Ice blanket machine dynamic temperature control method based on body temperature feedback

Abstract

The invention relates to the technical field of medical temperature control, and discloses an ice blanket machine dynamic temperature control method based on body temperature feedback. The method comprises the following steps: acquiring body surface temperature sensing data and core temperature simulation data of a target object through body temperature multi-source sensing, and constructing an original body temperature sensing data set; generating a physiological state feature vector according to the data, and performing physiological state evaluation; based on the feature vector, generating a multi-modal body temperature state tensor through multi-source sensing fusion modeling; then, an optimized temperature control strategy is generated according to the multi-modal body temperature state tensor by adopting a reinforcement learning method for dynamic binding of a clinical constraint rule component; performing integrity verification and clinical safety rule conflict detection on the optimization strategy to generate a safety control instruction sequence; and finally, the ice blanket machine is driven to execute temperature regulation and control operation according to the sequence. According to the method, the accuracy and safety of temperature regulation and control of the ice blanket machine are improved, and complex clinical scene requirements are met.

Description

Technical Field

[0001] This invention relates to the field of medical temperature control technology, specifically a dynamic temperature control method for an ice blanket machine based on body temperature feedback. Background Technology

[0002] In clinical medicine, ice blanket machines are widely used as important thermoregulatory devices for cooling patients with high fever and maintaining body temperature after surgery. Their core function is to intervene in the patient's body temperature by regulating the blanket surface temperature to achieve therapeutic or adjunctive therapeutic purposes. However, current temperature control methods for ice blanket machines still have many limitations, making it difficult to meet the precise control needs of complex clinical scenarios.

[0003] Current cooling blanket machines mostly rely on a single surface temperature sensor, which can only acquire temperature data from a localized area of ​​the patient's body. However, the distribution of body temperature is significantly complex, and there is often a large difference between surface temperature and core temperature, especially when the patient is in a special physiological state such as fever, shock, or anesthesia. A single surface temperature reading cannot fully reflect the patient's true body temperature status, easily leading to temperature control decisions that deviate from actual needs. For example, the cooling intensity might be reduced prematurely due to a temporary decrease in surface temperature, while ignoring situations where the core temperature remains at a higher level.

[0004] Regarding temperature regulation strategies, traditional cooling blanket machines mostly employ preset, fixed control modes, such as step-by-step adjustments based on set temperature thresholds. This mode lacks the ability to dynamically adapt to the individual physiological state of patients. Different patients exhibit significant differences in basal metabolic rate, disease severity, and temperature sensitivity, making it difficult for fixed modes to accommodate individual specificities. Furthermore, existing strategies rarely integrate clinical treatment rules systematically. For example, for patients with bleeding tendencies, excessive cooling may increase the risk of coagulation disorders, but current control logic often fails to incorporate such clinical constraints into decision-making, posing potential medical safety risks.

[0005] The existing control process of ice blanket machines lacks a comprehensive safety verification mechanism. Once the control strategy is generated, it directly drives the actuator without real-time monitoring of the strategy's integrity and compatibility with clinical safety rules. In complex clinical environments, factors such as abnormal sensor data, temporary equipment malfunctions, or sudden changes in the patient's physiological state may lead to conflicting or incomplete control commands, affecting temperature control effectiveness and even causing adverse events. For example, when sensor data drift occurs, control commands generated based on erroneous data may cause the blanket temperature to be too low, resulting in discomfort such as shivering in patients. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic temperature control method for an ice blanket machine based on body temperature feedback, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a dynamic temperature control method for an ice blanket machine based on body temperature feedback, the method comprising: Multi-source body temperature sensing acquires surface temperature sensing data and core temperature simulation data of the target object to construct an original body temperature sensing dataset. Physiological state assessment, based on the surface temperature sensing data and core temperature simulation data, generates a physiological state feature vector; Multimodal body temperature state modeling: Based on the physiological state feature vector, a multimodal body temperature state tensor is generated through multi-source sensing fusion modeling. The body temperature regulation strategy is optimized by generating an optimized temperature control strategy based on the multimodal body temperature state tensor using a reinforcement learning method that dynamically binds clinical constraint rule components. Control chain path topology verification: Perform integrity verification and clinical safety rule conflict detection on the optimized temperature control strategy, and generate a safety control instruction sequence. Dynamic temperature control is implemented, driving the ice blanket machine to perform temperature regulation operations according to the safety control command sequence.

[0008] Preferably, the multi-source body temperature sensing includes the following operations: The skin surface temperature time series is collected by a distributed temperature sensor, and environmental temperature and humidity compensation parameters are obtained simultaneously. The core temperature simulation value is generated by the core temperature estimation algorithm, and the original body temperature sensing dataset is constructed by combining the skin surface temperature time series and the environmental temperature and humidity compensation parameters.

[0009] Preferably, the physiological state assessment includes the following operations: Temperature time-series correction is performed on the original body temperature sensing dataset to generate a body surface temperature time-series sequence. The body surface temperature time series is input into the physiological state mapping component, and the core temperature shift feature is generated through the body surface-core temperature correlation model. By integrating the core temperature offset features with environmental temperature and humidity compensation parameters, a physiological state feature vector is generated.

[0010] Preferably, the multimodal body temperature state modeling includes the following operations: A hierarchical attention fusion mechanism was constructed, including a thermoregulation response attention layer, an environmental disturbance inhibition attention layer, and a clinical risk warning attention layer; The body temperature regulation response attention layer extracts the temperature regulation demand features from the physiological state feature vector; The environmental interference suppression attention layer filters out environmental temperature and humidity interference noise characteristics; The clinical risk warning attention layer generates a probability distribution of the risk of sudden body temperature changes. By aggregating the temperature regulation demand characteristics, environmental temperature and humidity interference noise characteristics, and the probability distribution of sudden body temperature changes, a multimodal body temperature state tensor is generated.

