Newborn incubator environment regulation method based on multi-sensor fusion

Abstract

The application provides a new-born incubator environment adjusting method based on multi-sensor fusion, comprising: collecting temperature, humidity, airflow state, radiation state and actual output of environment executing mechanism in the incubator, loading control watermark in stable state and safety boundary, extracting response direction, response delay, response gain and recovery time between environment executing mechanism and environment sensor, and constructing verified causal response matrix; generating non-interchangeable environment paths with same cumulative adjustment amount and different operation sequences according to the causal response matrix, and calculating output loop area, path end output difference, recovery time difference and control watermark reservation degree according to path response; thus, candidate abnormal sensors and common mode contaminated sensor groups are identified, internal environment change, external environment change and environment change with uncertain source are distinguished, and a trusted sensor set and a trusted environment change source record are formed.

Description

Technical Field

[0001] This invention relates to the field of medical device environmental control technology, and in particular to a method for regulating the environment of a neonatal incubator based on multi-sensor fusion. Background Technology

[0002] Neonatal incubators collect environmental data from temperature, humidity, and airflow sensors, and control the operation of heaters, humidifiers, circulating fans, and radiation regulation units to maintain a suitable neonatal care environment. Current technologies typically use multi-sensor averaging, fixed-weight fusion, or outlier removal methods to determine environmental conditions.

[0003] However, during incubator operation, phenomena such as water vapor condensation, disinfectant residue adhesion, uneven local airflow, and heat accumulation on the chamber walls may occur, causing common-mode deviations in the same direction to appear on multiple similar environmental sensors. These deviations are difficult to identify using conventional outlier detection and are easily mistaken for changes in the actual environment. Furthermore, there is a coupling relationship between heating, humidification, air supply, and radiation regulation; different regulation sequences may produce different dynamic responses. Existing methods typically only compare measurement results before and after regulation, making it difficult to distinguish between sensor contamination responses and the lag in actual environmental conditions such as heat accumulation and moisture absorption within the chamber.

[0004] Furthermore, changes in the incubator environment can be caused by either active adjustments from the environmental actuators or by external events such as the opening of doors or control ports. Existing control methods lack joint verification of the actual working relationship of the environmental actuators, the reliability of environmental sensors, and the sources of environmental changes. This makes it easy to implement unnecessary adjustments based on abnormal measurement results, leading to environmental fluctuations and increased energy consumption.

[0005] Therefore, this invention proposes a method for regulating the environment of a neonatal incubator based on multi-sensor fusion. The information disclosed in the background section is only for enhancing understanding of the background of this disclosure and may therefore contain prior art information that is not common knowledge to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for regulating the environment of a neonatal incubator based on multi-sensor fusion, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for regulating the environment of a neonatal incubator based on multi-sensor fusion includes the following steps: S1. Collect temperature, humidity, airflow, radiation, and actual output of environmental actuators inside the incubator to establish the initial environmental state; superimpose control watermarks within the stable state and safety boundary to form watermark response data; S2. Extract the response direction, response delay, response gain and recovery time from the watermark response data, construct the initial causal response matrix, and repeatedly verify it to form a verified causal response matrix and a set of trusted response channels; S3. Based on the verified causal response matrix, generate non-exchangeable environment paths that start from a common initial environmental state, call the same environmental actuators, have the same cumulative adjustment amount, and run in different orders. After execution, a non-exchangeable response feature set is formed. S4. Identify candidate abnormal sensors and common-mode contamination sensor groups based on the non-exchange response feature set. Combine control watermarks and causal response relationships to distinguish between internal environmental changes, external environmental changes, and environmental changes with uncertain sources, forming a set of credible sensors and a record of credible sources of environmental changes. S5. Reconstruct the state of the trusted environment using a set of trusted sensors, separate the internal environmental changes and external environmental disturbances based on the source of the trusted environmental changes, optimize the strategy, generate and execute optimized adjustment commands, and update the control closed loop based on the actual environmental response.

[0008] S1 specifically includes: collecting temperature, humidity, airflow, radiation, and actual output of the environmental actuator within the incubator; synchronizing the collected data with time and marking anomalies; determining the initial environmental state and baseline output under stable conditions; determining the control watermark safety boundary based on the initial environmental state, preset environmental range, and response slope; optimizing the watermark amplitude, period, phase offset, duration, and loading sequence using the bat algorithm to obtain the control watermark parameter set; and superimposing the control watermark onto the baseline adjustment command within the stable state and control watermark safety boundary, recording the environmental sensor measurements, watermark loading commands, and actual output of the environmental actuator to form watermark response data.

[0009] S2 specifically includes: performing benchmark correction and cross-correlation analysis on the watermark response data, extracting response direction, response delay, response gain, and recovery time to form a watermark response feature set; using environmental actuators as inputs and environmental sensors as outputs, writing the watermark response features into the response channels, and filtering valid response channels based on relevant thresholds, minimum identifiable gain, physical action direction, and maximum allowable response time to construct an initial causal response matrix; comparing response direction, response delay, response gain, and recovery time through multiple repeated verification windows to calculate causal confidence, filtering reliable response channels, and writing the median of the response features and the weights of the matrix units into the initial causal response matrix to update and obtain the verified causal response matrix, thus forming a set of reliable response channels.

[0010] S3 specifically includes: screening coupled environment actuators based on the verified causal response matrix and the set of reliable response channels, determining the common initial environment state and the common target environment state, and generating non-exchangeable environment paths with the same cumulative adjustment amount but different running order; controlling the operation of the environment actuators according to the non-exchangeable environment paths, superimposing control watermarks on the path adjustment amounts, collecting environmental sensor measurements, actual output amounts, and path stage information, forming a path response sequence, and performing path reset; performing delay compensation, stage matching, and normalization processing on the path response sequence, calculating the output loop area, path endpoint output difference, recovery time difference, current response delay, current response gain, and control watermark retention degree, forming a non-exchangeable response feature set.

[0011] S4 specifically includes: establishing a normal baseline distribution based on the non-exchange response feature set, calculating path anomaly scores, screening candidate anomaly sensors, and identifying common-mode pollution sensor groups based on the output loop direction, the output difference direction at the path endpoint, and feature similarity; calculating the source matching degree by combining the verified causal response matrix, current response delay, current response gain, control watermark retention degree, and actual output of the environmental actuator, distinguishing between internal environmental changes, external environmental changes, and environmental changes with uncertain sources; calculating the comprehensive credibility based on path credibility, source credibility, and data quality scores, classifying environmental sensors as credible, abnormal, and pending verification, forming a set of credible sensors, a set of abnormal sensors, a set of sensors pending verification, and a record of credible environmental change sources.

