Anti-interference control method and system for intelligent incubation warehouse based on error self-learning

Abstract

The present invention provides an intelligent incubator anti-interference control method and system based on error self-learning, which relates to the field of intelligent control technology. The method includes: calculating temperature deviation values ​​and humidity deviation values ​​based on optimized parameter sequences of temperature prediction error and humidity prediction error, respectively; performing normalized weighted summation on the temperature deviation values ​​and humidity deviation values, and dynamically scaling them using a Gaussian kernel function to obtain a correction factor; dynamically adjusting the activation function slope of the hidden layer of the neural network based on the correction factor and the BP neural network, and reconstructing the weights of the fuzzy rule base based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weights; constructing a three-dimensional parameter adjustment surface based on the corrected fuzzy rule base weights, and dynamically adjusting the proportional, integral, and differential parameters of the PID controller using a surface gradient search algorithm to generate a control signal. The present invention improves control accuracy.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to an intelligent incubation warehouse anti-interference control method and system based on error self-learning. Background Art

[0002] The current control technology for intelligent incubators mainly relies on traditional PID control algorithms, but it has some defects in complex interference scenarios, as shown below:

[0003] Traditional PID control relies on fixed parameter adjustments and is difficult to adapt to environmental disturbances and time-varying characteristics during the incubation process (such as sudden changes in temperature and humidity caused by ventilation, and changes in heat transfer efficiency due to equipment aging). Its core flaws include:

[0004] The ability to resist time-varying interference is insufficient, and it is unable to dynamically track the non-steady-state changes of temperature and humidity errors. When periodic interference (such as daily ambient temperature fluctuations) or sudden interference (such as equipment failure) occurs in the incubation chamber, the control quantity is prone to overshoot or adjustment lag, causing the incubation environment to deviate from the ideal range for a long time.

[0005] Temperature and humidity parameters are strongly coupled (for example, humidification is accompanied by heat loss). Traditional single-loop PID control adjusts them independently, ignoring any cross-effects and resulting in a control strategy that prioritizes one over the other. For example, humidity adjustment can cause temperature fluctuations, but PID controllers are unable to quantify and compensate for these coupled effects. Proportional, integral, and differential parameters require manual tuning, making it difficult to address the dynamic control requirements of different incubation stages (e.g., the differing sensitivity to temperature and humidity during early hatching and hatching), and lacking an adaptive adjustment mechanism. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an intelligent incubation warehouse anti-interference control method based on error self-learning, thereby improving the accuracy of control.

[0007] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0008] In a first aspect, an anti-interference control method for an intelligent incubation chamber based on error self-learning is provided, the method comprising:

[0009] Step 1: Real-time acquisition of temperature and humidity data streams for incubating chicken embryos in an incubation chamber; forming a sliding detection window with data from three consecutive sampling cycles, fusing the temperature and humidity data within the window, and outputting an optimized parameter sequence for temperature prediction error and an optimized parameter sequence for humidity prediction error;

[0010] Step 2: Calculate the temperature deviation value and the humidity deviation value according to the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error respectively;

[0011] Step 3: Perform normalized weighted summation of the temperature deviation and humidity deviation, and dynamically scale them using a Gaussian kernel function to obtain a correction factor;

[0012] Step 4: dynamically adjust the activation function slope of the hidden layer of the neural network according to the correction factor and the BP neural network, and reconstruct the weight of the fuzzy rule base based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight;

[0013] In step 5, based on the modified fuzzy rule base weights, a three-dimensional parameter adjustment surface is constructed, and the proportional, integral, and differential parameters of the PID controller are dynamically adjusted through the surface gradient search algorithm to generate a control signal.

[0014] Furthermore, the data of three consecutive sampling periods are used to form a sliding detection window, and the temperature and humidity data are fused within the window to output the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error, including:

[0015] In the sliding detection window, the temperature and humidity raw data at the current sampling moment and the two moments before it are constructed into time series vectors, each vector containing three consecutive data points;

[0016] A state transition model is established based on the temperature observations of the previous two cycles to generate the temperature prediction value at the current moment. The actual temperature data collected is subtracted from the prediction value to obtain the temperature prediction error value. The filter gain is dynamically adjusted according to the noise covariance matrix and the observation noise covariance matrix, and the prediction error is weighted and corrected to output the temperature prediction error optimization parameter at the current moment.

[0017] Repeat the prediction-correction process for the humidity time series vector to generate the corresponding humidity prediction error optimization parameters;

[0018] Arrange the temperature prediction error optimization parameters output from three consecutive sampling periods in chronological order to form an optimization parameter sequence for the temperature prediction error; arrange the humidity prediction error optimization parameters to form an optimization parameter sequence for the humidity prediction error;

[0019] The sliding average of three consecutive parameter values ​​in the optimized parameter sequence of the temperature prediction error is performed to obtain the temperature error mean; the square deviation of each parameter value is calculated based on the temperature error mean, and the arithmetic average of the three square deviations is used as the current element of the temperature error variance matrix; the same operation is performed on the optimized parameter sequence of the humidity prediction error to generate the current element of the humidity error variance matrix.

[0020] Furthermore, based on the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error, the temperature deviation value and the humidity deviation value are calculated respectively, including:

[0021] The optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error are aligned point by point according to the sampling timestamp. From the aligned optimized parameter sequence of the temperature prediction error, the rate of change of the data before and after the current sampling point is calculated to obtain the dynamic gradient eigenvalue reflecting the dynamic change trend of the temperature error sequence; from the optimized parameter sequence of the humidity prediction error, the absolute value of the difference between adjacent data is calculated to obtain the quantitative fluctuation eigenvalue representing the instantaneous fluctuation degree of the humidity error;

[0022] An initial weight is generated based on the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue, and the initial weight is corrected using the historical humidity deviation ratio to obtain a normalized weight coefficient. The difference between the current element of the temperature error variance matrix and the target threshold is multiplied by the normalized weight coefficient to obtain a basic deviation. The basic deviation is processed based on the instantaneous intensity value in the quantized fluctuation eigenvalue to obtain a temperature deviation value.

[0023] The ratio of the current element of the humidity error variance matrix to the target threshold is transformed into a natural logarithm to generate an initial deviation. The compensation factor is generated according to the product of the accumulated temperature amplitude and the instantaneous humidity fluctuation intensity. The initial deviation is dynamically compressed using the compensation factor to obtain the humidity deviation value.

[0024] Furthermore, an initial weight is generated based on the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue, and the initial weight is corrected using the historical humidity deviation ratio to obtain a normalized weight coefficient. The difference between the current element of the temperature error variance matrix and the target threshold is multiplied by the normalized weight coefficient to obtain a basic deviation. The basic deviation is processed based on the instantaneous intensity value in the quantized fluctuation eigenvalue to obtain a temperature deviation value, including:

[0025] Calculate the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue to obtain a ratio that reflects the relative intensity of the temperature and humidity error changes, which is used as the initial weight.

[0026] Calculate the ratio of humidity error historical data that deviates from the normal range, that is, the humidity historical deviation ratio, and use the humidity historical deviation ratio to adjust the initial weight to obtain the corrected initial weight;

[0027] The corrected initial weight is normalized to obtain a normalized weight coefficient;

[0028] Obtain the current element of the temperature error variance matrix, that is, the average value of the squared deviation of the temperature error in the current sliding window, calculate the difference between the current element and the preset temperature error target threshold, and fuse the difference with the normalized weight coefficient to obtain the basic deviation value that reflects the degree of deviation of the temperature error fluctuation from the target state;

[0029] The instantaneous intensity value at the current moment, that is, the fluctuation peak value of the current humidity error, is extracted from the quantized fluctuation characteristic value. The basic deviation is adjusted using the instantaneous intensity value to obtain the temperature deviation value reflecting the temperature error and humidity interference.

[0030] Furthermore, the ratio of the current element of the humidity error variance matrix to the target threshold is transformed into a natural logarithm to generate an initial deviation. A compensation factor is generated based on the product of the temperature cumulative amplitude value and the instantaneous humidity fluctuation intensity. The initial deviation is dynamically compressed using the compensation factor to obtain a humidity deviation value, including:

[0031] Get the current element of the humidity error variance matrix, that is, the average value of the square of the humidity error deviation in the current sliding window, and calculate the humidity ratio of the current element to the preset humidity error target threshold to reflect the deviation multiple of the current humidity error fluctuation relative to the target state;

[0032] The natural logarithm of the humidity ratio is taken and the numerical range is compressed to the logarithmic scale to obtain the initial deviation reflecting the degree of humidity error deviation;

[0033] Calculate the absolute value cumulative sum of the optimized parameter sequence of the humidity prediction error within the current sliding window to characterize the overall deviation degree and duration of the temperature error and form the temperature cumulative amplitude value;

[0034] Extract the instantaneous fluctuation peak value at the current moment from the optimized parameter sequence of humidity prediction error, that is, the maximum absolute value of the difference between adjacent sampling points, and record it as the humidity instantaneous fluctuation intensity;

[0035] Multiply the temperature cumulative amplitude value by the instantaneous humidity fluctuation intensity to obtain the compensation factor;

[0036] The initial deviation is adjusted by the compensation factor, and the numerical range of the initial deviation is narrowed by nonlinear mapping to obtain the humidity deviation value that balances the coupling effect of temperature and humidity.