[0011] Preferably, the optimization of the body temperature regulation strategy includes the following operations: Construct a reinforcement learning state parameter space, including the current temperature state of the ice blanket, the rate of change of body temperature, and the state of clinical constraint rule components; A multi-layered strategy network architecture is adopted, including an annual protocol generation layer, a clinical dynamic optimization layer, and a real-time execution decision layer; The annual plan generation layer outputs a long-term body temperature management framework protocol. The long-term body temperature management framework protocol is dynamically adjusted through the clinical dynamic optimization layer to generate hourly temperature control schemes. The real-time execution decision layer generates minute-level temperature adjustment commands.

[0012] Preferably, the training of the multi-layer policy network architecture includes the following operations: Design a multi-objective reward aggregation function that integrates temperature offset penalty factor, physical discomfort assessment factor and equipment energy consumption factor; A clinical disturbance simulation function is introduced to simulate abnormal body temperature fluctuations, sensor failures, and sudden environmental events. The optimized temperature control strategy is generated through a phased optimization process, including an initial protocol generation phase, a clinical disturbance adaptation phase, and a multi-objective strategy aggregation phase.

[0013] Preferably, the control chain path topology verification includes the following operations: Construct a control command dependency graph to record the sequential execution dependency of each temperature adjustment command in the optimized temperature control strategy; The conflict nodes between the temperature regulation command and the clinical safety rules component are detected by using a backreference mapping table; A depth-first search algorithm is used to verify the loop risk of the control command dependency graph; Generate a sequence of security control instructions that is conflict-free and loop-free.

[0014] Preferably, the conflict detection of the clinical safety rule component includes the following operations: Define body temperature safety boundary rules, temperature change rate threshold rules, and equipment maximum power limit rules; Iterate through each temperature adjustment command in the optimized temperature control strategy and match the violation identifiers of the body temperature safety boundary rule, temperature change rate threshold rule, and device maximum power limit rule using a bidirectional mapping tool; Delete the temperature adjustment command containing the aforementioned violating identifier.

[0015] Preferably, the generation of the security control instruction sequence includes the following operations: Based on the topological sorting algorithm, conflict-free temperature regulation commands are sorted according to their execution dependencies; Add device startup preheating commands and sensor calibration commands as front-end nodes; Output a sequence of safety control instructions arranged in order.

[0016] Preferably, the dynamic temperature execution includes the following operations: Analyze the temperature target value and execution time window in the safety control command sequence; The power adjustment component of the ice blanket machine compressor and the circulating water flow control valve are driven to gradually approach the target temperature value according to the execution time window; The deviation between the actual temperature value and the target temperature value is fed back in real time to the multi-source body temperature sensing step.

[0017] Compared with the prior art, the beneficial effects of the present invention are: At the level of body temperature perception, the method innovatively employs a multi-source sensing approach, simultaneously acquiring surface temperature sensing data and core temperature simulation data to construct a raw dataset. This dual-dimensional sensing mode overcomes the limitations of traditional single-source surface temperature sensing, enabling a more comprehensive capture of the patient's body temperature information. Surface temperature directly reflects the instantaneous temperature state of the blanket application area, while core temperature simulation data reflects the deep internal body temperature. Combining the two avoids misjudgments caused by biases in a single data point, allowing subsequent temperature control decisions to be based on a more complete foundation of body temperature information.

[0018] The physiological state assessment stage generates feature vectors through in-depth analysis of surface and core temperature data, providing an individualized basis for precise temperature control. Different patients vary in age, disease condition, and underlying physical constitution, resulting in different physiological responses to temperature stimuli. This stage extracts features related to physiological state, transforming abstract body temperature data into quantifiable state indicators. This allows the temperature control system to identify individual patient specificities, such as differentiating the inflammatory response stage in patients with high fever or the metabolic recovery status of postoperative patients, thus guiding subsequent strategy optimization.

[0019] Multimodal body temperature status modeling, by fusing multi-source sensor data to generate tensors, further improves the accuracy of body temperature status representation. Body temperature change is a dynamic process involving the interaction of multiple factors, and single-dimensional features are insufficient to fully describe its complex patterns. Multimodal modeling integrates multi-dimensional information such as surface temperature dynamics, core temperature trends, and physiological characteristic changes into a unified tensor structure. This allows for a more comprehensive characterization of the spatiotemporal distribution and evolution of the patient's body temperature status, providing richer state-based evidence for the generation of regulatory strategies and making these strategies more aligned with the patient's actual body temperature change needs.

[0020] The thermoregulation strategy optimization process employs a reinforcement learning method with dynamic binding of clinical constraint rules, making the regulation strategy more aligned with clinical practice requirements. Reinforcement learning possesses the ability to optimize decisions through dynamic learning, while the introduction of clinical constraint rules sets safety boundaries for the learning process. During strategy generation, the system can refer to clinical treatment guidelines in real time, such as temperature control targets for different diseases and temperature tolerance ranges for special populations. This ensures that the optimized strategy is both dynamically adaptable and adheres to safety principles in clinical practice, avoiding regulatory behaviors that do not conform to medical norms.

[0021] Control chain path topology verification adds a safety layer to temperature control operations. This step performs integrity checks and safety rule conflict detection on the generated optimized strategy, enabling timely identification of potential omissions or content that contradicts clinical safety requirements. For example, if a sudden temperature change instruction is detected in the strategy, its safety can be assessed in conjunction with the patient's cardiovascular status to avoid adverse reactions caused by inappropriate instructions. By generating a sequence of safe control instructions, the reliability of the control operation at the execution level is ensured, reducing risks in clinical applications.