[0012] S5 specifically includes: weighted fusion of measured values ​​in the environmental control area based on a trusted sensor set, causal response matrix, and trusted environmental change source records to reconstruct the trusted environmental state, and separating internal environmental changes, external environmental disturbances, and uncertain source disturbances; using the trusted environmental state as the initial prediction value, encoding the target output of the environmental actuator as candidate adjustment combinations, and using the Bat Algorithm to optimize the strategy under constraints of environmental safety range, output range, rate of change, and state uncertainty to generate optimized adjustment instructions; executing the optimized adjustment instructions, verifying the adjustment response based on the deviation between the environmental response and the predicted results of the causal response matrix, triggering response channel verification when the response does not match, and updating the control closed loop based on the trusted environmental state.

[0013] The beneficial effects of this invention are as follows: This invention, by superimposing control watermarks onto environmental actuators within safety boundaries, can identify the actual response direction, response delay, and response gain between the environmental actuators and environmental sensors, thereby improving the accuracy of causal response relationships. By constructing non-exchangeable environmental paths with the same cumulative adjustment but different operating sequences, it can amplify the path differences caused by condensation films, deposits, etc., affecting environmental sensors, providing a valid basis for common-mode pollution identification.

[0014] This invention, through joint analysis of output loop differences, path endpoint differences, and recovery time differences, can identify common-mode deviations of multiple similar environmental sensors, avoiding misinterpretation of real environmental changes due to mutual corroboration of abnormal sensors. By correlating the response characteristics of different types of reliable environmental sensors, it can distinguish between environmental sensor contamination and real environmental path memory caused by enclosure heat storage, surface moisture absorption, and airflow mixing lag, reducing the probability of misjudgment.

[0015] This invention, by matching control watermarks, causal response matrices, actual outputs of environmental actuators, and the status of hatches or operating ports, can distinguish between internal environmental changes, external environmental changes, and environmental changes of uncertain origin, thus improving the reliability of determining the source of environmental changes. By reconstructing a reliable environmental state through weighted fusion of trusted sensors and employing the bat algorithm for strategy optimization, it can reduce unnecessary adjustments caused by abnormal measurements and external disturbances, thereby improving the stability, safety, and energy efficiency of incubator environmental regulation. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the neonatal incubator environment adjustment method of the present invention; Figure 2 This is a schematic diagram of the system structure of the neonatal incubator environment regulation method of the present invention. Detailed Implementation

[0017] 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.

[0018] Example: Figure 1 and Figure 2 As shown, this embodiment provides a method for adjusting the environment of a neonatal incubator based on multi-sensor fusion, including the following steps: S1. Collect temperature, humidity, airflow, radiation, and actual output of environmental actuators inside the incubator to establish the initial environmental state; superimpose control watermarks within the stable state and safety boundary to form watermark response data; S2. Extract the response direction, response delay, response gain and recovery time from the watermark response data, construct the initial causal response matrix, and repeatedly verify it to form a verified causal response matrix and a set of trusted response channels; S3. Based on the verified causal response matrix, generate non-exchangeable environment paths that start from a common initial environmental state, call the same environmental actuators, have the same cumulative adjustment amount, and run in different orders. After execution, a non-exchangeable response feature set is formed. S4. Identify candidate abnormal sensors and common-mode contamination sensor groups based on the non-exchange response feature set. Combine control watermarks and causal response relationships to distinguish between internal environmental changes, external environmental changes, and environmental changes with uncertain sources, forming a set of credible sensors and a record of credible sources of environmental changes. S5. Reconstruct the state of the trusted environment using a set of trusted sensors, separate the internal environmental changes and external environmental disturbances based on the source of the trusted environmental changes, optimize the strategy, generate and execute optimized adjustment commands, and update the control closed loop based on the actual environmental response.

[0019] S1 specifically includes the following sub-steps: S110. Establish multi-source environmental data and initial environmental conditions. According to the incubator structure diagram, sensor installation diagram, and equipment factory configuration table, divide the interior of the incubator into the head adjacent area, torso adjacent area, foot adjacent area, air supply channel, and return air channel. Record the type, number, installation area, sampling period, range, and permissible error of each environmental sensor. Environmental sensors include temperature sensors, humidity sensors, airflow sensors, and radiation sensors; environmental actuators include heaters, humidifiers, circulating fans, and radiation conditioning units.

[0020] Temperature sensors are located near the head, torso, and feet, as well as in the air supply and return ducts. Humidity sensors are located at least near the torso and in the return duct. Airflow sensors are located at least in the air supply duct, near the infant, and in the return duct. Radiation sensors are located at the measurement position facing the infant and near the transparent cover to measure radiant heat flux density, in W / m³. 2 .

[0021] The system synchronously collects measurements from various environmental sensors, the opening and closing status of the hatches, the opening and closing status of the control ports, and the actual output of the environmental actuators. The actual output of the heater is obtained from the power feedback interface, the actual output of the humidifier is obtained from the atomization duty cycle or water supply feedback interface, the actual output of the circulating fan is obtained from the speed feedback interface, and the actual output of the radiation regulation unit is obtained from the power feedback interface.

[0022] All data are timestamped using a unified clock and resampled to a 1-second sampling period. When the original sampling period is less than 1 second, the average value within the corresponding 1 second is taken. When the continuous missing time does not exceed 2 seconds, linear interpolation of adjacent valid values ​​is used. When the continuous missing time exceeds 2 seconds, the missing value is retained.

[0023] The measurement sequence is processed using a sliding median filtering method with a window length of 7 sampling points. When the deviation of the measured value from the sliding median exceeds 3 times the absolute deviation of the median, the measured value is marked as an abnormal jump value, but the original value is retained.

[0024] A stable state is defined as follows: the hatch and operating ports remain closed; the temperature range does not exceed 0.2℃, the humidity range does not exceed 2%RH, and the airflow state range does not exceed 10% of the absolute value of the corresponding baseline average; and there are no fault indicators in the actual output of each environmental actuator. When calculating the airflow state range ratio, if the absolute value of the airflow baseline average is lower than a preset minimum airflow value, the preset minimum airflow value is used as the denominator. This preset minimum airflow value is the greater of the minimum effective measurement value of the airflow sensor and 10% of the average value of the no-load stable airflow.