[0037] Furthermore, the temperature deviation and humidity deviation are normalized and weighted and summed, and dynamically scaled using a Gaussian kernel function to obtain a correction factor, including:

[0038] Calculating the maximum and minimum values ​​of the temperature deviation value in a preset sampling period respectively, and calculating the first numerical deviation range thereof, that is, the maximum value minus the minimum value; subtracting the minimum value from the current temperature deviation value to obtain a first deviation difference, and calculating the deviation ratio between the first deviation difference and the first numerical deviation range to obtain a normalized temperature deviation value;

[0039] Calculating the maximum and minimum values ​​of the humidity deviation value in a preset sampling period respectively, and calculating its second numerical deviation range, that is, the maximum value minus the minimum value; subtracting the minimum value from the current humidity deviation value to obtain a second deviation difference, and calculating the deviation ratio between the second deviation difference and the second numerical deviation range to obtain a normalized humidity deviation value;

[0040] Preset the temperature influence weight coefficient and humidity influence weight coefficient, multiply the normalized temperature deviation value by the temperature influence weight coefficient to obtain the temperature contribution; multiply the normalized humidity deviation value by the humidity influence weight coefficient to obtain the humidity contribution;

[0041] Add the temperature contribution and the humidity contribution to obtain the temperature and humidity deviation index value; substitute the temperature and humidity deviation index value into the Gaussian kernel function and calculate its function value, which is the correction factor.

[0042] Furthermore, according to the correction factor and the BP neural network, the activation function slope of the hidden layer of the neural network is dynamically adjusted, and the weight of the fuzzy rule base is reconstructed based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight, including:

[0043] Determine the initial slope parameter corresponding to the activation function of the hidden layer of the BP neural network;

[0044] Determine the adjustment amount of the slope of the activation function according to the value of the correction factor to obtain the adjusted value of the slope;

[0045] Add the adjusted slope value to the initial slope parameter to obtain the adjusted activation function slope;

[0046] Analyze the numerical distribution characteristics of the correction factor and calculate the numerical distribution characteristics, which include the mean, variance, maximum value and minimum value of the correction factor;

[0047] The initial weight of the fuzzy rule base is determined, and the weight of each rule in the fuzzy rule base is adjusted according to the numerical distribution characteristics of the correction factor to obtain the corrected weight of the fuzzy rule base.

[0048] Furthermore, based on the weights of the modified fuzzy rule base, a three-dimensional parameter adjustment surface is constructed. The proportional, integral, and differential parameters of the PID controller are dynamically adjusted through the surface gradient search algorithm to generate a control signal, including:

[0049] Determine the three variables of the three-dimensional parameter adjustment surface, which are temperature deviation value, humidity deviation value and the weight of the modified fuzzy rule base;

[0050] Under different combinations of temperature deviation values, humidity deviation values ​​and modified fuzzy rule base weights, the corresponding proportional, integral and differential parameter values ​​of the PID controller are recorded;

[0051] According to the values ​​of the proportional, integral and differential parameters of the PID controller, a three-dimensional parameter adjustment surface is constructed through neural network fitting;

[0052] According to the current temperature deviation value, humidity deviation value and the weight of the modified fuzzy rule base in the incubator, the corresponding state point is found on the three-dimensional parameter adjustment surface. The state point represents the initial parameter setting of the PID controller under the current working condition.

[0053] At the current state point, the gradient of the three-dimensional parameter adjustment surface is calculated. The gradient represents the direction of change of the three-dimensional parameter adjustment surface at the state point. It contains three components, which correspond to the change directions of the proportional, integral and differential parameters of the PID controller respectively.

[0054] The gradient and search step of the surface are adjusted according to the three-dimensional parameters, and the proportional, integral and differential parameters of the PID controller are adjusted in the opposite direction of the gradient until the search step is reached to obtain the optimized proportional, integral and differential parameters of the PID controller;

[0055] The PID controller calculates the corresponding control signal based on the current temperature deviation value and humidity deviation value, combined with the adjusted proportional, integral and differential parameters.

[0056] In the second aspect, an intelligent incubation chamber anti-interference control system based on error self-learning includes:

[0057] The acquisition module is used to collect the temperature and humidity data stream of the chicken embryo incubation in the incubation chamber in real time; the data of three consecutive sampling cycles are used to form a sliding detection window, and the temperature and humidity data are fused within the window to output the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error;

[0058] A calculation module, used to calculate the temperature deviation value and the humidity deviation value respectively according to the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error;

[0059] A correction module is used to perform normalized weighted summation of the temperature deviation value and the humidity deviation value, and dynamically scale them using a Gaussian kernel function to obtain a correction factor;

[0060] An adjustment module is used to dynamically adjust the activation function slope of the hidden layer of the neural network according to the correction factor and the BP neural network, and reconstruct the weight of the fuzzy rule base based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight;

[0061] The processing module is used to construct a three-dimensional parameter adjustment surface based on the modified fuzzy rule base weights, and dynamically adjust the proportional, integral and differential parameters of the PID controller through the surface gradient search algorithm to generate a control signal.

[0062] The above solution of the present invention includes at least the following beneficial effects:

[0063] By forming a sliding detection window with data from three consecutive sampling cycles, the temporal characteristics of the temperature and humidity data streams can be captured, random noise interference can be effectively filtered out, the reliability of the original data can be improved, the temperature and humidity data can be fused within the window, and an optimized parameter sequence of the temperature / humidity prediction error can be output. Compared with the traditional single-cycle sampling method, the prediction error is reduced.

[0064] Based on the optimized parameter sequence, the temperature and humidity deviation values ​​are calculated in real time, and the degree of deviation between the environmental parameters and the set threshold is accurately quantified, providing a clear quantitative basis for interference response; the dimensional differences of temperature and humidity are eliminated through normalization processing, and the correction factor is dynamically adjusted based on the nonlinear mapping characteristics of the Gaussian kernel function, which improves the system's response speed to sudden interference (such as equipment start-up and shutdown, environmental fluctuations), and significantly enhances the interference suppression capability.

[0065] The slope of the activation function of the hidden layer of the neural network is dynamically adjusted by the correction factor, the fitting ability of the BP network to complex nonlinear error characteristics is optimized, the model convergence speed is improved, and the weights of the fuzzy rule base are reconstructed based on the numerical distribution characteristics of the correction factor, so that the fuzzy control rules can be adaptively adjusted according to the error characteristics, solving the limitations of fixed traditional fuzzy control rules.

[0066] A three-dimensional adjustment surface for the PID controller parameters was constructed, and the proportional, integral, and differential parameters were optimized in real time using the surface gradient search algorithm. Compared with traditional PID control, the temperature control fluctuation range was reduced to ±0.2°C, and the humidity control fluctuation range was reduced to ±3%RH. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart of an anti-interference control method for an intelligent incubation chamber based on error self-learning provided by an embodiment of the present invention.

[0068] Figure 2 This is a schematic diagram of an intelligent incubation warehouse anti-interference control system based on error self-learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0070] like Figure 1As shown, an embodiment of the present invention proposes an intelligent incubation warehouse anti-interference control method based on error self-learning, and the method includes the following steps:

[0071] Step 1: Real-time acquisition of temperature and humidity data streams for incubating chicken embryos in an incubation chamber; forming a sliding detection window with data from three consecutive sampling cycles, fusing the temperature and humidity data within the window, and outputting an optimized parameter sequence for temperature prediction error and an optimized parameter sequence for humidity prediction error;

[0072] Step 2: Calculate the temperature deviation value and the humidity deviation value according to the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error respectively;

[0073] Step 3: Perform normalized weighted summation of the temperature deviation and humidity deviation, and dynamically scale them using a Gaussian kernel function to obtain a correction factor;

[0074] Step 4: dynamically adjust the activation function slope of the hidden layer of the neural network according to the correction factor and the BP neural network, and reconstruct the weight of the fuzzy rule base based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight;

[0075] In step 5, based on the modified fuzzy rule base weights, a three-dimensional parameter adjustment surface is constructed, and the proportional, integral, and differential parameters of the PID controller are dynamically adjusted through the surface gradient search algorithm to generate a control signal.

[0076] In this embodiment of the present invention, a sliding detection window is constructed by combining data from three consecutive sampling periods to capture the temporal characteristics of the temperature and humidity data streams, effectively filtering out random noise interference and improving the reliability of the original data. Temperature and humidity data are fused within the window to output an optimized parameter sequence for temperature / humidity prediction errors. This reduces prediction errors compared to traditional single-period sampling methods. Based on the optimized parameter sequence, temperature and humidity deviations are calculated in real time to accurately quantify the degree of deviation between environmental parameters and set thresholds, providing a clear quantitative basis for interference response. Normalization eliminates dimensional differences in temperature and humidity, and dynamically adjusts correction factors based on the nonlinear mapping properties of the Gaussian kernel function. This improves the system's response to sudden interference (such as equipment startup and shutdown, environmental fluctuations), significantly enhancing interference suppression capabilities. Dynamically adjusting the slope of the neural network's hidden layer activation function using the correction factor optimizes the BP network's ability to fit complex nonlinear error characteristics, improving model convergence speed. Reconstructing the weights of the fuzzy rule base based on the numerical distribution characteristics of the correction factor enables fuzzy control rules to adapt to error characteristics, addressing the limitations of traditional fixed fuzzy control rules.

[0077] A three-dimensional adjustment surface for PID controller parameters is constructed, and a surface gradient search algorithm is used to optimize proportional, integral, and differential parameters in real time. Compared with traditional PID control, the temperature control fluctuation range is reduced to ±0.2°C, and the humidity control fluctuation range is reduced to ±3%RH. A complete closed loop is formed from data acquisition, error prediction, correction factor generation, and control parameter optimization. It can gradually improve the ability to predict complex interference scenarios through iterative learning through historical error data. By integrating the advantages of multiple technologies such as data fusion algorithms, neural networks, fuzzy control, and gradient search, it has both data-driven learning capabilities and the interpretability of rule-based reasoning, allowing the incubator to maintain optimal control at different incubation stages (such as early embryonic development and hatching), thereby improving the hatching success rate.