[0022] The dynamic temperature control system operates based on a sequence of safety commands, ensuring the immediacy and accuracy of temperature control. Upon receiving a safety command, the system quickly activates the cooling blanket machine to adjust the blanket surface temperature, allowing the control measures to be applied to the patient promptly. This rapid response capability is particularly important in clinical settings where body temperature fluctuates rapidly, enabling timely monitoring of patient temperature changes and maintaining the temperature within the target range to ensure the effectiveness of temperature control. Attached Figure Description

[0023] Figure 1 This is a timing diagram of the dynamic temperature control method for the ice blanket machine based on body temperature feedback as described in this invention. Figure 2 A flowchart for physiological state assessment; Figure 3 Flowchart for modeling multimodal body temperature states; Figure 4 This is a flowchart for controlling the topology verification of the chain path. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1 This invention provides a dynamic temperature control method for an ice blanket machine based on body temperature feedback, the method comprising: The system comprises six core modules: multi-source body temperature sensing, physiological state assessment, multimodal body temperature state modeling, body temperature regulation strategy optimization, control chain path topology verification, and dynamic temperature execution. The multi-source body temperature sensing module collects surface temperature data through distributed sensors and constructs a raw dataset by combining environmental parameters with a core temperature estimation algorithm. The physiological state assessment module generates feature vectors reflecting physiological states through temperature time-series correction and correlation models. The multimodal body temperature state modeling module employs a hierarchical attention mechanism to integrate temperature regulation needs, environmental interference suppression, and clinical risk warning features. The body temperature regulation strategy optimization module generates hierarchical temperature control strategies based on a reinforcement learning framework. The control chain path topology verification module ensures the clinical safety of the instruction sequence through dependency analysis and conflict detection. The dynamic temperature execution module ultimately translates safety instructions into specific control operations of the cooling blanket machine, forming a closed-loop control system.

[0026] Example 1: See Figure 2 The system encompasses two core processing stages: body temperature data acquisition and physiological state transition. A distributed temperature sensor array utilizes medical-grade platinum resistance thermometers and is deployed according to standard clinical temperature measurement protocols across six key anatomical areas: the forehead, left and right carotid arteries, left and right axillae, and the xiphoid process of the sternum. Sensor nodes are networked via Bluetooth Low Energy 5.0, employing a time-division multiple access (TDMA) mechanism for synchronous sampling at a frequency of 10 times per second. Each sensor unit integrates a digital temperature and humidity composite chip, simultaneously acquiring body surface temperature data and capturing real-time air temperature and relative humidity parameters at a distance of 2 cm from the skin surface, with a humidity measurement range covering 20% ​​to 95% RH. The data preprocessing unit automatically performs three-point calibration, eliminating sensor baseline drift errors and generating a timestamped raw temperature data stream.

[0027] The core temperature estimation algorithm uses a recurrent neural network as its computational framework, constructing a three-layer long short-term memory network with 128 hidden units. The input layer receives four types of parameters: the mean temperature of six regions on the body surface, the maximum temperature gradient between adjacent sensors, the heart rate variability index synchronously acquired from the medical monitor, and the environmental thermal resistivity calculated based on the thickness of the patient's clothing. The hidden layers employ a gating mechanism to filter transient interference signals, and the output layer generates the core temperature estimate and adds a dynamic confidence interval calculation. This confidence interval is constructed based on a sliding window statistical method, with a window width set to 300 consecutive sampling points, generating a probability distribution model based on standard deviation calculation. The original body temperature sensing dataset uses a time-division storage structure, generating a data packet every 60 seconds. Each packet contains 26 fields, including a timestamp array accurate to milliseconds, the six sensor node numbers and their corresponding temperature-humidity values, the core temperature estimate, and the upper and lower bounds of the confidence interval. The data storage format uses a lightweight binary encoding scheme, with each data packet having a fixed length of 512 bytes.

[0028] During the physiological state assessment phase, the data correction engine initiates a Kalman filter procedure. The filter's process noise covariance matrix is ​​dynamically adjusted based on the patient's activity level: when the accelerometer detects a change in body position, the noise tolerance is automatically increased by a factor of 3; in the resting state, the baseline covariance parameter is used. The temperature time-series correction module establishes a dual processing channel: the main channel performs a sliding window averaging on the raw data, while the sub-channel uses median filtering for verification. The 30-second wide sliding window advances in 5-second steps, recalculating the arithmetic mean of the data within the window at each shift. When the difference between the main channel and the sub-channel data exceeds 0.2 degrees Celsius, an abnormal data replacement mechanism is triggered.

[0029] The body surface-core temperature correlation model is implemented using a Gaussian process regression framework. The input feature set contains parameters across five dimensions: the 120-second moving average of the corrected body surface temperature, the slope of the body surface temperature change over the last 10 seconds, the confidence interval width of the core temperature estimate, the arithmetic mean of ambient humidity readings, and the historical trend of the difference between ambient and body surface temperatures. During model training, a hyperparameter optimization module is loaded, and the optimal combination of kernel functions is determined through Bayesian tuning. The model's real-time output includes two key indicators: the core temperature offset and its rate of change, with a data update frequency of twice per second.