[0025] The average value of each effective measurement and the actual output within the steady state is determined as the initial environmental state and the baseline output, respectively, and environmental sensor installation configuration, multi-source environmental data sequence, environmental actuator operation record and steady state measurement variance are formed.

[0026] S120. Determine the control watermark parameter set and safety boundaries. A control watermark is a low-amplitude, zero-mean, uniquely identified, and finite-duration detection sequence superimposed on the environmental actuator's baseline adjustment command, used to identify the dynamic response relationship between the environmental actuator and the environmental sensor. Based on the device's preset environmental range, initial environmental state, the environmental actuator's rated output range, and the response slope obtained from factory calibration, determine the upper limit of the control watermark amplitude for the j-th environmental actuator.

[0027] in, This represents the upper limit of the control watermark amplitude for the j-th environmental actuator. This represents the set of environmental variable types affected by this environmental enforcement agency. This represents the safety margin between the current state of the q-th type of environment variable and the nearest permissible boundary. This represents the maximum response slope of the j-th environmental actuator to the q-th environmental variable, obtained from factory calibration or no-load testing. This represents a safety factor ranging from 0.3 to 0.5. This indicates that positive numbers with a denominator of 0 should be prevented.

[0028] The control watermark parameter set includes watermark amplitude, period, phase offset, duration, and loading order. The complement of the maximum correlation coefficient between different control watermark sequences is used as the watermark separability, and the bat algorithm is employed to search the control watermark parameter set. Each artificial bat's position vector corresponds to one set of parameters. The fitness function is defined as follows:

[0029] in, Indicates fitness. Indicates the watermark separability. This represents the maximum normalized change of an environmental variable. Indicates the number of times the environmental execution agency has switched to normalized mode. , , These are non-negative weighting coefficients that sum to 1 (e.g., the default value is 1 when there is insufficient data). =0.5、 =0.3、 =0.2).

[0030] The population size is set to 30, and the maximum number of iterations is set to 100. Candidate parameters that violate the preset environmental range, rated output range, minimum retention time, or upper limit of the control watermark amplitude are judged as infeasible. The process terminates when the optimal fitness changes by less than 0.001 after 10 consecutive iterations. The output is the control watermark parameter set that satisfies the safety constraints and has the highest separability, as well as the control watermark safety boundary.

[0031] For example, when the initial temperature is 36.5℃, the allowable temperature range is 36.0℃ to 37.0℃, the maximum response slope of the heater is 0.04℃ / W, and the safety factor is 0.4, the corresponding upper limit of the control watermark amplitude is 5W.

[0032] S130, Load the control watermark and generate watermark response data. The control watermark is loaded only when the stable state determined in S110 is established, the hatch and operating port are closed, the environmental variables do not enter the boundary adjacent area equivalent to 10% of the width of the preset environmental range from the allowable boundary, and the environmental actuator has no fault or disabled indicator.

[0033] The baseline adjustment command of the j-th environmental actuator is added to its control watermark sequence point by point to form a watermark loading command. When the addition result exceeds the allowable output range of the environmental actuator, the truncated data is not used for identification; instead, the control watermark amplitude is reduced and reloaded. The data for the 60 seconds before the control watermark loading is used as the baseline window, and the duration of the control watermark is used as the loading window. Data is continuously collected after the control watermark stops until the relevant environmental variables re-enter the baseline fluctuation range, forming a recovery window.

[0034] The actual output is compared with the watermark loading command. If the actual output deviation does not exceed 2% of the rated output, it is marked as valid execution; otherwise, an execution anomaly flag is set. The environmental sensor number, measured value, measured value quality flag, environmental actuator number, reference output, instantaneous control watermark value, watermark loading command, actual output, watermark number, window number, hatch status, operating port status, amplitude limit flag, and execution anomaly flag are recorded according to a unified timestamp to form watermark response data.

[0035] If, during the watermark loading process, a hatch opens, an operating port opens, an unplanned adjustment command is issued, or continuous data loss exceeds 2 seconds, the current loading segment is terminated and marked as invalid. Watermark response data is output to S210; the actual output of the environmental actuator also serves as the data source for S420 to identify internal environmental changes and for S510 to calculate the amount of internal environmental change.

[0036] For example, if the heater's reference output is 40W and the instantaneous value of the control watermark is 3W, the watermark loading command is 43W; if the actual output is 42.8W and there are no other abnormalities, it is considered to be validly executed.

[0037] S2 specifically includes the following sub-steps: S210. Extract control watermark response features. Read the watermark response data generated in S130, and select only the loading segments where the control watermark is effectively loaded, the environmental sensor measurement value quality indicator is valid, there are no execution anomalies or amplitude limiting indicators, and the hatch and operation port remain closed.

[0038] The control watermark sequence of the j-th environmental actuator is used as... The measurement sequence of the i-th environmental sensor is used as... The average value of the valid measurements within the reference window is used as the reference value. Calculate the measurement increment:

[0039] in, This represents the change of the i-th environmental sensor relative to the baseline state at a uniform sampling time t.

[0040] Divide the difference between adjacent measurement increments within the reference window by a 1-second sampling period to obtain the reference change slope, and determine the 95th percentile of the absolute value of the reference increment as the reference fluctuation threshold for the i-th environmental sensor. Calculate within the range of 0 to the maximum permissible response time. and The time offset at which the normalized cross-correlation coefficient has the largest absolute value is determined as the candidate response delay. The maximum permissible response time is based on the equipment's factory test record; if no factory test record is available, 1.2 times the 95th percentile of the no-load commissioning response time is used.

[0041] Calculate the response gain:

[0042] in, This represents the response gain from the j-th environmental actuator to the i-th environmental sensor, with the sign indicating the direction of the response. This prevents positive numbers with a denominator of 0. From the start of the effective loading of the self-control watermark, when the measurement increment exceeds 3 consecutive sampling points... And the direction of change is the same as When they are consistent, the first sampling time that meets the condition is determined as the first valid response time; the measured response delay calculated from this time is... When the difference exceeds 2 sampling periods, the corresponding segment is marked as having an uncertain response.

