Unprocessed grain static storage online detection system based on multi-sensor fusion

Abstract

The invention relates to the technical field of unprocessed grain storage management, in particular to an unprocessed grain static storage online detection system based on multi-sensor fusion, and the system comprises the steps: collecting the purchased grain source initial detection data of a grain pile, and synchronously collecting the chemical gas data and environmental state data of the grain pile through a multi-source sensor network disposed in a granary; performing traceability analysis on heterogeneity risks of grain piles in combination with initial inspection data of purchased grain sources and environmental state data; performing dynamic simulation analysis on the respiratory heat accumulation risk of the grain pile in combination with chemical gas data and environmental state data of the grain pile; fusing the heterogeneity risk traceability analysis result and the respiratory heat accumulation risk dynamic simulation analysis result of the grain pile, and evaluating the grain storage risk of the stored unprocessed grains; according to the stored grain storage risk assessment result of the stored grain, making a disposal decision on the stored grain; therefore, the guarantee level and the management efficiency of grain storage safety are improved, the mildewing and corruption risk is reduced from the source, and the quality deterioration of the unprocessed grains in the storage process is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of raw grain storage management technology, and in particular to an online detection system for static storage of raw grain based on multi-sensor fusion. Background Technology

[0002] Static grain storage, as a strategic link in ensuring national food security and regulating market supply and demand, directly impacts the stability of grain quality, the effective preservation of grain resources, the safety of subsequent processed food, and its immense economic value. Grain faces two core threats during long-term storage: First, the heterogeneity of grain pile quality due to diverse sourcing sources, with initial differences in moisture content, impurities, and other indicators between different batches of grain, can become potential starting points for mold growth and spread under suitable temperature and humidity conditions, progressing from localized spot mold to regional spoilage. Second, the continuous respiration of grain as a living organism generates heat. In statically stacked, poorly ventilated storage structures, heat easily accumulates and is difficult to dissipate, leading to abnormally high local temperatures, thus accelerating grain aging and sugar loss. Failure to promptly detect the coupled evolution of heterogeneous hotspots and respiration heat accumulation can trigger serious incidents such as localized mold and spoilage, toxin accumulation, or large-scale condensation and heating in grain piles, causing not only significant economic losses but also jeopardizing national food quality and security.

[0003] However, current technologies for online monitoring of raw grain storage typically analyze various sensor data in isolation, failing to deeply integrate initial inspection data characterizing the grain's inherent properties with real-time storage environmental data to trace the source of quality heterogeneity risks. For example, when an increase in mold marker gas concentration is detected in a certain area, without considering the historical initial quality of the grain in that area and specific buyer information, it is difficult to accurately determine whether the mold risk stems from a problem with the grain itself or from a local environmental malfunction. This results in a lack of targeted control measures, failing to curb the risk at its source. Furthermore, current technologies lack dynamic simulation and prediction of the dynamic biological process of grain respiration and the mechanism of heat accumulation risk formation. They often only issue threshold alarms based on current temperature and humidity, without considering the temporal impact of factors such as respiration rate, storage time, and environmental interaction on heat accumulation. This makes it difficult to accurately predict the formation and evolution trend of heat accumulation risks.

[0004] To address these issues, this application presents an online monitoring system for static grain storage based on multi-sensor fusion. Summary of the Invention

[0005] The purpose of this invention is to provide an online detection system for static grain storage based on multi-sensor fusion. By integrating multi-source data such as initial inspection of purchased grain, environmental conditions, and chemical gases, a random forest model is constructed to trace the source of heterogeneous mold risk in grain piles, and an LSTM model is constructed to dynamically simulate and predict the risk of respiratory heat accumulation. Then, the two types of risks are weighted and fused to achieve a comprehensive assessment, realizing early warning and precise prevention and control of grain storage risks.