[0030] After the environmental temperature and humidity compensation parameters enter the feature fusion stage, a dynamic weighting algorithm is initiated. This algorithm automatically assigns weight coefficients based on the stability of environmental sensor readings: when the environmental temperature fluctuation is below 0.1℃ for three consecutive samples, it is assigned a weight value of 0.3; when a minute-level change in humidity exceeds 5%, the weight value automatically drops to 0.1. The fusion processor integrates data streams from three sources: core offset features from the temperature correlation model, environmental temperature and humidity weighted values, and the third derivative features of the body surface temperature itself. The feature vector builder ultimately generates a 128-dimensional floating-point array, with each element storing specific physiological state parameters in a fixed order. The first 64 dimensions carry temperature features, including temperature values ​​of various regions of the body surface, temperature difference matrices between different anatomical points, and temperature change acceleration parameters; the middle 32 dimensions store core temperature-related parameters, including offset, trend confidence, and abnormal fluctuation counts; the last 32 dimensions record environmental interaction parameters, including heat flux estimates, effective radiation area coefficient, and humidity influence factors. This feature vector is updated once per second and transmitted to downstream processing units through a double-buffering mechanism to avoid processing delays caused by data transmission.

[0031] A data integrity monitoring system operates synchronously throughout the entire processing chain. A numerical boundary checker is installed at the temperature acquisition layer to discard abnormal data exceeding the physiological range of 34℃-42℃. Cyclic redundancy check (CRC) technology is employed at the transmission layer, with a 16-bit checksum appended to each data frame. A null value detection unit is deployed at the feature vector generation layer; when missing data is detected, a historical data imputation mechanism is automatically activated, with the imputation algorithm selecting either linear extrapolation or mean imputation based on data type differences. The entire processing flow adopts a time-triggered architecture, with all processing nodes synchronized via a precision clock to ensure that the time alignment accuracy of data at each stage is controlled within 20 milliseconds.

[0032] Example 2: See Figure 3 This study focuses on the multi-dimensional modeling of body temperature and the generation mechanism of hierarchical optimization strategies. The body temperature modeling module constructs a three-layer parallel attention processing architecture. The body temperature regulation response layer is configured with eight independent self-attention computation units, each equipped with a weighted network consisting of 512 neural nodes. These units use a key-value query mechanism to process different subsets of data from physiological state feature vectors, focusing on the 15th to 45th dimensions of the feature vectors that reflect the heat flux of the body surface. Self-attention computation is executed every 200 milliseconds, outputting a temperature regulation demand index matrix. The matrix has a dimension of 8x64, with each row representing the intensity of control demand in different body regions.

[0033] The environmental interference suppression layer is implemented as a gated recurrent structure, consisting of three stacked gated recurrent units with 128, 64, and 32 hidden units respectively. This structure receives all 128-dimensional feature vector inputs while maintaining a 120-dimensional environmental noise template library in memory. A template matcher calculates the cosine similarity between the current environmental parameters and historical noise templates in real time, dynamically loading the three closest templates to form an adaptive filter. Filtering parameters include temperature drift compensation coefficients and environmental decoupling factors, with an operation frequency of 10Hz. Three sets of output gates connected through a weight-sharing mechanism ultimately generate a dimensionality-reduced anti-interference feature vector, compressing the dimension to 16.

[0034] The clinical risk warning layer is implemented as a probability generation network, with input ports receiving the body temperature regulation demand index matrix and filtered environmental feature data. The risk predictor comprises four parallel fully connected networks. Each network's output layer uses the sigmoid activation function to calculate probability estimates for hypothermia risk, febrile seizure risk, sudden temperature drop risk, and equipment overload risk. The probability output range is limited to 0 to 1, with a resolution of 0.001. Each risk probability is updated independently, with a response time of 0.5 seconds. The risk classifier includes a priority weighting mechanism; when multiple risks occur simultaneously, the weight of the equipment overload risk is automatically increased by 40%.

[0035] The multimodal tensor synthesizer receives the processing results from a three-layer attention mechanism: a 64-dimensional regulatory demand vector from the thermoregulation layer, 16-dimensional purification features from the environmental inhibition layer, and a 4-dimensional probability vector from the risk warning layer. The three types of data are projected onto a unified high-dimensional space through tensor product operations. The fusion processor employs cascaded convolution operations, configured with 3x3 cores and a stride of 1, and outputs a 32-channel feature map after applying three convolution transformations. The tensor reconstruction unit ultimately generates a 16x16x16 third-order tensor structure, with tensor elements stored in floating-point format and a dynamic range of -10.0 to +10.0.

[0036] In the regulation strategy generation module, the state space of the reinforcement learning environment is precisely configured with 53 observed variables. The ice blanket temperature status employs a five-level coding mechanism: the 34℃ to 40℃ range is divided into 12 sub-intervals with a width of 0.5℃, and each sub-interval has a status flag. The body temperature change rate parameter records the average gradient of change over the most recent 30 seconds with an accuracy of 0.01℃ / min. The clinical rule component status contains 19 binary indicators to track the activation status of various safety constraints in real time.

[0037] The annual treatment plan generation layer is implemented as a three-input deep neural network. Input sources include patients' historical body temperature fluctuation patterns, a database of environmental seasonal features, and clinical prescription constraints. The network structure contains seven fully connected layers with 256, 128, 64, 32, 16, 8, and 8 nodes respectively. The output layer generates 365 sets of temperature control baseline values, each set containing 96 preset temperature values ​​for a single day, with each preset value having a resolution of 0.1℃. The baseline curve undergoes data smoothing every minute, using cubic spline interpolation to correct for data jump points.

[0038] The clinical dynamic optimization layer employs a dual-delay policy gradient algorithm, configured with four independent value networks and two policy networks. Each value network layer has 128 nodes, and the policy networks use a 256-node hidden layer. The optimizer activates a policy update every 60 minutes, with inputs including real-time physiological state data, annual baseline curve offset, and equipment operating status logs. The optimizer output is a 240-element adjustment matrix, with each element corresponding to the temperature correction for the next 15 minutes, with an accuracy of 0.05℃.