[0043] The maximum absolute value of the measurement increment within the loading window and delay compensation window is determined as the response peak value; when the slope of the measurement increment for 10 consecutive sampling points is lower than the 95th percentile of the absolute value of the baseline slope, the average value of the measurement increment in that interval is determined as the stable response amplitude; the time required for the measured value to re-enter the baseline fluctuation range after the control watermark stops and maintain it for 10 consecutive sampling points is determined as the recovery time. A watermark response feature set is formed, including response direction, response delay, response gain, response peak value, stable response amplitude, recovery time, maximum cross-correlation coefficient, and data quality identifier.

[0044] For example, if the temperature increment exceeds 0.03℃ continuously from the 17th to the 19th second after the heater controls the watermark loading and the direction is positive, then the first effective response time is the 17th second.

[0045] S220. Construct the initial causal response matrix. The causal response matrix is ​​a data structure that records the dynamic response relationship between each input and each output after the active loading of the control watermark, with the environmental actuator as the input and the environmental sensor as the output.

[0046] The matrix cell in the i-th row and j-th column corresponds to the response channel from the j-th environmental actuator to the i-th environmental sensor, and contains the environmental actuator number, environmental sensor number, response direction, response delay, raw response gain, normalized response gain, peak response value, stable response amplitude, recovery time, maximum cross-correlation coefficient, channel status, and data quality identifier. The raw response gain retains the corresponding physical units, and the normalized response gain is only used for comparison between different response channels.

[0047] A physical action direction table is established based on the incubator structural design data, environmental actuator factory test data, and no-load commissioning data; for response channels where the action direction of circulating fans, local temperature, etc., changes with the installation area, the no-load commissioning results are used as the direction reference.

[0048] The relevant thresholds are determined based on the distribution of the maximum cross-correlation coefficients of loaded and unloaded control watermark segments during no-load commissioning, ensuring that the proportion of unloaded segments being misidentified as valid responses does not exceed 5%; the minimum identifiable gain is taken as the 5th percentile of the absolute value of the gain of the normal response channel of the same type.

[0049] If the maximum cross-correlation coefficient reaches the relevant threshold, the absolute value of the response gain reaches the minimum identifiable gain, the response direction conforms to the physical action direction table, the response delay does not exceed the maximum allowable response time, the data missing ratio does not exceed 5%, and there are no uncertain responses or execution anomalies, the corresponding matrix unit is marked as a valid response channel; otherwise, the invalid reasons such as unidentified watermark, insufficient response, abnormal direction, excessive delay, or insufficient data quality are recorded. An initial causal response matrix is ​​formed from all matrix units, and a set of valid response channels is formed from the valid matrix units.

[0050] For example, when the maximum cross-correlation coefficient between the heater and temperature sensor T03 is 0.82, the response gain is 0.012℃ / W, and the response delay is 17s, and these values ​​reach the corresponding thresholds, the response channel is marked as valid; when the maximum cross-correlation coefficient of another temperature sensor is only 0.31, it is marked as watermark unrecognized.

[0051] S230. Verify and update the causal response matrix. Perform no less than 3 repeated verification windows for the effective response channels. Each window must meet the steady-state conditions of S110. The interval between adjacent windows must be at least one recovery time from the corresponding environmental actuator to the farthest environmental sensor, and the same set of control watermark parameters must be used. When the difference between the reference output of the environmental actuator in different windows exceeds 10% of the rated output, verify according to the reference output interval.

[0052] For each valid response channel, causal confidence is calculated based on the watermark recognition result, response direction, response latency, response gain, and recovery time of each repeated validation window:

[0053] in, Indicates the credibility of causation. Indicates the proportion of responses that are consistent in direction. , and These represent the standard deviations of response delay, response gain, and recovery time, respectively. This represents the average response gain. and These represent the allowable response delay fluctuation scale and the recovery time fluctuation scale, respectively. to The sum is 1.

[0054] Each weight is determined by the no-load commissioning data; when there is insufficient data, an equal-weighted initial value is used (i.e., ...). to (All values ​​are set to 0.25). The minimum confidence threshold is set to the 5th percentile of the causal confidence of the normal response channel, and the verification threshold is set to 80% of the minimum confidence threshold.

[0055] Response channels with causal confidence levels not lower than the minimum confidence threshold are included in the set of trusted response channels, and the matrix unit weights are adjusted accordingly. Set as Response channels between the two thresholds are marked as response channels to be reviewed, and are not included in the path execution mechanism screening of S310 and the state prediction of S520, and will be re-verified in the next identification cycle; response channels below the review threshold are eliminated.

[0056] The response direction, median response delay, median response gain, median recovery time, causal confidence, and matrix unit weights of the trusted response channels are written into the initial causal response matrix to form a verified causal response matrix. This matrix, along with the set of trusted response channels, is output to S310, S420, S510, S520, and S530. If the response delays of the three verification windows for the same response channel are 16s, 17s, and 18s, and the response gains are 0.011℃ / W, 0.012℃ / W, and 0.013℃ / W, respectively, then 17s and 0.012℃ / W are taken as the verified median response delay and median response gain, respectively.

[0057] S3 specifically includes the following sub-steps: S310, Generate a non-exchangeable environmental path. A non-exchangeable environmental path refers to an environmental state change process that starts from a common initial environmental state, calls the same environmental actuators, applies the same cumulative adjustment amount, and reaches a common target environmental state, but the order in which the environmental actuators operate is different. Read the initial environmental state and environmental sensor installation configuration formed in S110, the control watermark parameter set and security boundary formed in S120, and call the empirically verified causal response matrix and trusted response channel set formed in S230.

[0058] For any two environmental actuators, when they act on the same environmental sensor through a reliable response channel, or when they act on environmental variables that have physical coupling relationships with temperature and humidity, temperature and airflow state, or radiation state and temperature respectively, they are identified as candidate coupled environmental actuators, and the coupling strength is calculated:

[0059] in, This represents the coupling strength between the j-th environmental actuator and the k-th environmental actuator. This refers to a set of environmental sensors that simultaneously form a reliable response channel with both of them. and This represents the weight of the corresponding matrix unit.

[0060] When the coupling strength reaches the 50th percentile of the candidate combination coupling strength during no-load commissioning, the corresponding environmental actuator is selected as the path actuator. Environmental data is continuously collected for 60 seconds before the start of both the first and second non-switched environmental paths; the difference between the baseline values ​​of each environmental variable before the start of the two paths does not exceed the corresponding baseline fluctuation threshold determined in S210. When the difference between the baseline output of each environmental actuator does not exceed 5% of the rated output, a common initial environmental state is determined.