[0078] In another preferred embodiment of the present invention, data from three consecutive sampling periods are used to form a sliding detection window, and the temperature and humidity data are fused within the window to output an optimized parameter sequence for the temperature prediction error and an optimized parameter sequence for the humidity prediction error, including:

[0079] Within the sliding detection window, the temperature and humidity raw data at the current sampling moment and the two moments before it are constructed as time series vectors. Each vector contains three consecutive data points. Specifically, the discrete sampling data is converted into vectors with time series association to capture short-term dynamic characteristics. The implementation includes:

[0080] At the current sampling time (denoted as t), extract the data of the last three consecutive sampling periods (i.e., time t-2, t-1, and t), and construct time series vectors of temperature and humidity respectively. Each vector contains three consecutive data points. For example, the temperature vector is , the humidity vector is ,In this way, the real-time collected data stream is converted into a 3D data unit with ,temporal context.

[0081] A state transition model is established based on the temperature observations of the previous two cycles to generate the temperature prediction value at the current moment. The actual temperature data collected is subtracted from the prediction value to obtain the temperature prediction error value. The filter gain is dynamically adjusted according to the noise covariance matrix and the observation noise covariance matrix, and the prediction error is weighted and corrected. The optimized parameters of the temperature prediction error at the current moment are output, which specifically include:

[0082] Using the temperature observations from the first two periods Establish a state transition model (such as linear extrapolation or first-order dynamic model) to generate the temperature prediction value at the current moment , the model captures the temperature change trend based on the time series correlation of historical data; the actual collected temperature value Tt is compared with the predicted value Subtract and get the original temperature prediction error value According to the preset noise covariance matrix (reflecting the internal noise level of the system) and the observed noise covariance matrix (reflecting the sensor measurement error), the filter gain coefficient is dynamically adjusted. This coefficient is used to perform weighted correction on the original error value, suppress high-frequency noise, and output the optimized temperature prediction error parameter ,This process improves the stability of error characteristics by balancing the prediction error and noise intensity.

[0083] When it is applied in practice, the first-order dynamic model is adopted, and the temperature state vector is set as Contains current temperature and temperature change rate ,Right now The state transition equation can be expressed as: ;

[0084] Among them, the state transfer matrix , is the sampling period; is the process noise vector, which obeys a Gaussian distribution with a mean of zero, that is, is the process noise covariance matrix, which reflects the internal noise level of the system.

[0085] From the temperature observations of the first two cycles and To estimate the current state vector , temperature change rate ,but Then, the state prediction value at the current moment is for:

[0086] ;

[0087] Therefore, the temperature forecast value at the current moment is for: .

[0088] Repeat the prediction-correction process for the humidity time series vector to generate the corresponding humidity prediction error optimization parameters, including: Repeat the above temperature prediction and correction process: Based on the humidity observations at the first two moments ( and ) Establish a state transition model to predict the current humidity value ; Calculate the actual humidity value Error from the predicted value ; Use the same filter gain mechanism to perform weighted correction on humidity error and generate humidity prediction error optimization parameters .

[0089] The temperature prediction error optimization parameters output from three consecutive sampling periods are arranged in chronological order to form an optimization parameter sequence for the temperature prediction error; the humidity prediction error optimization parameters are arranged to form an optimization parameter sequence for the humidity prediction error, specifically including: arranging the three consecutive sampling periods (such as ) Output temperature prediction error optimization parameter Arranged in chronological order, forming the optimized parameter sequence of temperature prediction error ; Humidity sequence construction, similarly, the humidity optimization parameters of the three cycles Arranged as the optimized parameter sequence of humidity prediction error .

[0090] Perform sliding average processing on three consecutive parameter values ​​in the optimized parameter sequence of temperature prediction error to obtain the temperature error mean; calculate the square deviation of each parameter value based on the temperature error mean, and take the arithmetic average of the three square deviations as the current element of the temperature error variance matrix; perform the same operation on the optimized parameter sequence of humidity prediction error to generate the current element of the humidity error variance matrix, specifically including:

[0091] The temperature error mean is obtained by performing arithmetic averaging on the three consecutive parameter values ​​in the optimized parameter sequence of the temperature prediction error. , the formula is , reflecting the overall level of error.

[0092] The square deviation calculation is to calculate the square deviation of each parameter value from the mean, that is, , which measures the degree of dispersion of each data point from the mean.

[0093] Variance calculation: Take the arithmetic mean of the three squared deviations to get the current element of the temperature error variance matrix , the formula is , characterizes the fluctuation intensity of the error;

[0094] Humidity error processing: perform the same sliding average, deviation square calculation and variance calculation process on the optimized parameter sequence of humidity prediction error to obtain the current element of the humidity error variance matrix .

[0095] In this embodiment of the present invention, by retaining data from three consecutive cycles and constructing a minimized time series analysis unit, the temporal characteristics of temperature and humidity changes (such as rising / falling trends and fluctuation amplitudes) are effectively extracted, avoiding random interference from single-cycle data. A fixed window length (three cycles) balances data timeliness with computational complexity, avoiding the lag caused by long windows while reducing data storage and processing pressure. The state transition model infers the current theoretical value from historical data, pre-emptively capturing temperature and humidity trends (such as inertial delay during heating / cooling processes). This allows the system to pre-adjust control parameters before interference occurs, achieving a "feedforward + feedback" composite control. The noise covariance matrix characterizes the statistical characteristics of system noise (such as variance and correlation). Combined with the observed noise covariance matrix, the measurement error is estimated in real time, achieving optimal filtering of random noise (similar to the Kalman filter principle), reducing the noise component in the original error signal and improving the signal-to-noise ratio. The filter gain dynamically adjusts based on noise characteristics. When environmental interference intensifies (e.g., noise covariance increases), the weight of observed values ​​is automatically increased, strengthening reliance on real-time data. When the system stabilizes, the weight of model predictions is increased, reducing the frequency of control actions and balancing control accuracy with actuator life. Model prediction errors are separated from measurement errors to prevent sensor noise (e.g., thermometer drift) from being misinterpreted as true environmental fluctuations, thereby reducing the probability of false actions. A sliding average (averaging three consecutive parameter values) filters short-term high-frequency noise and extracts low-frequency trend components of the error signal (e.g., a continuous increase in temperature / decrease in humidity), preventing over-control caused by random errors. Temperature and humidity error sequences are processed independently, preserving both coupling and independence (e.g., temperature fluctuations may be accompanied by humidity changes, but the interference sources are different). This allows the system to optimize control parameters for the different error characteristics of temperature and humidity, avoiding the "one loses the other" problem of single-variable control. By comparing the time series of changes in the temperature and humidity error variance matrix, we can assist in determining the type of interference (for example, a sudden increase in temperature variance may be due to a heating device failure, while a sudden increase in humidity variance may be due to an abnormality in the humidification equipment). This provides data support for fault diagnosis and maintenance, and reduces the cost of manual inspections.

[0096] In another preferred embodiment of the present invention, the temperature deviation value and the humidity deviation value are calculated respectively according to the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error, including:

[0097] The optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error are aligned point by point according to the sampling timestamp. From the aligned optimized parameter sequence of the temperature prediction error, the change rate of the data before and after the current sampling point is calculated to obtain the dynamic gradient eigenvalue reflecting the dynamic change trend of the temperature error sequence; from the optimized parameter sequence of the humidity prediction error, the absolute value of the difference between adjacent data is calculated to obtain the quantitative fluctuation eigenvalue representing the instantaneous fluctuation degree of the humidity error, which specifically includes:

[0098] Optimization parameter sequence for temperature prediction error Optimized parameter sequence with humidity prediction error All samples are taken at equal time intervals (the sampling period is ), timestamps are strictly synchronized (e.g. is the sampling time).

[0099] Implementation:

[0100] Directly align the parameter values ​​at the same time through the timestamp index to form an aligned two-dimensional sequence: ;

[0101] in, is the sliding window length (e.g. corresponds to the data of the first two moments).

[0102] Boundary processing:

[0103] Initial stage (t<2):

[0104] When the current sampling point When it is less than the window length (such as t = 0 or t = 1), forward difference or backward difference is used instead of bidirectional difference.

[0105] Outlier filtering:

[0106] If the sensor data at a certain moment is abnormal (such as the temperature and humidity jump exceeding the threshold), then skip this point and use the linear interpolation of the data before and after to complete it before participating in the calculation.

[0107] By calculating the slope of the optimized parameter sequence of the temperature prediction error before and after the current point, the rate of error change (upward / downward trend and speed) is quantified, and the dynamic gradient is extracted using the central difference method or sliding window linear regression. Specific implementation steps:

[0108] At the current sampling point As the center, take the previous moment and the next moment Data (3 points in total) form a local window .

[0109] Use the data of three points at adjacent moments to fit a straight line, and the slope reflects the trend near the current point (avoiding interference from sudden changes at a single point).

[0110] The central difference method calculates the gradient:

[0111] Assume that the coordinates of the three points are , then the central difference gradient is:

[0112] ;

[0113] The numerator represents the average of the backward difference and the forward difference (eliminating the one-sided bias); the denominator represents the time interval (normalized to the rate of change per unit time).

[0114] Boundary processing (t=0 or t=N-1, N is the sequence length):

[0115] Starting point (t=0): Use backward difference ;

[0116] End point (t=N−1): Use forward difference .