[0039] The real-time execution decision layer constructs a fine-grained control branch of the policy gradient framework, achieving second-level response based on the nearest neighbor policy optimization method. The state encoder acquires a multimodal body temperature state tensor every 15 seconds, compressing it into a 32-dimensional state vector through a feature extraction network. The execution policy network employs a two-layer, 256-node fully connected structure, with its output connected to three independent action channels: a temperature target value channel with a resolution of 0.01℃; a regulation rate channel with a setpoint range of 0.1-5.0℃ / min; and an expected settling time channel outputting command values ​​between 5 and 300 seconds in seconds. The action selector introduces a random exploration factor through Gaussian sampling, with the standard deviation set to 5% of the target temperature difference.

[0040] Example 3: See Figure 4 It covers the policy network training mechanism and the security control instruction generation system. The reinforcement learning framework configures a multi-objective reward aggregation function. :

[0041] This function updates the parameter values ​​at a discrete time step t. Represents the actual measured core temperature value, with an accuracy maintained at 0.01℃; This indicates the clinically set target temperature value, which is allowed to fluctuate within the range of 36.5℃ to 37.5℃; It is the Localized Thermal Discomfort Index calculated according to the ISO 7730 standard, derived by combining the body surface temperature gradient and the rate of sweat evaporation; This is the normalized value of the instantaneous power consumption of the ice blanket compressor, with a benchmark of 2000W rated power. The weighting coefficients employ an adaptive mechanism. It automatically adjusts based on the range of body temperature deviation, with a baseline value of 1.25 and an increase in the coefficient of magnitude of 0.3 for every 0.5°C increase. Maintain a constant value of 0.8; The value is dynamically adjusted within the range of 0.05-0.2 based on the equipment's operating mode.

[0042] The network training process employs a phased optimization workflow. The initial protocol generation phase loads a historical case database containing 2000 complete thermoregulation records for different patient conditions. The state transition model is constructed using a Markov decision process, with a transition probability matrix of 53×53 dimensions. The policy network pre-training performs 15000 iterations, sampling 128 sets of scenario data per iteration. The optimizer uses the RMSprop algorithm with an initial learning rate of 3×10^{-5}. This phase outputs the basic policy parameter set, stored in sectors 1-1024 of non-volatile memory.

[0043] A virtual training environment was constructed during the clinical perturbation adaptation phase. The body temperature abnormality fluctuation event simulator employed a piecewise random walk model: a maximum drift amplitude of ±0.2℃ was set under normal physiological conditions; when marked as pathological fluctuations, the drift amplitude increased to ±1.5℃. The sensor failure event injection module was designed with two failure modes: a random failure mode that shielded 10%-50% of sensor nodes using a Bernoulli distribution; and a regional failure mode that specified the continuous shielding of sensor groups in specific anatomical regions. The environmental mutation event generator applied a ±5℃ step disturbance every 300 steps, with a duration randomly distributed between 30-180 seconds. The perturbation training executed a three-loop mechanism: first, the strategy parameters were fine-tuned for 2000 iterations in a stable environment; then, three types of perturbations were injected progressively for 5000 iterations each; finally, a perturbation combination training was performed for 10000 iterations, with a total training step size of 3.2 × 10^5.

[0044] The parameter fusion process is initiated during the multi-objective strategy aggregation phase. The value network constructs four parallel evaluation channels: a body temperature control channel calculates the mean absolute error within a 30-second sliding window; a comfort evaluation channel records the cumulative value of the thermal discomfort index; an energy consumption monitoring channel calculates the total power consumption of the compressor; and a safety compliance channel tracks the number of safety constraint violations. The fusion controller employs a feature-cascaded connection method, generating value evaluation vectors through a three-layer, 256-node fully connected network. The policy network gradient update executes a double-buffering mechanism: the online network collects experience samples every 15 milliseconds and stores them in a replay buffer; the target network performs a soft update operation every 3000 steps, with a smoothing coefficient τ=0.005.

[0045] The control chain path topology verification module constructs a directed graph of instructions. Each temperature regulation instruction is modeled as a graph node, with node attributes including: target temperature value (resolution 0.01℃), execution time window (accuracy 0.1 seconds), and clinical rule association identifier (128-bit mask). Dependency edges are divided into two types: sequential dependency edges are represented by unidirectional solid lines to indicate a mandatory order relationship; resource mutual exclusion edges are represented by bidirectional dashed lines to indicate device conflict relationships. The dependency graph is allowed to contain a maximum of 256 nodes; if the threshold is exceeded, the graph segmentation algorithm is automatically activated.

[0046] The reverse reference mapping system employs a three-level index structure. The main index table stores the mapping relationship between temperature command parameter combinations and clinical rule addresses, with a hash bucket depth of 16. The conflict detector performs seven rule verifications when traversing each command: whether the target temperature exceeds the 34.0-40.0℃ boundary; whether the single temperature adjustment difference exceeds 0.8℃; whether the minute-level temperature change rate is higher than 0.5℃ / min; whether the cumulative temperature difference of continuous adjustment commands exceeds 3℃; whether the maximum power of the equipment exceeds 2200W; whether the compressor's continuous operating time is below the safety threshold; and whether the water flow rate is within the 10-300ml / min tolerance range. The violation identifier encoding rule is as follows: each binary flag corresponds to a specific rule violation status, forming a 16-bit conflict code.

