Search

CN-121998558-A - Early warning analysis method for microbial pollution risk of stored silkworm feed

CN121998558ACN 121998558 ACN121998558 ACN 121998558ACN-121998558-A

Abstract

The invention relates to a microbial contamination early warning technology based on a multi-mode sensing and intelligent model in a storage environment. Aiming at the problems of delayed detection and low early warning accuracy of microbial contamination in the existing silkworm group production, the accurate alignment and purification of multi-mode monitoring data are realized through highly coordinated time synchronization and noise filtering, and the biological behavior and physiological abnormal characteristics of the silkworm body are further extracted from the multi-mode monitoring data. Based on a graph attention network and LSTM-GCN mixed model, the microecological stress reaction and the microbial growth situation of the silkworm group are deduced, and a dynamic threshold and self-adaptive weight mechanism are introduced, so that real-time assessment and early warning of pollution risks are realized. The method effectively improves the sensitivity and accuracy of pollution early warning in a warehouse scene, has self-closing loop adjustment and real-time response capability, and improves the health guarantee level of silkworms.

Inventors

  • LIN JINWEI
  • ZHANG YAPING
  • LIN JINTAO
  • ZHANG YUEHUA
  • PAN ZHAOSHENG

Assignees

  • 深圳同益新中控实业有限公司

Dates

Publication Date
20260508
Application Date
20260410

Claims (9)