[0006] This invention is implemented as follows: In a first aspect, the present invention provides an online detection system for static storage of raw grain based on multi-sensor fusion, including a multi-source acquisition module, a grain source heterogeneity detection module, a grain accumulation heat detection module, a storage risk assessment module, and a disposal decision module; Among them, the multi-source acquisition module is used to collect the initial inspection data of the purchased grain pile, and through the multi-source sensor network deployed in the grain warehouse, it simultaneously collects the chemical gas data and environmental status data of the grain pile. The grain source heterogeneity detection module is used to combine the initial inspection data of the purchased grain source with environmental status data to conduct source tracing analysis of the heterogeneity risk of the grain pile; The grain accumulation heat detection module is used to dynamically simulate and analyze the risk of respiratory heat accumulation in grain piles by combining chemical gas data and environmental condition data. The storage risk assessment module is used to integrate the results of heterogeneity risk tracing analysis of grain piles and the results of dynamic simulation analysis of respiratory heat accumulation risk to assess the storage risk of stored raw grains. The disposal decision module is used to make disposal decisions for stored raw grains based on the results of the grain storage risk assessment.

[0007] As a preferred embodiment of the present invention, the heterogeneity risk of grain piles is traced and analyzed by combining the initial inspection data of the purchased grain source with environmental status data, including the following specific steps: S21. Obtain the initial inspection data of the purchased grain source corresponding to each monitoring point inside the grain warehouse from the warehouse management information system, including the initial moisture content, impurity content, imperfect grain rate, and the purchaser identification with unique thermal coding; obtain the environmental status data of each monitoring point within the same time window from the multi-source sensor network deployed in the grain warehouse, including temperature and relative humidity. S22. Based on the length of the time window in the current monitoring period, divide the historical time window and obtain the initial moisture content, impurity content, imperfect grain rate, buyer identification with unique thermal coding, temperature, relative humidity, and the concentration of mold marker gas measured by the gas sensor at each monitoring point inside the grain warehouse in multiple historical monitoring periods as a heterogeneity risk prediction dataset. Divide the heterogeneity risk prediction dataset into a heterogeneity risk prediction training set and a heterogeneity risk prediction validation set. S23. Construct a random forest model. Based on the heterogeneous risk prediction training set, use the initial moisture content, impurity content, imperfect grain rate, uniquely thermally encoded purchaser identifier, temperature, and relative humidity of each monitoring point inside the grain warehouse in each historical monitoring period as the input features of the random forest model. Use the concentration of mold marker gas at each monitoring point in the heterogeneous risk prediction training set in the corresponding historical monitoring period as the output target of the random forest model. Train the random forest model to obtain the initial heterogeneous risk prediction model. Validate the initial heterogeneous risk prediction model through the heterogeneous risk prediction validation set. The initial heterogeneous risk prediction model with an accuracy greater than or equal to the preset first model is taken as the heterogeneous risk prediction model.

[0008] As a preferred embodiment of the present invention, the heterogeneity risk of grain piles is traced and analyzed by combining the initial inspection data of the purchased grain source with environmental status data, and specifically includes the following steps: S24. Input the initial moisture content, impurity content, imperfect particle rate, buyer identification with unique thermal coding, temperature and relative humidity of each monitoring point in the current monitoring cycle into the heterogeneity risk prediction model, and output the predicted concentration of mold marker gas at each monitoring point. S25. The monitoring points with predicted mold marker gas concentrations greater than the mold marker gas concentration threshold are identified as quality abnormality monitoring points, and the monitoring points with predicted mold marker gas concentrations less than or equal to the mold marker gas concentration threshold are identified as quality normal monitoring points. All monitoring points identified as quality abnormality monitoring points are statistically clustered based on the corresponding spatial location and the associated purchaser identifiers. The grain source corresponding to the purchaser identifier with the highest frequency is identified as the traceability direction of the batch's storage heterogeneity risk within the current monitoring period. S26. Obtain the variation range of predicted mold characteristic gas concentrations at all monitoring points, calculate the difference between the average and minimum values ​​of predicted mold characteristic gas concentrations at all monitoring points, and use the ratio of the difference to the variation range of predicted mold characteristic gas concentrations at all monitoring points as the heterogeneity risk of the grain pile in the current monitoring period.