[0061] The common target environment state is determined based on the common initial environment state, the equipment setting target, and the control watermark safety boundary. All target values ​​are within the preset environment range and maintain a safe distance from the nearest allowable boundary that is not less than the maximum environmental change that the control watermark may cause.

[0062] The cumulative adjustment of the j-th environmental actuator in a single path is:

[0063] in, Indicates the cumulative adjustment amount. This represents the actual output of the nth sampling point. Indicates the reference output quantity. The sampling period is represented by N, and the number of sampling points along the path is represented by N.

[0064] Two paths are generated with the same adjustment range, holding time, and cumulative adjustment amount, but in reverse order. The cumulative adjustment amount deviation of the same environmental actuator in the two paths must not exceed 5%. For example, the first path first increases the heater output by 4W and holds it for 40s, then increases the humidifier atomization duty cycle by 5% and holds it for 40s; the second path executes in the reverse order.

[0065] S320. Execute the non-switched environment path and collect the path response sequence. Divide each path into an initial stabilization phase, a first environment actuator adjustment phase, an inter-phase maintenance phase, a second environment actuator adjustment phase, a common target environment state maintenance phase, and a recovery phase, and set path phase identifiers for each phase.

[0066] The control path actuator is controlled according to the operating sequence determined in S310, using the formal path adjustment amount as the baseline adjustment command, and superimposed with the control watermark determined in S120. If the calculated superimposed output exceeds the allowable output range of the environmental actuator or the safety boundary of the control watermark, the formal path adjustment amount is reduced by the same proportion and the cumulative adjustment amount is recalculated; if the equal quantity condition is still not met after adjustment, the current path pair is canceled.

[0067] The path response sequence is formed by synchronously recording path identifier, path stage identifier, unified timestamp, environmental sensor number and measurement value, measurement value quality identifier, environmental actuator number, baseline output, formal path adjustment, instantaneous value of control watermark, watermark loading command, actual output, cumulative adjustment, hatch status, operating port status, abort identifier and abnormal identifier.

[0068] After the first non-switched environment path is completed, each path actuator is restored to the baseline output. When the temperature, humidity, airflow, and radiation conditions re-enter the baseline fluctuation range of the common initial environmental state and remain there continuously for 60 seconds, the path reset is considered complete. The maximum reset time is twice the median maximum value of the recovery time of the trusted response channels participating in the path. If the path is not reset after the maximum reset time, the current path pair is abandoned.

[0069] If any environmental variable exceeds the preset environmental range, a hatch or control port is opened, an environmental actuator malfunctions, the actual cumulative adjustment deviates from the planned cumulative adjustment by more than 5%, or the reliable response channel required to form the path fails during path execution, path execution will be immediately stopped and the path will be marked as invalid. Valid path response sequences will be output to S330, while invalid path response sequences will only be retained as part of the running record.

[0070] S330. Construct a non-commutative response feature set. Read two valid path response sequences, match the corresponding stages according to the path stage identifier, and perform delay compensation on the measurement sequences of each environmental sensor according to the median response delay determined in S230. Then, resample each stage into the same number of normalized adjustment progress points.

[0071] The normalized path output of the i-th environmental sensor is:

[0072] in, This indicates the normalized path output. Represents the original measurement value. This represents the baseline value under the common initial environmental conditions. This represents the baseline fluctuation threshold, and t represents the uniform sampling time. This indicates that positive numbers with a denominator of 0 should be avoided.

[0073] The normalized outputs of the same environmental sensor at corresponding progress points in two paths are connected to form a closed loop. The closed loop integral is calculated using a trapezoidal integral, and its absolute value is used as the area of ​​the output loop. The positive and negative values ​​are respectively defined as the positive loop and the negative loop. When the absolute value is lower than the 95th percentile of the area of ​​the unpolluted unloaded path loop, it is determined to have no significant direction.

[0074] Calculate the difference between the measured values ​​of the two path endpoints to obtain the path endpoint output difference. If the deviation of any path endpoint from the common target environmental state exceeds the corresponding baseline fluctuation threshold, that path is marked as invalid as the endpoint. The absolute difference in the time required for the two paths to re-enter the common initial environmental state baseline fluctuation range after completion is defined as the recovery time difference. .

[0075] Using the same calculation method as S210, the current response delay is calculated from the path response sequence. and current response gain The maximum normalized cross-correlation coefficient between the delayed-compensated environmental sensor output and the corresponding control watermark is determined as the control watermark retention level. .

[0076] A non-commutative response feature set is generated, including environmental sensor number, output loop area, output loop direction, output difference at the end of the path, deviation of each end of the path, recovery time difference, current response delay, current response gain, control watermark retention degree, valid path identifier, combination of path actuators and matrix unit weights, and output to S410 and S420.

[0077] For example, if the humidity endpoints of the two paths are 63.0%RH and 60.8%RH respectively and form a significant homing curve, the corresponding features will be used as the basis for identifying candidate anomalies in S410.

[0078] S4 specifically includes the following sub-steps: S410: Identify candidate abnormal sensors and common-mode contamination sensor groups. Read the non-exchangeable response feature set formed in S330, and select only data with valid paths, no invalid endpoint markers, and where both non-exchangeable environmental paths have completed reset.

[0079] A normal baseline distribution is established based on the combination of environmental sensor types and path actuators. The baseline data comes from the effective path data under uncontaminated sensor conditions during the factory no-load commissioning of the incubator, the path data confirmed to be normal after sensor cleaning or calibration, and the path data verified as reliable by S430 in the historical identification cycle. During operation, only reliable sensor data is used to update the normal baseline distribution.

[0080] The 95th percentile values ​​of the output loop area, the absolute value of the output difference at the path endpoint, and the recovery time difference in the normal baseline distribution are respectively determined as the output loop area threshold, the path endpoint output difference threshold, and the recovery time difference threshold. The path anomaly score for the i-th environmental sensor is then calculated.

[0081] in, Indicates path anomaly score, , and These represent the output loop area, the difference between the output values ​​at the path endpoints, and the difference in recovery time, respectively. Quantities with the subscript 'th' indicate the corresponding threshold. to The sum is 1.

[0082] Each weight is determined based on no-load commissioning data and pre- and post-cleaning comparative test data; equal initial weights are used when insufficient data is available. An environmental sensor is included in the candidate abnormal sensor set when its path anomaly score reaches the 95th percentile of the corresponding score for the normal baseline data, and at least two of the three characteristics exceed their corresponding thresholds.