  1. 1. A method for early warning and analyzing the microbial pollution risk of stored silkworm feed specifically comprises the following steps: s1, acquiring millimeter wave radar signals and environmental parameters at a plurality of different positions in a storage unit, and uniformly defining acquisition results as original monitoring data; s2, performing cross-mode time sequence alignment and noise filtering processing on the original monitoring data to eliminate signal interference and obtain preprocessed monitoring data; S3, extracting the decay rate of the feeding rhythm and the body surface micro-temperature low-frequency amplitude based on the preprocessing monitoring data, and uniformly defining the extraction result as a silkworm body biological characteristic variable; s4, inputting the silkworm body biological characteristic variable into a lightweight graph attention network for mapping calculation to obtain silkworm body stress response intensity; S5, based on the timestamp synchronous environment parameters of the stress response intensity of the silkworm body, deducing a microorganism growth situation by using an LSTM-GCN mixed model, and generating a microorganism growth potential index; S6, performing weighted summation operation based on the microorganism growth potential index and the silkworm stress response intensity, and generating a dynamic threshold value by combining a change rate compensation factor; s7, comparing and judging the real-time risk value with the dynamic threshold value, and generating a pollution risk early warning signal if the real-time risk value exceeds the threshold value; And S8, carrying out on-line correction on the weight coefficient in the weighted summation operation based on consistency verification of the pollution risk early warning signal and the actual detection result, and generating an updated weight coefficient.
  2. 2. The method for analyzing the risk of microbial contamination of the stored silkworm feed according to claim 1, wherein the step S1 specifically includes: S1.1, deploying millimeter wave radar sensor arrays and environmental parameter sensing nodes on a plurality of preset space coordinate points in a storage unit, and generating radar emission control instructions with uniform time stamps and environmental sampling trigger signals based on a pulse repetition frequency synchronization mechanism so as to establish a space-time reference frame for multi-source data acquisition; S1.2, driving a millimeter wave radar sensor array to emit frequency modulation continuous wave signals to an active area of the silkworm group based on the radar emission control instruction, and receiving millimeter wave radar echo signals formed by reflection on the surface of the silkworm group so as to obtain an original radar data stream containing the silkworm group micro Doppler frequency shift information; S1.3, driving an environmental parameter sensing node to execute multipoint parallel sampling operation based on the environmental sampling trigger signal, and capturing temperature field distribution data, humidity field distribution data, carbon dioxide concentration data and ammonia concentration data in a storage space in real time to generate a multidimensional environmental parameter data set; s1.4, performing frame-level packaging processing based on uniform time stamps on the original radar data stream and the multidimensional environment parameter data set, and integrating the heterogeneous signal stream into a data packet with a fixed frame structure by utilizing data packaging so as to form a standardized original monitoring data sequence; S1.5, executing integrity check and abnormal value marking processing based on the standardized original monitoring data sequence, removing invalid data fragments generated by instantaneous faults of the sensor through sliding window statistical check, and finally outputting original monitoring data for confirmation to be used for calling by a subsequent preprocessing module.
  3. 3. The method for analyzing the risk of microbial contamination of the stored silkworm feed according to claim 1, wherein the step S2 specifically includes: S2.1, performing linear interpolation resampling processing on a millimeter wave Lei Dadian cloud sequence and an environmental sensor timestamp in original monitoring data to unify the sampling frequency of multi-source heterogeneous data and generate a synchronous monitoring data stream with a unified time reference; s2.2, calculating the peak position of a cross-correlation function based on the synchronous monitoring data stream so as to quantify the transmission delay difference between the millimeter wave radar signal and the environment parameter signal and generate a time sequence offset correction vector containing an accurate time delay compensation quantity; S2.3, performing sliding window translation operation on the synchronous monitoring data stream by using the time sequence offset correction vector so as to eliminate the phase asynchronous phenomenon among the multi-mode sensors and generate a cross-mode alignment data set with strictly aligned time axes; s2.4, applying a self-adaptive wavelet threshold denoising algorithm to the cross-modal alignment data set to separate and filter high-frequency electromagnetic interference and mechanical vibration noise in a storage environment and generate a purified monitoring signal matrix with obviously improved signal-to-noise ratio; S2.5, performing outlier rejection and missing data linear filling processing based on the purification monitoring signal matrix so as to repair data break points caused by instantaneous faults of the sensor, and finally outputting complete and continuous preprocessing monitoring data which meet the analysis precision requirement.
  