[0047] The loop risk detection module implements an optimized depth-first search. The graph traversal is set to a maximum depth of 100 layers, with 16KB of memory pre-allocated in the stack. The search process maintains two auxiliary data structures: a real-time access marker array to record node access status; and a loop path cache to store the node sequence that may form a loop. When the length of the detected instruction sequence exceeds 12, a parallel search is initiated: the main thread executes a standard depth-first traversal; auxiliary threads use a breadth-first strategy to verify critical node dependencies. When a potential loop is detected, a reconstruction engine is triggered: first, it attempts to add virtual barrier nodes to cut the loop; if a complex loop exists, a sequence reconstruction algorithm is initiated to regenerate the acyclic topology sequence. Finally, the verified control instruction sequence is labeled with a timestamp, a cyclic redundancy check code (32-bit polynomial), and a digital signature (based on elliptic curve algorithm), forming a complete security control instruction data packet.

[0048] Example 4: Focusing on Clinical Safety Rule Validation and Instruction Sequence Reconstruction Mechanisms. The clinical safety rule engine deploys dual processing channels: a forward inference channel monitors physical parameters in real time, and a reverse validation channel processes control instructions in batches. The body temperature safety boundary rule divides the 34℃ to 42℃ range into 32 sub-ranges with a width of 0.25℃, and each sub-range is assigned a hazard level score. Below 34℃, a hypothermia risk flag is activated, and the score increases linearly with decreasing temperature; above 39℃, a hyperthermia risk flag is triggered, and the score gradient increases by 40%. The rule engine maintains a hazard level lookup table, which dynamically updates the hazard thresholds based on real-time vital sign data, with an update frequency of 5 times per second.

[0049] The temperature change rate threshold rule is implemented as a dual-detection system. The main detector uses a 60-second wide annular buffer to store recent temperature values ​​and calculates the rate of change acceleration parameter once per second. When the standard deviation within the detection window exceeds the set value of 0.15℃, the auxiliary detection channel is automatically activated: this channel is configured with a more sensitive 40-second moving window, combined with first-order derivative analysis to identify abnormal fluctuations. The equipment power limiting rule achieves millisecond-level real-time monitoring through current transformers and voltage sampling circuits, establishing a power-temperature mapping table. This mapping table is initially calibrated during the equipment preheating phase, recording typical power consumption curves of the compressor under different temperature setpoints.

[0050] Table 1: Safety rule conflict detection parameters.

[0051]

[0052] The conflict detector initiates a seven-step analysis process during command verification. Taking command #207 (temperature adjustment) as an example: this command plans to reduce the ice blanket temperature from 35.8℃ to 34.2℃ within 180 seconds and includes three key parameters. The rule engine first analyzes the target temperature of 34.2℃, finding it falls within the 8th danger zone (34.0-34.25℃), corresponding to a low-temperature danger level of 0.78. Next, it calculates the temperature change rate: (35.8-34.2) / 3 = 0.533℃ / min, exceeding the 0.5℃ / min threshold. Finally, it predicts the compressor power requirement: a lookup in the mapping table reveals that this condition requires 2310W of power, exceeding the rated value of 2200W. The detector generates a conflict code 0x1A (binary 011010), indicating that rules 1, 3, and 5 are simultaneously triggered and violated.

[0053] The instruction sequence reconstruction engine initiates a topology sorting process. The initial instruction sequence contains 89 nodes. A depth-first scan reveals a loop: node E (cooling down to 36.0℃) depends on node G (preheating to 38.0℃), while node G requires node E to complete before it can execute. The reconstruction process first removes the hot node #207 that violates the three rules; then, it uses a layered grouping algorithm for topology sorting: dividing the remaining 86 nodes into 12 levels, allowing nodes at the same level to execute in parallel. The level division is based on instruction time window overlap analysis: if the execution periods of two instructions overlap by more than 30% and the target temperature difference is greater than 1.0℃, they are assigned to different levels.

[0054] The device initialization module adds two types of pre-heating commands: the compressor preheating command comprises three stages: for the first 30 seconds, it heats up to 32℃ at a rate of 0.3℃ / second; then it maintains a constant temperature for 120 seconds; and finally, for the last 60 seconds, it adjusts to the initial operating temperature of 35.5℃. The sensor calibration command activates the closed-loop test program: it applies stimuli at three standard temperature points (30.0℃, 35.0℃, and 40.0℃) for 30 seconds, acquiring the response curves of each sensor. The calibration data analyzer calculates the offset compensation value; for example, if a forehead sensor detects a reference temperature of 35.0℃ and displays 35.3℃, then a compensation parameter of -0.3℃ is written.

[0055] The safety control command sequence generation adopts a framed structure. Each frame contains a 256-byte payload: the first 8 bytes are the command sequence number and version identifier; the next 96 bytes store the command group after topology layering (12 bytes per command); the following 64 bytes record the device initialization parameters; and the last 32 bytes contain safety verification data. The frame structure is output through a time-triggered mechanism: a 5-millisecond communication window is reserved in the compressor's operating cycle, and this window appears every 200 milliseconds. The data transmitter injects 3 frames of data when the communication window opens, forming a data pipeline.

[0056] The digital signature system employs a hierarchical key architecture. Frame-level signatures use a 256-bit elliptic curve algorithm, with the private key stored in a hardware security module. Taking a specific instruction frame as an example: the frame data with sequence number #8912 generates a digest value of 0x7A3F. The signature generator uses the device-specific private key to encrypt the digest, outputting a 512-bit signature code. System-level certificates are bound to the hospital ID and device serial number, with a 24-hour certificate renewal cycle and an automatic renewal process initiated 30 minutes before expiration. A redundant verification channel is established at the transport layer: when the main channel transmits instruction data via the CAN bus, the backup channel synchronously transmits the instruction sequence number and simplified temperature setpoint parameters via an infrared link, enabling cross-channel data verification.