[0117] The humidity quantitative fluctuation characteristic value calculation (reflecting instantaneous fluctuations) is calculated by calculating the absolute value of the difference between adjacent data in the optimized parameter sequence of the humidity prediction error, quantifying the instantaneous change amplitude of the error, and reflecting the suddenness of environmental interference (such as sudden changes in humidity caused by the start and stop of humidification equipment).

[0118] The specific implementation steps are as follows:

[0119] Adjacent difference calculation, for the optimized parameter sequence of humidity prediction error, calculate the absolute difference between the current point and the previous moment:

[0120] ;

[0121] Single-step difference reflects the degree of mutation between adjacent moments, and the absolute value eliminates the influence of positive and negative directions and focuses on the amplitude of fluctuation.

[0122] Multi-step differential expansion (enhanced robustness):

[0123] If you need to reflect the short-term fluctuation trend, you can calculate the average of the absolute values ​​of the consecutive n-step differences:

[0124] ;

[0125] Typical value: When n = 3, the absolute value of the difference between the current point and the previous three moments is calculated to suppress occasional noise.

[0126] Dynamic threshold filtering:

[0127] Set the fluctuation amplitude threshold (such as determined based on the standard deviation of historical data), when When , it is judged as significant fluctuation, otherwise it is set to zero: .

[0128] Dynamic threshold filtering is used to distinguish normal fluctuations from sudden interference to avoid false triggering of control actions.

[0129] An initial weight is generated based on the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue, and the initial weight is corrected using the historical humidity deviation ratio to obtain a normalized weight coefficient. The difference between the current element of the temperature error variance matrix and the target threshold is multiplied by the normalized weight coefficient to obtain a basic deviation. The basic deviation is processed based on the instantaneous intensity value in the quantized fluctuation eigenvalue to obtain a temperature deviation value.

[0130] The ratio of the current element of the humidity error variance matrix to the target threshold is transformed into a natural logarithm to generate an initial deviation. The compensation factor is generated according to the product of the accumulated temperature amplitude and the instantaneous humidity fluctuation intensity. The initial deviation is dynamically compressed using the compensation factor to obtain the humidity deviation value.

[0131] In embodiments of the present invention, by calculating the rate of change of data before and after the current sampling point in the optimized parameter sequence for temperature prediction error, the dynamic trend of the temperature error sequence can be keenly captured. This allows the system to not only monitor the current temperature error but also predict future trends, enabling it to proactively implement appropriate control measures, thereby enhancing the system's foresight and adaptability. For example, when a positive and large dynamic gradient eigenvalue of the temperature error is detected, it indicates that the temperature error is rapidly increasing. The system can promptly adjust the power of the heating or cooling equipment to prevent the temperature from further deviating from the target value. Calculating the absolute value of the difference between adjacent data points in the optimized parameter sequence for humidity prediction error accurately quantifies the instantaneous fluctuation of the humidity error. This helps the system promptly detect sudden changes in humidity and monitor and evaluate humidity fluctuations in real time. For example, a large quantified fluctuation eigenvalue indicates that humidity has fluctuated significantly within a short period of time, allowing the system to quickly implement measures to stabilize humidity. Humidity environments are often susceptible to interference from various factors, such as the start and stop of ventilation and humidification equipment. The quantified fluctuation eigenvalue can effectively reflect the impact of these interferences on the humidity error, enabling the system to promptly adjust its control strategy and enhance its resistance to humidity interference.

[0132] Initial weights are generated by calculating the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue, and then corrected based on the historical humidity deviation ratio. The resulting normalized weight coefficient comprehensively considers the dynamic changes and historical data of temperature and humidity errors. This allows for a more comprehensive consideration of the relationship between temperature and humidity and their changing trends when calculating temperature deviation values, improving the rationality and accuracy of the weight coefficient. The normalized weight coefficient adaptively adjusts based on the dynamic changes in temperature and humidity errors. When the temperature error changes significantly, the weight will be biased toward the temperature factor accordingly; when the humidity error fluctuates significantly, the weight will appropriately consider the influence of the humidity factor. This adaptive adjustment mechanism enables the system to better adapt to different environmental conditions and error changes.

[0133] The base deviation is calculated by multiplying the difference between the current element of the temperature error variance matrix and the target threshold by a normalized weight coefficient. This calculation method fully considers the magnitude and weight of the temperature error, and the base deviation more accurately reflects the degree of deviation of the temperature error from the target threshold. The base deviation is processed based on the instantaneous intensity value of the quantized fluctuation eigenvalue to obtain the temperature deviation value, further accounting for the impact of humidity fluctuations on temperature control. In actual incubation environments, humidity fluctuations may interfere with temperature measurement and control. This processing method can, to a certain extent, eliminate the impact of humidity fluctuations on the temperature deviation calculation, thereby improving the stability of temperature control.

[0134] The initial deviation is generated by applying a natural logarithm transformation to the ratio of the current element of the humidity error variance matrix to the target threshold. This logarithmic transformation can compress a wide range of humidity error variance ratios, making the initial deviation more reasonable and easier to handle. Furthermore, the logarithmic transformation can highlight relative changes in the humidity error variance ratios, more accurately reflecting the degree of humidity error deviation from the target threshold. A compensation factor is generated by multiplying the temperature cumulative amplitude by the instantaneous humidity fluctuation intensity. This compensation factor is used to dynamically compress the initial deviation. This approach accounts for the interaction between temperature and humidity and prevents excessive fluctuations in humidity deviation. For example, when the temperature cumulative amplitude is large and the instantaneous humidity fluctuation intensity is also high, the compensation factor significantly compresses the initial deviation, preventing system instability caused by over-response. Dynamic range compression keeps the humidity deviation within a reasonable range, avoiding drastic adjustments to control parameters due to sudden increases or decreases in humidity error. This helps improve the stability of humidity control and maintains the humidity in the incubation environment within a relatively stable range, promoting healthy hatching of chick embryos.

[0135] In another preferred embodiment of the present invention, an initial weight is generated based on the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue, and the initial weight is corrected using the humidity historical deviation ratio to obtain a normalized weight coefficient; the difference between the current element of the temperature error variance matrix and the target threshold is multiplied by the normalized weight coefficient to obtain a basic deviation; and the basic deviation is processed based on the instantaneous intensity value in the quantized fluctuation eigenvalue to obtain a temperature deviation value, including:

[0136] Calculate the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue to obtain a ratio that reflects the relative intensity of the temperature and humidity error changes, which is used as the initial weight.

[0137] Calculate the ratio of humidity error historical data that deviates from the normal range, that is, the humidity historical deviation ratio, and use the humidity historical deviation ratio to adjust the initial weight to obtain the corrected initial weight;

[0138] The corrected initial weight is normalized to obtain a normalized weight coefficient;

[0139] Obtain the current element of the temperature error variance matrix, that is, the average value of the squared deviation of the temperature error in the current sliding window, calculate the difference between the current element and the preset temperature error target threshold, and fuse the difference with the normalized weight coefficient to obtain the basic deviation value that reflects the degree of deviation of the temperature error fluctuation from the target state;

[0140] The instantaneous intensity value at the current moment, that is, the fluctuation peak value of the current humidity error, is extracted from the quantized fluctuation characteristic value. The basic deviation is adjusted using the instantaneous intensity value to obtain the temperature deviation value reflecting the temperature error and humidity interference.

[0141] In this embodiment of the present invention, the relative impact of the temperature error is quantified by calculating the ratio of the dynamic gradient eigenvalue of the temperature error (reflecting the error change rate, such as the heating / cooling trend) to the quantitative fluctuation eigenvalue of the humidity error (reflecting the instantaneous fluctuation amplitude, such as the start-stop interference of the humidification equipment). The formula is:

[0142] ,in, Minimize the value to avoid the denominator being zero. Temperature and humidity are often coupled in the incubation environment (for example, heating may be accompanied by a decrease in humidity), but the interference sources are different (temperature fluctuations are mostly caused by heating / cooling equipment, and humidity fluctuations are mostly caused by humidification / ventilation). The current dominant interference type can be dynamically determined by the ratio:

[0143] If the ratio is greater than 1, it indicates that the temperature error change rate is dominant (for example, the temperature continues to deviate due to the continuous operation of the heating device), and the weight is tilted towards temperature;

[0144] If the ratio is less than 1, it means that the instantaneous fluctuation of humidity is more significant (such as sudden ventilation causing a sudden drop in humidity), and the weight is tilted towards humidity correction.

[0145] The dynamic gradient eigenvalue contains trend information in the time dimension (such as the rising trend of temperature error for three consecutive cycles). Compared with single-cycle data, it can predict the risk of temperature deviation 2-3 sampling cycles in advance, allowing the system to adjust control parameters before the interference expands.

[0146] Calculate the proportion of historical humidity error data that deviates from the normal range (for example, the proportion of cycles in which the humidity error exceeds the threshold in the past N cycles). Use this proportion to correct the initial weight. The corrected initial weight = initial weight × (1 − historical humidity deviation proportion).

[0147] If the historical humidity deviation ratio is low (e.g., <20%), it indicates that the current humidity fluctuation may be accidental noise (e.g., instantaneous sensor error). The initial weight after correction is more dependent on the temperature trend to avoid excessive control triggered by accidental humidity fluctuations. If the ratio is high (e.g., >50%), it indicates that there is a persistent humidity anomaly (e.g., humidification equipment failure). The weight is tilted towards humidity correction, giving priority to stabilizing the humidity environment.

