state change trend term according to the environmental hidden state vector: ； Calculating the single-direction gating factor: ； Use the single-direction gating factor as a dynamic disturbance quantity, intervene when updating the candidate hidden state inside the vital sign layer, and construct a causally fused sign representation; ； ； Among them, represents the change amount of the environmental hidden state vector; represents the moment of the environmental hidden state vector; represents the moment of the environmental hidden state vector; represents the unidirectional gating factor; represents the Sigmoid activation function; represents the weight matrix of the unidirectional gating factor; represents the bias vector; represents the life candidate hidden state vector; represents the hyperbolic tangent activation function; represents the weight matrix of the life candidate hidden state vector; represents the moment of the vital sign input; represents the moment of the weight matrix of the life hidden state vector; represents the reset gate; represents the moment of the life hidden state vector; represents the gating intervention weight matrix; represents the update gate; represents the Hadamard product; According to the calculation result of single-item gating, perform asymmetric training on the dual-channel control model to realize the adaptive update of the model.

2. The intelligent ward control method based on the Internet of Things according to claim 1, characterized in that: The characteristic data of the ward includes vital sign data of patients and environmental parameter data in the ward. The vital sign data includes body temperature, heart rate, and respiratory rate. The environmental parameter data includes temperature, humidity, and illuminance.

3. The intelligent ward control method based on the Internet of Things according to claim 2, wherein: The ward control includes a preset combined status representation vector of the control threshold , and the control determination layer detects according to the combined status representation vector . If indicates that the characteristic data of the ward is normal, continue the detection without performing ward control; If The characteristic data of the ward is abnormal, triggering the determination of the main cause attribution; The determination of the main cause attribution includes setting a threshold for the one-way gating factor and , and when , it is determined as the environmental main cause, and the equipment in the ward is adjusted according to the environmental parameter data. After the adjustment, continue to detect within cycles. If indicates that the risk is lifted, and if then trigger the hybrid anomaly process; When it is judged as the main pathological cause, a risk alarm is triggered and pushed to the nurse station; When it is judged as a mixed factor, the equipment in the ward is adjusted according to the environmental parameter data, and at the same time, a risk alarm is triggered, and the detection continues and the main cause attribution determination is triggered again in the next cycle.

4. The intelligent ward control method based on the Internet of Things according to claim 3, wherein: The asymmetric training includes performing asymmetric training according to the result of main cause attribution determination, and the loss function is expressed as: ； Among them, represents the overall loss; represents the local auxiliary loss output by the physical sign channel; represents the local auxiliary loss output by the environmental channel; represents the main classification loss of the control decision output after splicing; , represents the weighting factor that changes dynamically according to the gating attribution The local auxiliary loss includes adding a small classification head separately at the end of the output of the environment perception layer and the vital sign layer, and calculating the local risk classification probability only using the output of this layer; Dynamically set for samples of different main cause categories , , calculate the local auxiliary loss and the main classification loss respectively, and combine them into an overall loss according to the weighted combination ; When training samples dominated by the environmental main cause, preferentially update the branch parameters of the environment perception layer; When training samples dominated by the pathological main cause, preferentially update the branch parameters of the vital sign layer; For the gated intervention weight matrix , optimize jointly over all samples, giving higher weights to the sign-dominated samples.

5. A smart ward control system based on the Internet of Things, which is applied to a smart ward control method based on the Internet of Things described in any one of claims 1 to 4, and is characterized in that, Including, A collection module that collects the characteristic data of the ward, encapsulates the characteristic data according to the time series structure to form a multi-dimensional input tensor, and performs channel decoupling processing on the multi-dimensional input tensor; A processing module that establishes a dual-channel control model, calculates the mutual influence between characteristic data through single-item gating, and performs ward control according to the output of the dual-channel control model; A training module that performs asymmetric training on the dual-channel control model according to the calculation result of single-item gating to realize the adaptive update of the model.

6. A computer device, comprising: A memory and a processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of a smart ward control method based on the Internet of Things as described in any one of claims 1-4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of a smart ward control method based on the Internet of Things as described in any one of claims 1-4 are implemented.

Images