[0057] Example 5: Focusing on the precise control and real-time feedback mechanism of the temperature execution system of the ice blanket machine. The temperature command parser adopts a dual-mode working architecture. In standard mode, it directly reads the target parameters from the security control command sequence, while in security mode, it additionally activates digital signature verification and redundant channel comparison procedures. When the command decoding unit processes each frame of 256 bytes of data, it first extracts the target temperature value field, which is stored as a 16-bit unsigned integer with a resolution of 0.01℃ and an effective range covering 30.00℃ to 45.00℃. The execution time window parameter includes two 32-bit values: a start timestamp and a duration. The time base is synchronized to a medical-grade atomic clock, with a minimum scheduling unit of 10 milliseconds. When the time window span is detected to exceed 300 seconds, it is automatically divided into multiple sub-tasks for batch execution.

[0058] The compressor power regulation system consists of three parallel controllers. The main controller implements an improved proportional-integral-derivative algorithm with a sampling period of 50 milliseconds, outputting a pulse-width modulation signal to drive an insulated-gate bipolar transistor. The auxiliary controller runs a fuzzy logic algorithm, dynamically adjusting control parameters based on recent temperature trends, updating 20 times per second. A safety monitor continuously tracks the compressor coil temperature and radiator status, intervening immediately when an overheating risk is detected. The power output stage employs four-quadrant control technology, allowing for 0.1% accuracy in both forward cooling and reverse heating regulation. Under typical operating conditions, the compressor takes 90 seconds to cool from 35.0°C to 34.0°C, during which time the power output exhibits a parabolic characteristic of first increasing and then decreasing.

[0059] The circulating water flow control system is equipped with a high-precision stepper motor-driven valve assembly. The valve position controller receives 0-100% opening commands, with each 1% opening corresponding to a 200-microstep movement of the stepper motor. The actual water flow calibration curve is stored in non-volatile memory. The flow feedback system includes dual measurement channels: a Hall effect sensor and a turbine flow meter. The data fusion processor outputs the corrected flow value at a frequency of 250Hz. When a cooling command is executed, the valve first opens rapidly to the estimated position (e.g., from 30% to 65%), then enters a fine-tuning phase, with each adjustment not exceeding 2%. The heating process employs the opposite strategy: conservative adjustment initially followed by accelerated compensation later, to avoid sudden temperature changes that could cause patient discomfort.

[0060] The real-time feedback system establishes a multi-level data acquisition network. The main channel transmits equipment operating parameters via the controller area network bus, including 12 types of data such as actual compressor power, instantaneous water temperature, and flow meter readings, sending a complete status frame every 100 milliseconds. The backup channel uses Bluetooth Low Energy to transmit a simplified dataset containing only key temperature values ​​and anomaly flags; the transmission interval is dynamically adjustable, with a minimum setting of 50 milliseconds. The data synchronizer compares the timestamp differences between the two channels, triggering a clock calibration procedure when the deviation exceeds 20 milliseconds.

[0061] The deviation analysis module executes a triple verification process. The first level of verification calculates the instantaneous difference between the target temperature and the actual temperature, maintaining an accuracy of 0.01℃. The second level of verification analyzes the cumulative temperature difference integral within a 300-second sliding window to identify systematic deviation trends. The third level of verification combines ambient temperature and patient surface characteristics to predict the theoretically permissible temperature difference range. When three consecutive samples detect a deviation exceeding the threshold, the system automatically responds in stages: a 0.3℃ deviation triggers parameter fine-tuning; a 0.5℃ deviation initiates a recalculation of the control algorithm; and a deviation above 1.0℃ executes an emergency protocol, suspending the current command and reinitializing the control system.

[0062] Data transmission optimization employs differential coding technology. Temperature data streams use 35.0℃ as a baseline value, with subsequent transmissions only recording changes, achieving a data compression rate of 60%. Timestamps use relative encoding, transmitting the complete time in the first frame and millisecond-level offsets in subsequent frames. The data packet structure is meticulously designed: the header contains a 4-byte synchronization word and a 2-byte length identifier; the payload is encapsulated in TLV format; and a 32-bit cyclic redundancy check (CRC) code is appended to the packet tail. The network scheduler manages communication time slots for multiple devices, ensuring that transmissions from the cooling blanket machine, monitor, and central control console do not interfere with each other, with the maximum end-to-end latency controlled within 80 milliseconds.

[0063] The anomaly handling mechanism comprises five recovery levels. Level 1 anomalies involve transient communication interruptions, in which the system automatically switches to a backup channel and retransmits the three most recent frames of data. Level 2 anomalies involve single sensor failures, triggering a data replacement algorithm from adjacent sensors. Level 3 anomalies manifest as actuator response delays, triggering a timeout retry mechanism with a maximum of three attempts. Level 4 anomalies correspond to severe parameter deviations, executing a safety protocol to abort the current operation and rollback to the most recent stable state. Level 5 anomalies involve multiple system failures, immediately cutting off actuator power, activating audible and visual alarms, and sending an emergency interrupt signal to the medical monitoring system. Each anomaly level corresponds to a specific recovery script, stored in tamper-proof memory, and can be forcibly invoked via a physical switch.