[0148] Short-term noise is filtered out by the statistical characteristics of historical data. For example, when the hatchery ventilation system starts and stops periodically (for example, ventilating for 5 minutes every hour), causing regular fluctuations in humidity, the historical deviation ratio can identify this pattern and adaptively adjust the weight to avoid frequent oscillations in control parameters.

[0149] The corrected initial weights are normalized (e.g. scaled to the interval [0, 1]) using the formula:

[0150] ;

[0151] Temperature error variance (unit: ℃ 2 ) has a different dimension from the humidity error fluctuation (unit: %RH). The normalized weight coefficient is a dimensionless value, ensuring that the temperature and humidity influences are weighted and summed in the same dimension, thus avoiding control deviations caused by dimensional differences.

[0152] At different stages of incubation (e.g., strict temperature control in the early stage and attention to humidity during hatching), the maximum / minimum weights are dynamically updated through historical data to achieve phased adaptation of the control strategy. For example:

[0153] Early embryonic development: Normalized weight coefficients are biased towards temperature (e.g., 0.7-0.9), ensuring that temperature fluctuations are ≤±0.2°C;

[0154] Shelling stage: The weight coefficient is biased towards humidity (such as 0.5-0.7), and the humidity is prioritized to be within the range of ±3%RH.

[0155] Calculate the current element of the temperature error variance matrix (denoted as ) and the target threshold (denoted as ), multiplied by the normalized weight coefficient (denoted as ω), the formula is:

[0156] ;

[0157] A positive difference indicates that the temperature error fluctuation exceeds the target (e.g., uneven heating causes increased temperature fluctuation), and control efforts need to be strengthened. A negative difference indicates that the fluctuation is within the controllable range, and control actions can be reduced to extend the life of the actuator.

[0158] Weight modulation controls sensitivity:

[0159] When the normalization weight coefficient ω is high (such as 0.8), the basic deviation is more sensitive to temperature errors and is suitable for scenarios with frequent interference. When ω is low (such as 0.3), the control is smoother and is suitable for the stable stage.

[0160] Extract the instantaneous intensity value at the current moment (denoted as , such as the absolute peak value of humidity error in adjacent cycles), and correct the basic deviation:

[0161] , where k is the adjustment coefficient, 0<k<1.

[0162] Humidity transient fluctuations may indirectly affect temperature measurement and control through physical coupling (such as changes in the thermal conductivity of moist air). For example, during humidification, moist air absorbs heat, causing the temperature sensor to misjudge temporarily. Increase, the correction item makes the temperature deviation value appropriately reduced to avoid the system misjudging it as real temperature interference. When temperature and humidity are abnormal at the same time (such as heating equipment failure + humidification pump leakage), through The basic deviation is compressed to avoid the deterioration of another variable caused by the control of a single variable (such as excessive heating may aggravate the decrease in humidity), and multi-variable coordinated control is achieved.

[0163] In another preferred embodiment of the present invention, a natural logarithm transformation is performed on the ratio of the current element of the humidity error variance matrix to the target threshold to generate an initial deviation; a compensation factor is generated according to the product of the temperature cumulative amplitude value and the instantaneous humidity fluctuation intensity, and the initial deviation is dynamically range compressed using the compensation factor to obtain a humidity deviation value, including:

[0164] Get the current element of the humidity error variance matrix, that is, the average value of the square of the humidity error deviation in the current sliding window, and calculate the humidity ratio of the current element to the preset humidity error target threshold to reflect the deviation multiple of the current humidity error fluctuation relative to the target state;

[0165] The natural logarithm of the humidity ratio is taken and the numerical range is compressed to the logarithmic scale to obtain the initial deviation reflecting the degree of humidity error deviation;

[0166] Calculate the absolute value cumulative sum of the optimized parameter sequence of the humidity prediction error within the current sliding window to characterize the overall deviation degree and duration of the temperature error and form the temperature cumulative amplitude value;

[0167] Extract the instantaneous fluctuation peak value at the current moment from the optimized parameter sequence of humidity prediction error, that is, the maximum absolute value of the difference between adjacent sampling points, and record it as the humidity instantaneous fluctuation intensity;

[0168] Multiply the temperature cumulative amplitude value by the instantaneous humidity fluctuation intensity to obtain the compensation factor;

[0169] The initial deviation is adjusted by the compensation factor, and the numerical range of the initial deviation is narrowed by nonlinear mapping to obtain the humidity deviation value that balances the coupling effect of temperature and humidity.

[0170] In the embodiment of the present invention, the humidity error variance current element is calculated With target threshold to obtain the humidity ratio; take the natural logarithm of the humidity ratio to generate the initial deviation:

[0171] ;

[0172] Numerical range compression and dynamic adaptation, when the humidity error variance significantly exceeds the target threshold (such as ), the humidity ratio is 10, and the natural logarithm is compressed to 2.3 to avoid the drastic fluctuation of the control parameters caused by the original ratio of 10;

[0173] When the error is small (such as ), the logarithmic output is -0.69, retaining the negative deviation information (insufficient humidity fluctuation), which makes it easier for the system to determine whether enhanced humidity regulation is needed.

[0174] The logarithmic transformation converts the "absolute error" into a "relative error multiple", for example:

[0175] The target threshold is 1%RH 2 When the current variance is 4%RH 2 (ratio 4) and 9%RH 2 The logarithmic difference of (ratio 9) is , reflecting that the deviation of the latter is 2.25 times that of the former, rather than the absolute difference of 5%RH 2 , which is more in line with people’s intuitive judgment of “error severity”.

[0176] Calculate the absolute value cumulative sum of the optimized parameter sequence of the humidity prediction error within the current sliding window (3 periods). The formula is: Optimize parameters for temperature prediction error.

[0177] A temperature error in a single cycle may be occasional noise (such as sensor jitter), but the cumulative sum of three consecutive cycles can reflect a persistent deviation (such as the aging of the heating element causing the temperature to be continuously high). For example:

[0178] like , , The cumulative amplitude value is 0.45℃, which shows that the temperature error is increasing and we need to be vigilant about the risk of subsequent loss of control.

[0179] Temperature and humidity are physically coupled (for example, water evaporation is accelerated in high temperature environments), and the cumulative amplitude value can quantify the long-term impact of temperature on humidity. For example:

[0180] Continuous high temperature in the incubator (large cumulative amplitude value) will cause the humidity to naturally drop. At this time, the humidity deviation value needs to be dynamically adjusted based on the temperature impact to avoid the system mistakenly judging it as a humidification equipment failure.

[0181] The maximum absolute value of the difference between adjacent sampling points in the optimized parameter sequence of humidity prediction error is extracted. The formula is:

[0182] ;

[0183] in, For the Optimization parameters of humidity prediction error at each sampling point.

[0184] When events such as the humidification equipment starting or stopping, or the vents suddenly opening, occur, the humidity error will experience instantaneous spikes (for example, the absolute value of the difference suddenly increases to 5%RH). The instantaneous fluctuation intensity can quickly identify such sudden interference.

[0185] Differentiate the types of interference:

[0186] If the instantaneous fluctuation intensity is high but the cumulative amplitude value is low (such as a single ventilation causing a sudden drop in humidity), it means that the interference is of short-term pulse type, and the control strategy is mainly based on rapid compensation; if both are high (such as the humidification pump continues to leak water, causing continuous fluctuations in humidity), it indicates that there is a persistent fault, and the control efforts need to be strengthened and the fault warning needs to be triggered.

[0187] The compensation factor is obtained by multiplying the temperature cumulative amplitude value by the instantaneous humidity fluctuation intensity:

[0188] Compensation factor = temperature cumulative amplitude value × humidity instantaneous fluctuation intensity;

[0189] The initial deviation is compressed by the compensation factor:

[0190] , where λ is the adjustment coefficient, λ>0.

[0191] When both temperature and humidity disturbances are present (e.g., heating and ventilation), the compensation factor increases, compressing the initial deviation range. This prevents the system from overreacting to a single variable and causing degradation of the other variable. For example, if the cumulative temperature amplitude is 0.5°C (high) and the instantaneous humidity fluctuation is 4%RH (a sudden drop), the compensation factor is 2.0, and the compression coefficient is 1 / (1 + 2λ). If λ = 0.5, the compressed deviation is 1 / 2 of the initial value, preventing the heating action from exacerbating the humidity drop. The compression function is hyperbolic, exhibiting a linear response for small errors and a saturated response for large errors. When the compensation factor is small (weak disturbance), the compression coefficient approaches 1, and the deviation is close to the initial value, resulting in sensitive control. When the compensation factor is large (strong disturbance), the compression coefficient approaches 0, preventing control parameters from exceeding the actuator's capabilities (e.g., the maximum power limit of a humidifier), thus ensuring system safety.

[0192] In another preferred embodiment of the present invention, the temperature deviation value and the humidity deviation value are normalized and weighted and summed, and dynamically scaled by a Gaussian kernel function to obtain a correction factor, including:

[0193] Calculating the maximum and minimum values ​​of the temperature deviation value in a preset sampling period respectively, and calculating the first numerical deviation range thereof, that is, the maximum value minus the minimum value; subtracting the minimum value from the current temperature deviation value to obtain a first deviation difference, and calculating the deviation ratio between the first deviation difference and the first numerical deviation range to obtain a normalized temperature deviation value;

[0194] Calculating the maximum and minimum values ​​of the humidity deviation value in a preset sampling period respectively, and calculating its second numerical deviation range, that is, the maximum value minus the minimum value; subtracting the minimum value from the current humidity deviation value to obtain a second deviation difference, and calculating the deviation ratio between the second deviation difference and the second numerical deviation range to obtain a normalized humidity deviation value;

[0195] Preset the temperature influence weight coefficient and humidity influence weight coefficient, multiply the normalized temperature deviation value by the temperature influence weight coefficient to obtain the temperature contribution; multiply the normalized humidity deviation value by the humidity influence weight coefficient to obtain the humidity contribution;

[0196] Add the temperature contribution and the humidity contribution to obtain the temperature and humidity deviation index value; substitute the temperature and humidity deviation index value into the Gaussian kernel function and calculate its function value, which is the correction factor.