4. 4. The method for analyzing the risk of microbial contamination of the stored silkworm feed according to claim 1, wherein the step S3 specifically includes: S3.1, performing short-time Fourier transform processing on the millimeter wave radar echo signal sequence in the preprocessing monitoring data to obtain a time-frequency energy distribution map containing silkworm group micro Doppler frequency shift information, wherein the time-frequency energy distribution map is used as a basic data source for identifying feeding action frequency change; s3.2, executing a peak detection and track association algorithm based on the time-frequency energy distribution map so as to extract an effective feeding action counting sequence in a unit time window, and calculating a slope change value of the counting sequence along with time through linear regression fitting so as to generate a feeding rhythm attenuation rate representing the descending trend of the feeding activity of the silkworm group; S3.3, band-pass filtering is carried out on the infrared thermal imaging temperature field data flow in the preprocessing monitoring data, and a Butterworth filter with the cut-off frequency of zero one hertz to zero one hertz is utilized to separate high-frequency environmental noise and low-frequency physiological fluctuation so as to obtain a pure silkworm body surface micro-temperature low-frequency fluctuation signal sequence; S3.4, performing Hilbert-Huang transformation and decomposition processing based on the silkworm body surface micro-temperature low-frequency fluctuation signal sequence, so as to extract an instantaneous amplitude envelope corresponding to a stress reaction frequency band in the eigenmode function components, and calculating root mean square statistic of the envelope in a sliding window, thereby generating body surface micro-temperature low-frequency amplitude representing the metabolic disturbance degree of the silkworm body; and S3.5, performing space-time coordinate binding and format packaging processing on the feeding rhythm attenuation rate and the body surface micro-temperature low-frequency amplitude, integrating the two types of heterogeneous scalar and vector data into a structured data object by using a unified timestamp, and finally outputting a standard data set defined as a silkworm body biological characteristic variable for the subsequent graph and meaning network to call.
  5. 5. The method for analyzing the risk of microbial contamination of the stored silkworm feed according to claim 1, wherein the step S4 specifically includes: S4.1, constructing isomerism map data containing space position information and time sequence characteristics based on the feeding rhythm attenuation rate and the body surface micro-temperature low-frequency amplitude in the silkworm body biological characteristic variables, defining each millimeter wave radar sensor node in a storage unit as a map node, defining the space adjacent relation and the signal correlation between the nodes as map edges, and generating an initial topological map structure for representing silkworm group micro-ecological disturbance distribution; S4.2, performing node characteristic embedding processing on the initial topological graph structure, and mapping a feeding rhythm attenuation rate scalar value and a body surface micro-temperature low-frequency amplitude spectrum vector to Gao Weiyin space by using a linear transformation matrix to generate a node characteristic embedding vector carrying multi-dimensional biological behavior semantic information so as to eliminate interference of different physical dimensions on subsequent attention mechanism calculation; S4.3, based on the node characteristic embedded vector, executing multi-head self-attention mechanism calculation, quantifying the association weight of biological behavior characteristics among adjacent nodes through dot product operation of query vectors, key vectors and value vectors, and generating a dynamic attention coefficient matrix reflecting the abnormal propagation path of local silkworm groups so as to identify a group stress response mode caused by microbial contamination; s4.4, carrying out weighted aggregation operation on feature embedded vectors of neighbor nodes by utilizing the dynamic attention coefficient matrix, and fusing original features of the current nodes and neighborhood context information by combining a residual error connection mechanism to generate updated node hiding state representation containing global space dependency relationship so as to enhance the capturing capability of the model on weak pollution signals; and S4.5, performing full-connection layer mapping and normalization processing based on the updated node hidden state representation, compressing the high-dimensional hidden state into a single-dimensional scalar output, and generating a final evaluation value representing the whole silkworm stress response intensity of the storage unit at the current moment as a biofeedback correction factor in a subsequent dynamic threshold adjustment mechanism.
  6. 6. The method for early warning and analyzing the microbial contamination risk of the stored silkworm feed according to claim 5, wherein the three-dimensional space accurate positioning is carried out on the sensor deployment by adopting a coordinate mapping algorithm, a pulse repetition frequency synchronization mechanism is adopted, 10Hz is used as a PRF frequency, millimeter wave radar signals work in a frequency band of 76 GHz-81 GHz, the modulation period is 100ms, the high-precision analysis on silkworm body inching is realized by adopting fast Fourier transform and Doppler frequency shift extraction, the sampling period of environmental parameters is 2 seconds, and the number of parallel threads is equal to the number of sensor nodes.
  7. 7. The method for analyzing the microbial contamination risk of the stored silkworm feed according to claim 3, wherein cross-correlation function analysis is performed, the maximum hysteresis range is + -120 seconds, the cross-correlation peak value normalization interval is [ -1,1], and sensing signal delay compensation is performed based on the peak value position, short-time Fourier transform is adopted, the window type is hanning window, the window length is 256 points, the frame shift is 128 points, and the FFT point number is 512.
  8. 8. The method for early warning and analyzing the microbial contamination risk of the stored silkworm feed according to claim 6 is characterized in that millimeter wave radar nodes are bound with biological characteristic variables through space mapping and time sequence fusion, a pearson correlation coefficient threshold value is adopted for node signal correlation to determine weighted edges in a screening mode, the environmental state diagram structure is formed by establishing node adjacency by a Euclidean distance threshold value of 2 meters, weighted average of temperature and humidity, carbon dioxide concentration differences and ammonia concentration differences is obtained by the edges, and a multi-head attention mechanism is adopted.
  9. 9. The method for early warning and analyzing the microbial contamination risk of the stored silkworm feed according to claim 1, wherein the basic risk reference value is obtained by weighting the current microbial growth potential index and the silkworm stress response intensity by the on-line reinforcement learning weights alpha and beta, the interpretability analysis adopts the SHAP sample number of 1000, the L2 norm is normalized, the upper limit and the lower limit of the dynamic threshold boundary constraint interval are dynamically adjusted by the multivariate statistical analysis of the stored historical data according to time windows and environmental gain coefficients, and the final dynamic threshold is output.