[0083] For candidate anomaly sensors of the same type, the output loop direction, the output difference direction at the path endpoint, and three normalized features are compared. When at least two environmental sensors have the same loop direction and difference direction, and their feature similarity reaches the 5th percentile of the similarity of common pollution samples in an empty-load pollution simulation experiment, they are grouped into a common-mode pollution sensor group. A common-mode pollution sensor group refers to a collection of similar environmental sensors that exhibit consistent non-exchange responses when affected by the same deposits, condensate films, or polluted environments.

[0084] If other types of trusted environmental sensors that are physically associated with the environmental variable also show path differences in direction and timing that conform to the physical process, then this phenomenon is marked as real environmental path memory and not considered common-mode contamination. Real environmental path memory refers to the phenomenon that different operating sequences produce differences in real environmental states due to heat storage of the enclosure material, surface moisture absorption, or airflow mixing lag. The candidate abnormal sensor set, the common-mode contamination sensor group, and the real environmental path memory identifier are output to S430.

[0085] S420: Identify the source of environmental change. Read the empirically verified causal response matrix generated in S230, and call the current response delay output in S330. Current response gain and control the degree of watermark retention It also reads the actual output of the environmental actuator recorded in S130 and the status of the hatch and operating ports recorded in S110.

[0086] Calculate the source matching degree for the current response from the j-th environmental actuator to the i-th environmental sensor:

[0087] in, Indicates the source matching degree. and These represent the verified response delay and response gain, respectively. The value is 1 if the current response direction matches the verified direction, and 0 otherwise. to The sum is 1. This indicates that positive numbers with a denominator of 0 should be avoided.

[0088] Each weight is determined by normal internal adjustment data; if insufficient data is available, an equal-weighted initial value is used. The source matching threshold is taken as the 5th percentile of the source matching degree of normal internal adjustment data. When the actual output feedback of the environmental actuator confirms that the control watermark is effectively loaded, the source matching degree of at least two trusted response channels reaches the threshold, and the direction of change conforms to the direction of physical action, the corresponding environmental change is determined as an internal environmental change. When multiple environmental actuators meet the conditions simultaneously, each internal source is recorded separately.

[0089] When no environmental actuator reaches the source matching threshold, the actual output does not change sufficiently to explain the current change, at least two trusted environmental sensors in different installation areas detect a change in direction consistent with the change, and the hatch or operating port is opened during the corresponding time period, the change is identified as an external environmental change, and the associated event is recorded. If both internal source evidence and external event evidence are insufficient, it is identified as an environmental change with an uncertain source.

[0090] If the control watermark disappears only in one environmental sensor, but other associated trusted environmental sensors can still identify the control watermark, it is given priority as evidence of environmental sensor anomaly and is not considered an external environmental change. An initial record of the source of the environmental change is generated, including the time of change, the affected environmental variable, the type of change, the environmental enforcement agency number or external event identifier, the source matching degree, and the trusted response channels involved in the judgment, and is output to S430.

[0091] S430: Complete cross-validation of environmental sensor reliability and sources of environmental change. Determine path reliability based on the path anomaly score from S410. The higher the path anomaly score, the lower the path reliability; when it belongs to the common-mode contamination sensor group, the path reliability is 0.

[0092] The highest valid source matching degree corresponding to the i-th environmental sensor is determined as the source credibility. For changes in the external environment, the source reliability is determined based on the proportion of reliable environmental sensors exhibiting the same direction of change and the temporal consistency between events at hatches or operating ports and environmental changes. A data quality score is calculated based on the proportion of missing data, anomalous jumps, and sampling anomalies recorded in S110. And limit it to between 0 and 1.

[0093] Calculate the overall credibility:

[0094] in, Indicates overall credibility. to The sum is 1. Each weight is determined by the no-load debugging data and the historical operation data that has been confirmed to be normal. When there is insufficient data, the initial value of equal weight is used.

[0095] The confidence threshold is set at the 5th percentile of the overall confidence level of normal environmental sensors, and the anomaly threshold is set at 80% of the confidence threshold. Environmental sensors not included in the candidate anomaly sensor set, whose overall confidence level reaches the confidence threshold, and which satisfy path endpoint consistency and control watermark causality consistency are included in the confident sensor set. Environmental sensors belonging to the common-mode pollution sensor group, or whose overall confidence level is below the anomaly threshold for two consecutive identification cycles, are included in the anomaly sensor set. The remaining environmental sensors are included in the sensor set to be verified. Sensors to be verified do not participate in data fusion in the current control cycle and are re-verified in the next identification cycle.

[0096] The initial record of environmental change sources generated in S420 is compared with the set of trusted sensors. Source conclusions supported only by anomalous sensors are eliminated, and source credibility, real environment path memory identifiers, and judgment validity periods are added to form a record of trusted environmental change sources. The set of trusted sensors, the set of anomalous sensors, the set of sensors to be verified, and the record of trusted environmental change sources are output to S510.

[0097] For example, when two humidity sensors are identified as a common-mode contamination sensor group, they are still included in the abnormal sensor set even if their current measurement values ​​are within the allowable range; humidity sensors that do not form abnormal hysteresis loops but have insufficient source evidence are included in the sensor set to be reviewed.

[0098] S5 specifically includes the following sub-steps: S510: Reconstruct the credible environmental state and separate environmental changes. Based on the environmental sensor installation and configuration, environmental control area division, and steady-state measurement variance formed in S110, call the empirically verified causal response matrix formed in S230, and obtain the credible sensor set, abnormal sensor set, sensor set to be verified, and credible environmental change source records from S430.

[0099] The reliable environmental state refers to the environmental state reconstructed from the current measurements in the reliable sensor set, which includes temperature, humidity, airflow and radiation conditions in each environmental control area. Measurements from abnormal sensors and sensors to be verified are not included in the data fusion of the current control cycle.

[0100] The confidence state of the q-th type of environmental variable within the r-th environmental control area is:

[0101] in, Indicates a trusted state. This represents the set of reliable sensors that measure the q-th type of environmental variable within the corresponding region. Indicates overall credibility. This represents the variance of the steady-state measurement obtained from S110. Indicates the current measurement value. This indicates that positive numbers with a denominator of 0 should be avoided.