[0197] In the embodiment of the present invention, the temperature deviation value (denoted as ) and humidity deviation (denoted as ) are normalized respectively, and the formula is: ;

[0198] in, Indicates the temperature deviation value in the preset sampling period. ;in, It indicates the humidity deviation value in the preset sampling period. The preset sampling period is the first 3-5 sampling periods, reflecting the short-term fluctuation characteristics.

[0199] The temperature deviation value (unit: °C) and the humidity deviation value (unit: %RH) have different dimensions. After normalization, they are both mapped to the interval [0, 1] to ensure that their effects are comparable when weighted summation is performed. For example:

[0200] The temperature deviation value of +0.3°C (0.8 after normalization) and the humidity deviation value of -4%RH (0.7 after normalization) can be directly superimposed according to the weights to avoid the problem of "the larger temperature dimension leads to biased control".

[0201] Dynamically adjust the normalization range based on the maximum / minimum values ​​of the preset data, rather than a fixed threshold, to adapt to the error fluctuation characteristics of different incubation stages:

[0202] Early embryonic development: Temperature and humidity are strictly controlled, with a small error range (e.g., temperature deviation of ±0.1°C), and even subtle changes can be captured after normalization;

[0203] Shelling period: The allowable range of humidity fluctuation is expanded (such as ±5%RH), and after normalization, control saturation due to large absolute values ​​is avoided.

[0204] Preset the temperature influence weight coefficient (α) and humidity influence weight coefficient (β), satisfying α + β = 1, and calculate the temperature and humidity contribution:

[0205] Temperature contribution = normalized temperature deviation × α;

[0206] Humidity contribution = normalized humidity deviation × β;

[0207] Temperature and humidity deviation index value = temperature contribution + humidity contribution.

[0208] By adjusting α and β, the control focus can be switched dynamically:

[0209] Early incubation: set α = 0.7, β = 0.3, and prioritize temperature control (the embryo is more sensitive to temperature);

[0210] Late incubation period: set α = 0.5, β = 0.5, and balance the temperature and humidity control (appropriate humidity is required for hatching).

[0211] When temperature and humidity deviate at the same time, weighted summation can avoid a single variable dominating the control. For example:

[0212] The normalized deviation value of temperature is 0.9 (high deviation) and the humidity is 0.9 (high deviation). After summing the values ​​with equal weights, the index value is 0.9, and the system triggers a "strong interference" response. If the temperature is 0.9 but the humidity is 0.1, the index value is 0.5. The system determines that the temperature interference is the main factor, and avoids amplifying the control action due to slight fluctuations in humidity.

[0213] The Gaussian kernel function is used to perform nonlinear transformation on the temperature and humidity deviation index value (denoted as S). The formula is:

[0214] ;

[0215] Among them, σ is the kernel function bandwidth, which reflects the sensitivity to interference.

[0216] The function curve is bell-shaped. When S=0, the output is 1 (no interference), and the output decays exponentially as S increases.

[0217] Nonlinear response characteristics:

[0218] Sensitive to small disturbances: When S is in the range [0, 0.5], the function value decreases gently (for example, the output is about 0.86 when S = 0.3), ensuring that the system responds promptly to early minor disturbances;

[0219] Saturation in case of large disturbance: When S>1, the function value approaches 0 (e.g., the output is about 0.14 when S=2), thus avoiding oscillation caused by excessive adjustment of control parameters under strong disturbance.

[0220] Dynamically adjust interference sensitivity:

[0221] By adjusting the kernel function bandwidth σ, it can adapt to different interference scenarios:

[0222] Stable scenario (e.g., no personnel operating at night): Set σ = 0.3, the function is sensitive to small fluctuations, and maintains high-precision control;

[0223] High-interference scenarios (such as frequent opening and closing of warehouse doors during the day): Set σ = 0.6. The function responds more smoothly to large fluctuations, reducing ineffective control actions.

[0224] Through normalization and weighted summation, the two-dimensional deviation problem of temperature and humidity is converted into a one-dimensional index value, which simplifies the subsequent control algorithm design (for example, the BP neural network only requires a single input).

[0225] Physical mapping of interference strength:

[0226] The correction factor value is negatively correlated with the severity of the interference (the smaller the value, the greater the interference), which is consistent with the control logic intuition. For example:

[0227] Correction factor = 0.9, indicating slight interference, only fine-tuning of PID parameters is required;

[0228] Correction factor = 0.2 indicates strong interference, which triggers a significant adjustment of the fuzzy rule base weights and a rapid optimization of the PID parameters.

[0229] The smoothness of the Gaussian kernel function has a natural inhibitory effect on outliers. For example:

[0230] At a certain moment, due to a sensor failure, the temperature and humidity deviation values ​​suddenly increased (all normalized to 1, S = 1). The correction factor output was 0.606 (non-zero), avoiding false action caused by a single bad point. Compared with directly using the index value control, the false alarm rate was reduced by 50%. The correction factor provided a normalized, low-noise comprehensive interference index for the subsequent BP neural network and fuzzy control, which improved the convergence speed of the neural network and the efficiency of the weight reconstruction of the fuzzy rule base.

[0231] In another preferred embodiment of the present invention, the activation function slope of the hidden layer of the neural network is dynamically adjusted according to the correction factor and the BP neural network, and the weight of the fuzzy rule base is reconstructed based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight, including:

[0232] Determine the initial slope parameter corresponding to the activation function of the hidden layer of the BP neural network;

[0233] Determine the adjustment amount of the slope of the activation function according to the value of the correction factor to obtain the adjusted value of the slope;

[0234] Add the adjusted slope value to the initial slope parameter to obtain the adjusted activation function slope;

[0235] Analyze the numerical distribution characteristics of the correction factor and calculate the numerical distribution characteristics, which include the mean, variance, maximum value and minimum value of the correction factor;

[0236] The initial weight of the fuzzy rule base is determined, and the weight of each rule in the fuzzy rule base is adjusted according to the numerical distribution characteristics of the correction factor to obtain the corrected weight of the fuzzy rule base.

[0237] In the embodiment of the present invention, the hidden layer activation function is assumed to be a Sigmoid function, and the initial slope parameter is k0 (which determines the steepness of the function curve).

[0238] The adjustment amount is calculated based on the correction factor (denoted as γ, with a value range of [0, 1]) to design the adjustment amount Δk = η × (1 − γ) (η is the learning rate, > 0).

[0239] Adjusted slope: .

[0240] When the correction factor γ is small (interference is strong), As it increases, the slope k increases, the Sigmoid function curve becomes steeper, and the neural network's ability to fit strong nonlinear error features (such as sudden changes in temperature and humidity) is enhanced. For example:

[0241] When γ = 0.2, k increases from its initial value of 2.0 to 2.8, and the derivative of the function around an input value of 0 increases from 0.5 to 0.68, making it more sensitive to subtle errors. When γ is larger (weaker interference), the slope k decreases, and the function curve flattens, preventing the network from overfitting to noise and improving generalization.

[0242] Dynamically adjusting the slope can accelerate model convergence. For example:

[0243] In strong interference scenarios (such as device startup and shutdown), increasing the slope enables the network to capture error characteristics faster and reduces the number of iterations by 25%. In stable scenarios, decreasing the slope reduces the gradient update amplitude, prevents oscillation, and makes the convergence process smoother.

[0244] Calculate the statistical characteristics of the correction factor, mean μ, variance σ 2 Reflects the fluctuation degree of interference, extreme value (γ max , γ min ), used to identify extreme disturbance events (such as γ min <0.1 indicates sudden strong interference).

[0245] If the mean μ is less than 0.5 and the variance σ 2 >0.1, indicating that the system is in a state of "medium-strong interference and high volatility" (such as frequent opening and closing of warehouse doors), and the dynamic adaptability of fuzzy rules needs to be enhanced;

[0246] If the mean μ>0.8 and the variance σ 2 <0.05 indicates that the system is stable and the frequency of rule adjustment can be reduced to save computing resources. Statistical features are calculated by sliding windows (such as the first 10 cycles), taking into account short-term interference (such as the current cycle γ min ) and long-term trends (such as a downward trend in the mean) to avoid misjudgment of single-period data.

[0247] Assume that the fuzzy rule base has m rules and the initial weight is w j ∈[0,1](j=1,2,…,m), usually uniformly distributed (such as w j =1 / m).

[0248] Dynamic adjustment strategy:

[0249] For each rule, adjust the weight according to the statistical characteristics of the correction factor :

[0250] ;

[0251] in, is the historical correction factor sample corresponding to rule j, λ is the adjustment coefficient, ϵ is the minimum value, and the denominator is zero).

[0252] For rules that perform well in strong interference scenarios (e.g., rule j has good control effect when γ < 0.3), its weight Increase, trigger priority;

[0253] For outdated or invalid rules (such as rules that are only applicable to the early incubation stage), the weight is reduced or even disabled, which solves the defects of fixed traditional fuzzy control rules.