[0009] As a preferred embodiment of the present invention, a dynamic simulation analysis of the respiratory heat accumulation risk of the grain pile is performed by combining chemical gas data and environmental condition data of the grain pile, including the following specific steps: S31. Obtain chemical gas data and environmental status data of each monitoring point inside the grain warehouse from the multi-source sensor network deployed inside the grain warehouse. The chemical gas data is carbon dioxide concentration, and the environmental status data includes temperature and relative humidity. Calculate the grain respiration rate at each monitoring point based on the rate of change of carbon dioxide concentration per unit time. S32. Obtain the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days the raw grain has been stored at each monitoring point in multiple historical monitoring periods, as well as the actual temperature deviation calculated after the end of the corresponding historical monitoring period, as a respiration heat accumulation risk prediction dataset; divide the respiration heat accumulation risk prediction dataset into a respiration heat accumulation risk prediction training set and a respiration heat accumulation risk prediction validation set; wherein, the actual temperature deviation is the difference between the average temperature of the monitoring point and the average of the average temperatures of all monitoring points in the monitoring period; S33. Construct a long short-term memory network model and train and validate it based on the respiratory heat accumulation risk prediction training set. Use the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days of raw grain storage at each monitoring point in each historical monitoring period as input features of the long short-term memory network model. Use the actual temperature deviation after the end of the historical monitoring period corresponding to the input features as the output target of the long short-term memory network model to train the model and obtain the initial respiratory heat accumulation risk prediction model. Validate the initial respiratory heat accumulation risk prediction model through the respiratory heat accumulation risk prediction validation set. The model with an accuracy greater than or equal to the preset model is used as the final respiratory heat accumulation risk prediction model. S34. Input the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days of raw grain storage at each monitoring point within the current monitoring period into the trained respiration heat accumulation risk prediction model, and output the predicted value of temperature deviation at each monitoring point at the end of the current monitoring period.

[0010] As a preferred embodiment of the present invention, the dynamic simulation analysis of the respiration heat accumulation risk of the grain pile, combining chemical gas data and environmental condition data, further includes the following specific steps: S35. The monitoring points with predicted temperature deviation values ​​greater than the temperature deviation threshold are identified as abnormal heat accumulation monitoring points, and the monitoring points with predicted temperature deviation values ​​less than or equal to the temperature deviation threshold are identified as normal heat accumulation monitoring points. S36. Obtain the variation range of the predicted temperature deviation values ​​of all monitoring points, calculate the difference between the average and minimum values ​​of the predicted temperature deviation values ​​of all monitoring points, and use the ratio of the difference to the variation range of the predicted temperature deviation values ​​of all monitoring points as the risk of respiratory heat accumulation in the grain pile during the current monitoring period.

[0011] As a preferred embodiment of the present invention, the storage risk of stored grain is assessed by integrating the results of heterogeneity risk tracing analysis of grain piles and the results of dynamic simulation analysis of respiratory heat accumulation risk, including the following specific steps: S41. Extract the heterogeneity risk and respiratory heat accumulation risk of the grain pile within the current monitoring period; S42. The storage risk of the grain pile is obtained by weighted summation of the heterogeneity risk and the respiratory heat accumulation risk within the current monitoring period.

[0012] In a preferred embodiment of the present invention, a disposal decision is made on the stored raw grain based on the results of the grain storage risk assessment, including the following specific contents: S51. Obtain the storage risk of raw grain in the current monitoring period, and extract the source of grain and heat accumulation anomaly monitoring point for the traceability of heterogeneity risk of this batch of stored grain in the current monitoring period. S52. Preset grain storage risk threshold. When the grain storage risk of the stored raw grain in the current monitoring period is greater than the grain storage risk threshold, an emergency release procedure will be executed for the grain source with traceability direction of the heterogeneity risk of this batch in the current monitoring period, and an emergency quality inspection will be carried out on the grain pile at the heat accumulation anomaly monitoring point. When the grain storage risk of the stored raw grain in the current monitoring period is less than or equal to the grain storage risk threshold, a priority release procedure will be executed for the grain source with traceability direction of the heterogeneity risk of this batch in the current monitoring period, and a random inspection will be carried out on the grain pile at the heat accumulation anomaly monitoring point.