Description

Early warning analysis method for microbial pollution risk of stored silkworm feed Technical Field The invention relates to the technical field of early warning of microbial contamination risk of silkworm feed storage, in particular to an early warning analysis method of microbial contamination risk of storage silkworm feed. Background Currently, the field of early warning of the microbial pollution risk of silkworm feed storage mainly depends on a static threshold method driven by environmental parameters to judge the pollution risk. In the conventional risk early warning system, environmental parameters such as temperature, humidity, carbon dioxide, ammonia gas and the like, and physicochemical indexes such as feed water activity, pH and the like are acquired through an Internet of things sensor in the warehouse monitoring process, and a fixed threshold or interval is set as a pollution risk judging standard according to statistical experience. When the detected parameters exceed the set range, the method is judged to be potential microbial contamination risk and gives out early warning, and the method has the advantages of simplicity and convenience in implementation and high reasoning speed, so that the method is widely applied to storage safety monitoring scenes of large amounts of materials such as grains, pastures, mushrooms and the like. Along with the development of technologies such as big data analysis and artificial intelligence, new paths such as multi-source parameter weighted fusion, multi-model integrated prediction, knowledge-graph rule reasoning, transfer learning threshold self-adaption and the like are explored in the industry so as to improve the perception capability of an early warning system on abnormal fine granularity fluctuation. For example, the pollution risk generalization assessment under different storage environments, feed categories and season periods is realized by constructing a multi-factor weight model, or a statistical mean value/standard deviation self-adaptive sliding adjustment threshold value is adopted to cope with false alarm or missing alarm caused by external disturbance to a certain extent. Although the optimization methods have a certain practical effect in the specific field, the overall optimization method still follows the static parameter experience or data-driven unidirectional modeling thought, and omits a key feed-microorganism-silkworm three-element biofeedback link in a warehouse feed system. Practical application shows that the dynamic evolution process of microbial contamination is influenced by multi-factor nonlinear coupling, and single environment or physical and chemical parameters are used as early warning threshold judgment standards, so that high-precision and low-delay capturing of risk events in complex scenes is difficult to realize. The method has the following outstanding problems and technical bottlenecks: The fixed threshold model is difficult to adapt to dynamic change situations such as different storage environments, season alternation, feed batch heterogeneity and the like. When external climate, in-warehouse layout, batch raw material state and the like are changed, the original static threshold value often leads to remarkable increase of risk misjudgment rate, and the increase of misinformation rate and exacerbation of missing information rate are shown, so that the practicability and generalization capability of the early warning system are severely restricted. The prior art route fails to effectively exploit the unique advantages of silkworm bodies as high sensitivity "biological indicators". The traditional method only focuses on environmental parameters or feed physical and chemical characteristics, ignores the sensitive physiological behavior response of silkworm groups to microbial pollution, such as feeding activity attenuation, group disorder, body surface stress temperature change and the like, and the feedback signals and the pollution risk have direct time sequence causal relationship. Although the threshold value can be dynamically corrected to a certain extent based on sliding adjustment of statistical mean value, standard deviation, quantile and the like or multi-source data weighted optimization, the threshold value is still driven by environmental data, the criterion of truly fusing biofeedback is lacking, response is lagged when sudden and hidden pollution events are encountered, and the judgment reliability is poor. Most systems do not establish a closed loop feedback mechanism of risk judgment-biofeedback-model correction, and cannot realize the self-adaptive online evolution of risk threshold values. The threshold parameters are not regulated along with the actual performance of the scene for a long time, the stability and the long-term precision of the system are obviously reduced, and the system is difficult to quickly self-optimize especially in high-risk and low-morbidity scenes. Therefore, a technical pa