[0254] Suppress rule conflicts:

[0255] When multiple rule outputs conflict (e.g., "heating" and "humidity reduction" rules are triggered at the same time), the rule with the higher weight takes the lead in control to avoid control failure caused by "averaging". For example:

[0256] During the hatching period, the weight of the "increase humidity" rule is increased from 0.1 to 0.4, taking precedence over the "temperature fine-tuning" rule to ensure that the humidity meets the standard.

[0257] By adjusting the slope of the activation function, it can quickly adapt to the nonlinear changes of error characteristics and reduce the fitting error of complex interference (such as temperature and humidity coupled oscillations); reconstructing the rule weights based on the correction factor distribution upgrades the fuzzy logic from "fixed threshold judgment" to "data-driven dynamic reasoning", thereby improving the accuracy of rule matching.

[0258] Multi-level response to interference intensity:

[0259] Short-term response: The activation function slope is adjusted in real time (calculation delay < 10ms) to quickly suppress sudden interference;

[0260] Long-term learning: It iteratively updates the fuzzy rule weights (once every 5 minutes) by modifying the distribution characteristics of the factors, and gradually optimizes the control strategy.

[0261] Slope adjustment only involves the hidden layer parameters (the calculation amount is O(n h ), n h is the number of hidden layer neurons), which is much lower than the full parameter update of traditional BP network; the fuzzy rule weights are adjusted in batches according to statistical characteristics, without the need for manual debugging of each rule.

[0262] In another preferred embodiment of the present invention, a three-dimensional parameter adjustment surface is constructed based on the modified fuzzy rule base weights, and the proportional, integral, and differential parameters of the PID controller are dynamically adjusted by a surface gradient search algorithm to generate a control signal, including:

[0263] Determine the three variables of the three-dimensional parameter adjustment surface, which are temperature deviation value, humidity deviation value and the weight of the modified fuzzy rule base;

[0264] Under different combinations of temperature deviation values, humidity deviation values ​​and modified fuzzy rule base weights, the corresponding proportional, integral and differential parameter values ​​of the PID controller are recorded;

[0265] According to the values ​​of the proportional, integral and differential parameters of the PID controller, a three-dimensional parameter adjustment surface is constructed through neural network fitting;

[0266] According to the current temperature deviation value, humidity deviation value and the weight of the modified fuzzy rule base in the incubator, the corresponding state point is found on the three-dimensional parameter adjustment surface. The state point represents the initial parameter setting of the PID controller under the current working condition.

[0267] At the current state point, the gradient of the three-dimensional parameter adjustment surface is calculated. The gradient represents the direction of change of the three-dimensional parameter adjustment surface at the state point. It contains three components, which correspond to the change directions of the proportional, integral and differential parameters of the PID controller respectively.

[0268] The gradient and search step of the surface are adjusted according to the three-dimensional parameters, and the proportional, integral and differential parameters of the PID controller are adjusted in the opposite direction of the gradient until the search step is reached to obtain the optimized proportional, integral and differential parameters of the PID controller;

[0269] The PID controller calculates the corresponding control signal based on the current temperature deviation value and humidity deviation value, combined with the adjusted proportional, integral and differential parameters.

[0270] In the embodiment of the present invention, the temperature deviation value , humidity deviation value , modified fuzzy rule weight As input variables, construct PID parameters The three-dimensional mapping relationship is as follows:

[0271] ;

[0272] The neural network is trained with historical data to fit the function and form a continuous and smooth parameter adjustment surface.

[0273] Traditional PID parameter tuning only considers single variable error (such as temperature), while the surface considers temperature and humidity coupling interference Adaptability to fuzzy rules Take into consideration at the same time.

[0274] For example: When the temperature deviates greatly , humidity deviation is small And the fuzzy rule weight is biased towards temperature When the surface automatically outputs high K p (rapid heating), low K iParameter combinations (to avoid integral saturation). Through training with historical operating data, the surface covers possible interference combinations throughout the incubation cycle (such as temperature ±0.5°C, humidity ±8%RH, and rule weights of 0.3-0.9). Thousands of sets of optimized parameters are pre-stored to avoid high latency in online real-time calculations.

[0275] According to the current temperature and humidity deviation value and the fuzzy rule weight, the corresponding state point is located on the three-dimensional surface. The coordinates of this point are , output the initial PID parameters .

[0276] Compared with traditional Ziegler-Nichols and other trial-and-error methods (which require tens of seconds to minutes of iteration), surface interpolation can obtain initial parameters within 10ms, and the response speed to sudden interference (such as sudden changes in temperature and humidity caused by the opening of the warehouse door) is increased by 90%. When the incubation stage switches (such as from early to mid-stage), the fuzzy rule weights Changes trigger state point migration, and PID parameters change continuously along the surface, avoiding temperature overshoot caused by parameter jumps in traditional segmented control (overshoot amplitude is reduced by 60%).

[0277] Calculate the surface gradient at the state point , the gradient direction indicates the direction in which the parameter is improved the fastest, and the opposite direction indicates the direction in which the error decreases the fastest.

[0278] Parameter adjustment: Update parameters in the opposite direction of the gradient:

[0279] ;

[0280] in, is the step size, which controls the adjustment amplitude.

[0281] Based on the initial parameters, it takes 3-5 gradient iterations to converge to the local optimal solution, which is 80% less than the number of random search iterations. For example:

[0282] When the initial temperature fluctuation is ±0.5℃, the fluctuation is reduced to within ±0.2℃ after 4 iterations; the gradient direction guides the parameter adjustment along the steepest descent path of the surface, avoiding the oscillation caused by blind parameter adjustment in the traditional trial and error method. For example: in humidity control, the differential parameter K d It gradually increases in the opposite direction of the gradient to smoothly suppress overshoot rather than repeatedly increasing and decreasing to cause fluctuations.

[0283] like Figure 2 As shown, an embodiment of the present invention further provides an intelligent incubation warehouse anti-interference control system based on error self-learning, comprising:

[0284] The acquisition module 100 is used to collect the temperature and humidity data stream of the chicken embryo incubation in the incubation chamber in real time; the data of three consecutive sampling cycles are used to form a sliding detection window, the temperature and humidity data are integrated within the window, and the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error are output;

[0285] A calculation module 200 is used to calculate a temperature deviation value and a humidity deviation value according to an optimized parameter sequence of a temperature prediction error and an optimized parameter sequence of a humidity prediction error, respectively;

[0286] Correction module 300, for performing normalized weighted summation on the temperature deviation value and the humidity deviation value, and dynamically scaling them using a Gaussian kernel function to obtain a correction factor;

[0287] An adjustment module 400 is used to dynamically adjust the slope of the activation function of the hidden layer of the neural network according to the correction factor and the BP neural network, and reconstruct the weights of the fuzzy rule base based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weights;

[0288] The processing module 500 is used to construct a three-dimensional parameter adjustment surface based on the modified fuzzy rule base weights, and dynamically adjust the proportional, integral and differential parameters of the PID controller through a surface gradient search algorithm to generate a control signal.

[0289] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0290] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. An intelligent incubation warehouse anti-interference control method based on error self-learning, characterized in that: The method comprises: Step 1: Real-time acquisition of temperature and humidity data streams for incubating chicken embryos in an incubation chamber; forming a sliding detection window with data from three consecutive sampling cycles, fusing the temperature and humidity data within the window, and outputting an optimized parameter sequence for temperature prediction error and an optimized parameter sequence for humidity prediction error; Step 2, according to the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error, respectively calculating the temperature deviation value and the humidity deviation value, including: aligning the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error point by point according to the sampling timestamp, calculating the change rate of the data before and after the current sampling point from the aligned optimized parameter sequence of the temperature prediction error, and obtaining the dynamic gradient eigenvalue reflecting the dynamic change trend of the temperature error sequence; calculating the absolute value of the difference between adjacent data from the optimized parameter sequence of the humidity prediction error, and obtaining the quantitative fluctuation eigenvalue representing the instantaneous fluctuation degree of the humidity error; generating an initial weight according to the ratio of the dynamic gradient eigenvalue to the quantitative fluctuation eigenvalue, and correcting the initial weight according to the historical deviation ratio of the humidity to obtain a normalized weight coefficient; multiplying the difference between the current element of the temperature error variance matrix and the target threshold by the normalized weight coefficient to obtain a basic deviation; processing the basic deviation based on the instantaneous intensity value in the quantitative fluctuation eigenvalue to obtain the temperature deviation value; The ratio of the current element of the error variance matrix to the target threshold is transformed into a natural logarithm to generate an initial deviation; a compensation factor is generated according to the product of the temperature cumulative amplitude value and the instantaneous humidity fluctuation intensity, and the initial deviation is dynamically compressed by the compensation factor to obtain the humidity deviation value, wherein the three consecutive parameter values ​​in the optimized parameter sequence of the temperature prediction error are subjected to sliding average processing to obtain the temperature error mean; the square deviation is calculated for each parameter value according to the temperature error mean, and the arithmetic average of the three square deviations is used as the current element of the temperature error variance matrix; the same operation is performed on the optimized parameter sequence of the humidity prediction error to generate the current element of the humidity error variance matrix; wherein the absolute value accumulation sum of the optimized parameter sequence of the temperature prediction error in the current sliding window is calculated to characterize the overall deviation degree and duration of the temperature error, forming the temperature cumulative amplitude value; the instantaneous fluctuation peak at the current moment is extracted from the optimized parameter sequence of the humidity prediction error, that is, the maximum absolute value of the difference between adjacent sampling points, which is recorded as the humidity instantaneous fluctuation intensity; Step 3: Perform normalized weighted summation of the temperature deviation and humidity deviation, and dynamically scale them using a Gaussian kernel function to obtain a correction factor; Step 4: dynamically adjust the activation function slope of the hidden layer of the neural network according to the correction factor and the BP neural network, and reconstruct the weight of the fuzzy rule base based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight; In step 5, based on the modified fuzzy rule base weights, a three-dimensional parameter adjustment surface is constructed, and the proportional, integral, and differential parameters of the PID controller are dynamically adjusted through the surface gradient search algorithm to generate a control signal.