[0102] Environmental sensors with high overall reliability and low stable-state measurement variance are given higher fusion weights. When the number of reliable sensors is lower than the preset minimum number of reliable sensors determined by S110 based on the installation configuration, a reliable sensor shortage flag is set, the reliable state corresponding to the previous control cycle is used as the transient estimate, the state uncertainty is increased, and the maximum adjustment range of the relevant environmental actuators is limited to 50% of the normal upper limit.

[0103] Based on the trusted environmental change source records, the internal environmental change is calculated only for environmental actuators identified as sources of internal environmental change, using their actual output change and verified response gain and response delay:

[0104] in, Indicates the first The internal environmental changes at each control sampling point This represents the set of environmental enforcement agencies that have reliable response channels to this region and environmental variables. This represents the change in the actual output quantity relative to the reference output quantity. and These represent the overall response gain and overall response delay obtained by summing the weights of the matrix units, respectively.

[0105] The actual change in the credible state is subtracted from the change in the internal environment; the remaining portion is recorded as an external environmental disturbance if there is evidence of an event involving a hatch or control port, and as a disturbance of uncertain origin if there is no evidence of an external event. For environmental changes with a real environmental path memory identifier, the lagged portion of the change is included in the internal environmental change and is not compensated for as an environmental sensor anomaly.

[0106] Based on the weighted dispersion of reliable environmental sensors, the degree of reliability of sensors, the residual between actual changes and causal response predictions, and the degree of source reliability deficiency, the state uncertainty of each environmental control area and environmental variable is calculated. The reliable environmental state, internal environmental changes, external environmental disturbances, source uncertain disturbances, state uncertainty, and reliable sensor reliability deficiency indicators are generated and output to S520.

[0107] S520. Generate optimized adjustment instructions through strategy optimization. Strategy optimization refers to the process of determining the target output of the heater, humidifier, circulating fan, and radiation control unit in the next prediction time domain under the constraints of the safe range of environmental variables, the output range of environmental actuators, the rate of output change, and state uncertainty.

[0108] The credible environmental state formed by S510 is used as the initial value for prediction, and the equipment setpoint is used as the target environmental state. The internal environmental changes generated by the candidate control combination are predicted based on the verified causal response matrix. The external environmental disturbance is extrapolated in a short time according to the change slope of the most recent 3 control cycles, and the source of the uncertain disturbance is incorporated into the state uncertainty safety margin without deterministic compensation.

[0109] The position vector of each artificial bat sequentially encodes the target power of the heater, the target atomization duty cycle or water supply of the humidifier, the target speed of the circulating fan, and the target power of the radiation regulation unit; the prediction time domain is set to 3 to 10 control steps, with each control step being 5 to 30 seconds.

[0110] The objective function of the candidate adjustment combination is:

[0111] Where F represents the objective function, h represents the predictive control step index, and H represents the predictive control step number. This represents the normalized deviation between the predicted environmental state and the target environmental state. This indicates the loss caused by unsuppressed external environmental disturbances. Indicates energy loss. This represents the output change and switching loss. This represents the loss due to state uncertainty. to The sum is 1. The deviation of each environmental variable is divided by the corresponding allowable range width before being included in the calculation. Each weight is determined by no-load commissioning data and reliable historical control data. When there is insufficient data, equal-weighted initial values ​​are used.

[0112] The bat population size is set to 30. The frequency, velocity, and position of the b-th artificial bat in the m-th iteration are updated according to the following formula:

[0113]

[0114]

[0115] Where b represents the artificial bat ID, and m represents the iteration number. Represents a random number between 0 and 1. Indicates search frequency. and These represent the velocity vector and the position vector, respectively. This represents the current optimal position; and local candidate solutions are generated near the current optimal position based on the current loudness, with a search frequency lower bound. and upper limit Example of value range: =0、 =2.

[0116] Candidate solutions that violate environmental safety limits, rated output limits, permissible output change rates, or control watermark safety boundaries are first subjected to boundary repair. If the constraints are still not met after repair, the solution is deemed infeasible. The control contribution of environmental actuators without reliable response channels to the corresponding environmental variables is set to 0. When the state uncertainty exceeds the corresponding threshold, the candidate adjustment range must not exceed 50% of the normal maximum adjustment range.

[0117] The maximum number of iterations is set to 100. The iteration terminates when the change in the optimal objective function is less than 0.001 after 10 consecutive iterations. The candidate adjustment combination with the smallest objective function that satisfies all constraints is selected as the optimized adjustment command. For example, if the opening of the door causes a temperature drop and only the heater and the circulating fan have reliable temperature response channels, the method of increasing the humidifier output is not used to compensate for the temperature change.

[0118] S530: Execute the optimization adjustment command and update the control closed loop. Convert the position vector output by S520 into the target power of the heater, the target atomization duty cycle or water supply of the humidifier, the target speed of the circulating fan, and the target power of the radiation regulation unit, and execute them step by step according to the allowable output change rate of the equipment; when the control watermark continues to be applied, the output after the optimization adjustment command and the control watermark are superimposed must not exceed the safety boundary determined by S120.

[0119] The response verification window is set according to the median response delay of the corresponding trusted response channel in S230. The actual environmental response is compared with the prediction results of the verified causal response matrix. When the absolute value of the prediction residual does not exceed the 95th percentile of the normal historical adjustment residual, the adjustment response is determined to be matched.

[0120] When the actual output feedback deviates from the optimized adjustment command by more than 2% of the rated output, the actual response direction is opposite to the verified direction, the response delay exceeds the maximum allowable response time, the responses of two consecutive verification windows do not match, or the state uncertainty continues to rise, the output of the relevant environmental actuators will be stopped, the most recent safe output will be maintained, the corresponding response channel will be marked as a response channel to be reviewed, and S230 will be triggered for re-verification; when multiple environmental variables are close to the allowable boundary, the strategy optimization control will be exited and the device will switch to the preset safety control mode.

[0121] When all trusted environmental states fall within the allowable range corresponding to the target environmental state, the change amplitude over 60 consecutive sampling cycles is lower than the 95th percentile of the S110 steady-state data, the state uncertainty is lower than the corresponding threshold, and there is no response mismatch indicator, the current adjustment cycle ends. Repeating S510 to S530 constitutes the control cycle; re-executing S110 to S430 constitutes the identification cycle.

[0122] If the causal response matrix is ​​within its validity period, the set of trusted sensors has not changed, and there is no response mismatch, the next control cycle returns to S510; if the preset identification interval is reached, the overall trustworthiness decreases, or the response channel drifts, the cycle returns to S110.