[0064] The device status visualization interface presents a multi-layered information structure. The basic layer displays the current temperature setpoint, actual value, and change curve, with a refresh rate of 10 frames per second. The middle layer displays device parameters such as compressor power percentage, water flow rate, and remaining operating time. The advanced layer requires authorization and provides diagnostic information such as the internal status of the control algorithm, network quality indicators, and anomaly logs. All display elements have been optimized for human factors, with key parameters using high-contrast color coding. When numerical changes exceed thresholds, the display automatically enlarges and adds dynamic border alerts. The operation log is stored in a circular buffer structure, recording the most recent 1000 important events, including command execution status, mode switch times, and anomaly handling records, supporting quick retrieval by time range or event type.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic temperature control method for an ice blanket machine based on body temperature feedback, characterized in that, Includes the following steps: Multi-source body temperature sensing acquires surface temperature sensing data and core temperature simulation data of the target object to construct an original body temperature sensing dataset. Physiological state assessment, based on the surface temperature sensing data and core temperature simulation data, generates a physiological state feature vector; Multimodal body temperature state modeling: Based on the physiological state feature vector, a multimodal body temperature state tensor is generated through multi-source sensing fusion modeling. The body temperature regulation strategy is optimized by generating an optimized temperature control strategy based on the multimodal body temperature state tensor using a reinforcement learning method that dynamically binds clinical constraint rule components. Control chain path topology verification: Perform integrity verification and clinical safety rule conflict detection on the optimized temperature control strategy, and generate a safety control instruction sequence. Dynamic temperature control is implemented, driving the ice blanket machine to perform temperature regulation operations according to the safety control command sequence.

2. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 1, characterized in that, The multi-source body temperature sensing includes the following operations: The skin surface temperature time series is collected by a distributed temperature sensor, and environmental temperature and humidity compensation parameters are obtained simultaneously. The core temperature simulation value is generated by the core temperature estimation algorithm, and the original body temperature sensing dataset is constructed by combining the skin surface temperature time series and the environmental temperature and humidity compensation parameters.

3. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 2, characterized in that, The physiological state assessment includes the following operations: Temperature time-series correction is performed on the original body temperature sensing dataset to generate a body surface temperature time-series sequence. The body surface temperature time series is input into the physiological state mapping component, and the core temperature shift feature is generated through the body surface-core temperature correlation model. By integrating the core temperature offset features with environmental temperature and humidity compensation parameters, a physiological state feature vector is generated.

4. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 3, characterized in that, The multimodal body temperature state modeling includes the following operations: A hierarchical attention fusion mechanism was constructed, including a thermoregulation response attention layer, an environmental disturbance inhibition attention layer, and a clinical risk warning attention layer; The body temperature regulation response attention layer extracts the temperature regulation demand features from the physiological state feature vector; The environmental interference suppression attention layer filters out environmental temperature and humidity interference noise characteristics; The clinical risk warning attention layer generates a probability distribution of the risk of sudden body temperature changes. By aggregating the temperature regulation demand characteristics, environmental temperature and humidity interference noise characteristics, and the probability distribution of sudden body temperature changes, a multimodal body temperature state tensor is generated.

5. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 4, characterized in that, The optimization of the body temperature regulation strategy includes the following operations: Construct a reinforcement learning state parameter space, including the current temperature state of the ice blanket, the rate of change of body temperature, and the state of clinical constraint rule components; A multi-layered strategy network architecture is adopted, including an annual protocol generation layer, a clinical dynamic optimization layer, and a real-time execution decision layer; The annual plan generation layer outputs a long-term body temperature management framework protocol. The long-term body temperature management framework protocol is dynamically adjusted through the clinical dynamic optimization layer to generate hourly temperature control schemes. The real-time execution decision layer generates minute-level temperature adjustment commands.

6. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 5, characterized in that, The training of the multi-layer policy network architecture includes the following operations: Design a multi-objective reward aggregation function that integrates temperature offset penalty factor, physical discomfort assessment factor and equipment energy consumption factor; A clinical disturbance simulation function is introduced to simulate abnormal body temperature fluctuations, sensor failures, and sudden environmental events. The optimized temperature control strategy is generated through a phased optimization process, including an initial protocol generation phase, a clinical disturbance adaptation phase, and a multi-objective strategy aggregation phase.

7. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 6, characterized in that, The control chain path topology verification includes the following operations: Construct a control command dependency graph to record the sequential execution dependency of each temperature adjustment command in the optimized temperature control strategy; The conflict nodes between the temperature regulation command and the clinical safety rules component are detected by using a backreference mapping table; A depth-first search algorithm is used to verify the loop risk of the control command dependency graph; Generate a sequence of security control instructions that is conflict-free and loop-free.

8. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 7, characterized in that, The conflict detection of the clinical safety rules component includes the following operations: Define body temperature safety boundary rules, temperature change rate threshold rules, and equipment maximum power limit rules; Iterate through each temperature adjustment command in the optimized temperature control strategy and match the violation identifiers of the body temperature safety boundary rule, temperature change rate threshold rule, and device maximum power limit rule using a bidirectional mapping tool; Delete the temperature adjustment command containing the aforementioned violating identifier.

9. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 8, characterized in that, The generation of the security control command sequence includes the following operations: Based on the topological sorting algorithm, conflict-free temperature regulation commands are sorted according to their execution dependencies; Add device startup preheating commands and sensor calibration commands as front-end nodes; Output a sequence of safety control instructions arranged in order.

10. The dynamic temperature control method for an ice blanket machine based on body temperature feedback according to claim 9, characterized in that, The dynamic temperature execution includes the following operations: Analyze the temperature target value and execution time window in the safety control command sequence; The power adjustment component of the ice blanket machine compressor and the circulating water flow control valve are driven to gradually approach the target temperature value according to the execution time window; The deviation between the actual temperature value and the target temperature value is fed back in real time to the multi-source body temperature sensing step.

Images