2. The anti-interference control method for an intelligent incubation chamber based on error self-learning according to claim 1 is characterized in that: The data of three consecutive sampling periods are used to form a sliding detection window. The temperature and humidity data are fused within the window to output the optimized parameter sequence of temperature prediction error and the optimized parameter sequence of humidity prediction error, including: In the sliding detection window, the temperature and humidity raw data at the current sampling moment and the two moments before it are constructed into time series vectors, each vector containing three consecutive data points; A state transition model is established based on the temperature observations of the previous two cycles to generate the temperature prediction value at the current moment. The actual temperature data collected is subtracted from the prediction value to obtain the temperature prediction error value. The filter gain is dynamically adjusted according to the noise covariance matrix and the observation noise covariance matrix, and the prediction error is weighted and corrected to output the temperature prediction error optimization parameter at the current moment. Repeat the prediction-correction process for the humidity time series vector to generate the corresponding humidity prediction error optimization parameters; The temperature prediction error optimization parameters output from three consecutive sampling periods are arranged in chronological order to form an optimization parameter sequence for the temperature prediction error; the humidity prediction error optimization parameters are arranged to form an optimization parameter sequence for the humidity prediction error.

3. The anti-interference control method for an intelligent incubation chamber based on error self-learning according to claim 2 is characterized in that: The initial weight is generated according to the ratio of the dynamic gradient eigenvalue to the quantitative fluctuation eigenvalue, and the initial weight is corrected by the humidity history deviation ratio to obtain the normalized weight coefficient; Multiply the difference between the current element of the temperature error variance matrix and the target threshold by the normalized weight coefficient to obtain the basic deviation; The basic deviation is processed based on the instantaneous intensity value in the quantized fluctuation characteristic value to obtain the temperature deviation value, including: Calculate the ratio of the dynamic gradient eigenvalue to the quantized fluctuation eigenvalue to obtain a ratio that reflects the relative intensity of the temperature and humidity error changes, which is used as the initial weight. Calculate the ratio of humidity error historical data that deviates from the normal range, that is, the humidity historical deviation ratio, and use the humidity historical deviation ratio to adjust the initial weight to obtain the corrected initial weight; The corrected initial weight is normalized to obtain a normalized weight coefficient; Obtain the current element of the temperature error variance matrix, that is, the average value of the squared deviation of the temperature error in the current sliding window, calculate the difference between the current element and the preset temperature error target threshold, and fuse the difference with the normalized weight coefficient to obtain the basic deviation value that reflects the degree of deviation of the temperature error fluctuation from the target state; The instantaneous intensity value at the current moment, that is, the fluctuation peak value of the current humidity error, is extracted from the quantized fluctuation characteristic value. The basic deviation is adjusted using the instantaneous intensity value to obtain the temperature deviation value reflecting the temperature error and humidity interference.

4. The anti-interference control method for an intelligent incubation chamber based on error self-learning according to claim 3 is characterized in that: The ratio of the current element of the humidity error variance matrix to the target threshold is transformed into a natural logarithm to generate an initial deviation. A compensation factor is generated based on the product of the temperature cumulative amplitude value and the instantaneous humidity fluctuation intensity. The initial deviation is dynamically compressed using the compensation factor to obtain the humidity deviation value, including: Get the current element of the humidity error variance matrix, that is, the average value of the square of the humidity error deviation in the current sliding window, and calculate the humidity ratio of the current element to the preset humidity error target threshold to reflect the deviation multiple of the current humidity error fluctuation relative to the target state; The natural logarithm of the humidity ratio is taken and the numerical range is compressed to the logarithmic scale to obtain the initial deviation reflecting the degree of humidity error deviation; Multiply the temperature cumulative amplitude value by the instantaneous humidity fluctuation intensity to obtain the compensation factor; The initial deviation is adjusted by the compensation factor, and the numerical range of the initial deviation is narrowed by nonlinear mapping to obtain the humidity deviation value that balances the coupling effect of temperature and humidity.

5. The anti-interference control method for an intelligent incubation chamber based on error self-learning according to claim 4 is characterized in that: The temperature deviation and humidity deviation values ​​are normalized and weighted and then dynamically scaled using a Gaussian kernel function to obtain a correction factor, including: Calculating the maximum and minimum values ​​of the temperature deviation value in a preset sampling period respectively, and calculating the first numerical deviation range thereof, that is, the maximum value minus the minimum value; subtracting the minimum value from the current temperature deviation value to obtain a first deviation difference, and calculating the deviation ratio between the first deviation difference and the first numerical deviation range to obtain a normalized temperature deviation value; Calculating the maximum and minimum values ​​of the humidity deviation value in a preset sampling period respectively, and calculating its second numerical deviation range, that is, the maximum value minus the minimum value; subtracting the minimum value from the current humidity deviation value to obtain a second deviation difference, and calculating the deviation ratio between the second deviation difference and the second numerical deviation range to obtain a normalized humidity deviation value; Preset the temperature influence weight coefficient and humidity influence weight coefficient, multiply the normalized temperature deviation value by the temperature influence weight coefficient to obtain the temperature contribution; multiply the normalized humidity deviation value by the humidity influence weight coefficient to obtain the humidity contribution; Add the temperature contribution and the humidity contribution to obtain the temperature and humidity deviation index value; substitute the temperature and humidity deviation index value into the Gaussian kernel function and calculate its function value, which is the correction factor.

6. The anti-interference control method for an intelligent incubation chamber based on error self-learning according to claim 5, characterized in that: According to the correction factor and the BP neural network, the activation function slope of the hidden layer of the neural network is dynamically adjusted, and the weight of the fuzzy rule base is reconstructed based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight, including: Determine the initial slope parameter corresponding to the activation function of the hidden layer of the BP neural network; Determine the adjustment amount of the slope of the activation function according to the value of the correction factor to obtain the adjusted value of the slope; Add the adjusted slope value to the initial slope parameter to obtain the adjusted activation function slope; Analyze the numerical distribution characteristics of the correction factor and calculate the numerical distribution characteristics, which include the mean, variance, maximum value and minimum value of the correction factor; The initial weight of the fuzzy rule base is determined, and the weight of each rule in the fuzzy rule base is adjusted according to the numerical distribution characteristics of the correction factor to obtain the corrected weight of the fuzzy rule base.

7. The anti-interference control method for an intelligent incubation chamber based on error self-learning according to claim 6, characterized in that: Based on the modified fuzzy rule base weights, a three-dimensional parameter adjustment surface is constructed. The proportional, integral, and differential parameters of the PID controller are dynamically adjusted through the surface gradient search algorithm to generate a control signal, including: Determine the three variables of the three-dimensional parameter adjustment surface, which are temperature deviation value, humidity deviation value and the weight of the modified fuzzy rule base; Under different combinations of temperature deviation values, humidity deviation values ​​and modified fuzzy rule base weights, the corresponding proportional, integral and differential parameter values ​​of the PID controller are recorded; According to the values ​​of the proportional, integral and differential parameters of the PID controller, a three-dimensional parameter adjustment surface is constructed through neural network fitting; According to the current temperature deviation value, humidity deviation value and the weight of the modified fuzzy rule base in the incubator, the corresponding state point is found on the three-dimensional parameter adjustment surface. The state point represents the initial parameter setting of the PID controller under the current working condition. At the current state point, the gradient of the three-dimensional parameter adjustment surface is calculated. The gradient represents the direction of change of the three-dimensional parameter adjustment surface at the state point. It contains three components, which correspond to the change directions of the proportional, integral and differential parameters of the PID controller respectively. The gradient and search step of the surface are adjusted according to the three-dimensional parameters, and the proportional, integral and differential parameters of the PID controller are adjusted in the opposite direction of the gradient until the search step is reached to obtain the optimized proportional, integral and differential parameters of the PID controller; The PID controller calculates the corresponding control signal based on the current temperature deviation value and humidity deviation value, combined with the adjusted proportional, integral and differential parameters.

8. The intelligent incubation warehouse anti-interference control system based on error self-learning is characterized by: The system is used to perform the method according to any one of claims 1 to 7, comprising: The acquisition module is used to collect the temperature and humidity data stream of the chicken embryo incubation in the incubation chamber in real time; the data of three consecutive sampling cycles are used to form a sliding detection window, and the temperature and humidity data are fused within the window to output the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error; A calculation module, used to calculate the temperature deviation value and the humidity deviation value respectively according to the optimized parameter sequence of the temperature prediction error and the optimized parameter sequence of the humidity prediction error; A correction module is used to perform normalized weighted summation of the temperature deviation value and the humidity deviation value, and dynamically scale them using a Gaussian kernel function to obtain a correction factor; An adjustment module is used to dynamically adjust the activation function slope of the hidden layer of the neural network according to the correction factor and the BP neural network, and reconstruct the weight of the fuzzy rule base based on the numerical distribution characteristics of the correction factor to generate the corrected fuzzy rule base weight; The processing module is used to construct a three-dimensional parameter adjustment surface based on the modified fuzzy rule base weights, and dynamically adjust the proportional, integral and differential parameters of the PID controller through the surface gradient search algorithm to generate a control signal.

Images