[0123] The adjusted credible environmental state serves as the prior state for the next control cycle S510. The environmental actuator operation record serves as the data source for calculating the internal environmental changes in the next cycle and for the next identification cycle S130. The adjusted response residual is used to update the response matching threshold. The credible state of the environmental sensor serves as the prior classification for the credibility update in the next identification cycle S430.

[0124] For example, if the expected temperature increase of 0.05℃ is achieved after the heater output is increased by 4W, and the actual temperature increase is 0.047℃ with the residual below the 0.015℃ threshold, then the regulation response matching is determined.

[0125] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.

[0126] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0128] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for regulating the environment of a neonatal incubator based on multi-sensor fusion, characterized in that, Includes the following steps: S1. Collect temperature, humidity, airflow, radiation, and actual output of environmental actuators inside the incubator to establish the initial environmental state; superimpose control watermarks within the stable state and safety boundary to form watermark response data; S2. Extract the response direction, response delay, response gain and recovery time from the watermark response data, construct the initial causal response matrix, and repeatedly verify it to form a verified causal response matrix and a set of trusted response channels; S3. Based on the verified causal response matrix, generate non-exchangeable environment paths that start from a common initial environmental state, call the same environmental actuators, have the same cumulative adjustment amount, and run in different orders. After execution, a non-exchangeable response feature set is formed. S4. Identify candidate abnormal sensors and common-mode contamination sensor groups based on the non-exchange response feature set. Combine control watermarks and causal response relationships to distinguish between internal environmental changes, external environmental changes, and environmental changes with uncertain sources, forming a set of credible sensors and a record of credible sources of environmental changes.

2. The method for regulating the neonatal incubator environment based on multi-sensor fusion according to claim 1, characterized in that, Also includes: S5. Reconstruct the state of the trusted environment using a set of trusted sensors, separate the internal environmental changes and external environmental disturbances based on the source of the trusted environmental changes, optimize the strategy, generate and execute optimized adjustment commands, and update the control closed loop based on the actual environmental response.

3. The method for regulating the neonatal incubator environment based on multi-sensor fusion according to claim 1, characterized in that, S1 specifically includes: Collect temperature, humidity, airflow, radiation, and actual output of environmental actuators in the incubator; synchronize the collected data with time and mark anomalies; and determine the initial environmental state and baseline output under stable conditions. The control watermark security boundary is determined based on the initial environmental state, preset environmental range and response slope. The bat algorithm is used to optimize the watermark amplitude, period, phase offset, duration and loading order to obtain the control watermark parameter set. Within the stable state and the control watermark safety boundary, the control watermark is superimposed on the reference adjustment command, and the environmental sensor measurement values, watermark loading commands, and actual output of the environmental actuator are recorded to form watermark response data.

4. The method for regulating the neonatal incubator environment based on multi-sensor fusion according to claim 1, characterized in that, S2 specifically includes: Benchmark correction and cross-correlation analysis are performed on the watermark response data to extract the response direction, response delay, response gain and recovery time, forming a watermark response feature set; Using environmental actuators as input and environmental sensors as output, watermark response features are written into the response channel, and effective response channels are selected based on relevant thresholds, minimum identifiable gain, physical action direction, and maximum allowable response time to construct an initial causal response matrix; By comparing response direction, response delay, response gain, and recovery time through multiple repeated validation windows, causal confidence is calculated, and reliable response channels are selected.

5. The method for regulating the neonatal incubator environment based on multi-sensor fusion according to claim 4, characterized in that, Also includes: The median of the response features and the weights of the matrix units are written into the initial causal response matrix to update the verified causal response matrix and form a set of reliable response channels.

6. The method for regulating the neonatal incubator environment based on multi-sensor fusion according to claim 1, characterized in that, S3 specifically includes: Based on the verified causal response matrix and the set of trusted response channels, the coupled environment actuators are screened, the common initial environment state and the common target environment state are determined, and non-exchange environment paths with the same cumulative adjustment amount but different running order are generated. The system operates according to the non-exchange environment path control environment actuator, superimposes control watermarks on the path adjustment amount, collects environmental sensor measurement values, actual output quantities and path stage information, forms a path response sequence and executes path reset; The path response sequence is processed by delay compensation, stage matching and normalization, and the output loop area, path endpoint output difference, recovery time difference, current response delay, current response gain and control watermark retention degree are calculated to form a non-exchangeable response feature set.

7. The method for regulating the neonatal incubator environment based on multi-sensor fusion according to claim 1, characterized in that, S4 specifically includes: A normal baseline distribution is established based on the non-exchange response feature set, path anomaly scores are calculated, candidate anomaly sensors are screened, and common-mode contamination sensor groups are identified based on the output loop direction, the output difference direction at the path endpoint, and feature similarity. By combining the validated causal response matrix, current response delay, current response gain, control watermark retention level, and actual output of the environmental actuator, the source matching degree is calculated to distinguish between internal environmental changes, external environmental changes, and environmental changes with uncertain sources.

8. The method for regulating the neonatal incubator environment based on multi-sensor fusion according to claim 7, characterized in that, Also includes: Based on path credibility, source credibility, and data quality score, a comprehensive credibility score is calculated. Environmental sensors are classified as credible, abnormal, and pending verification, forming a set of credible sensors, a set of abnormal sensors, a set of sensors pending verification, and a record of credible environmental change sources.

9. The method for regulating the environment of a neonatal incubator based on multi-sensor fusion according to claim 2, characterized in that, S5 specifically includes: Based on a set of trusted sensors, a causal response matrix, and records of trusted sources of environmental change, the measured values ​​of the environmental control area are weighted and fused to reconstruct the trusted environmental state, and the internal environmental change, external environmental disturbance, and uncertain source disturbance are separated. Using the credible environmental state as the initial prediction value, the target output of the environmental actuator is encoded as a candidate adjustment combination. The Bat algorithm is used to optimize the strategy under the constraints of environmental safety range, output range, rate of change and state uncertainty, and generate optimized adjustment instructions.

10. The method for regulating the environment of a neonatal incubator based on multi-sensor fusion according to claim 9, characterized in that, Also includes: The system executes optimization and adjustment commands, verifies and adjusts the response based on the deviation between the environmental response and the predicted results of the causal response matrix, triggers response channel verification when the response does not match, and updates the control closed loop based on the credible environmental state.

Images