[0013] Secondly, the present invention provides an online detection method for static storage of raw grains based on multi-sensor fusion, comprising the following specific steps: The system collects preliminary inspection data of the purchased grain from the grain piles and simultaneously collects chemical gas data and environmental status data of the grain piles through a multi-source sensor network deployed in the grain warehouse. By combining preliminary inspection data of the acquired grain sources with environmental status data, a source tracing analysis was conducted on the heterogeneity risk of the grain piles. By combining chemical gas data and environmental condition data of grain piles, a dynamic simulation analysis of the risk of respiratory heat accumulation in grain piles is conducted. By integrating the results of heterogeneity risk tracing analysis of grain piles and dynamic simulation analysis of respiratory heat accumulation risk, the storage risk of stored raw grains is assessed. Based on the risk assessment results of the stored grain, a decision is made on the disposal of the stored grain.

[0014] Thirdly, the present invention provides an electronic device, comprising: a processor and a memory, wherein the memory stores a computer program that can be called by the processor, and the processor executes an online detection method for static storage of raw grain based on multi-sensor fusion by calling the computer program stored in the memory.

[0015] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. This invention integrates preliminary inspection data of purchased grain sources with real-time environmental data to construct a predictive model for quantitative analysis and source tracing of heterogeneous mold risk in grain piles. This significantly improves the early warning capability for mold risk and the accuracy of prevention and control measures, thereby reducing the risk of systemic mold and spoilage from the source. 2. This invention utilizes an LSTM network to dynamically simulate the temporal relationship between multiple factors such as grain respiration rate, environmental and temporal characteristics, and heat accumulation, thereby achieving dynamic prediction and spatial location of the risk of heat accumulation during grain respiration, effectively avoiding quality deterioration caused by local temperature rise. 3. This invention forms a unified comprehensive assessment index for grain storage risk by weighted integration of heterogeneity risk and respiratory heat accumulation risk, and triggers a graded disposal decision-making mechanism accordingly, thereby realizing the optimized allocation of storage management resources and intelligent closed-loop risk response, and comprehensively improving the level of grain storage safety and management efficiency. Attached Figure Description

[0016] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the overall process of the online detection method for static storage of raw grain based on multi-sensor fusion according to the present invention; Figure 2 This is a schematic diagram of the structure of the online detection system for static storage of raw grain based on multi-sensor fusion according to the present invention; Figure 3 This is an analytical flowchart of step S2 in the online detection method for static storage of raw grain based on multi-sensor fusion of the present invention; Figure 4 This is an analytical flowchart of step S3 in the online detection method for static storage of raw grain based on multi-sensor fusion of the present invention; Figure 5 This is a schematic diagram of the grain storage structure for the online detection method of static grain storage based on multi-sensor fusion according to the present invention. Figure 6 This is an analysis flowchart of the respiratory heat accumulation risk prediction model of the online detection method for static storage of raw grain based on multi-sensor fusion according to the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0018] Example 1 like Figure 1 As shown, this embodiment provides an online detection method for static storage of raw grains based on multi-sensor fusion, including the following specific steps: S1. Collect preliminary inspection data of the purchased grain piles, and simultaneously collect chemical gas data and environmental status data of the grain piles through a multi-source sensor network deployed in the grain warehouse. S2. Combine the initial inspection data of the acquired grain source with the environmental status data to conduct a source tracing analysis of the heterogeneity risk of the grain pile; S3. Combining chemical gas data and environmental status data of grain piles, conduct dynamic simulation analysis on the risk of respiratory heat accumulation in grain piles. S4. By integrating the results of the heterogeneity risk tracing analysis of grain piles and the results of the dynamic simulation analysis of respiratory heat accumulation risk, the storage risk of stored raw grains is assessed. S5. Based on the risk assessment results of the stored grain, make disposal decisions for the stored grain.

[0019] In this embodiment, as Figure 3 As shown, step S2 combines the initial inspection data of the acquired grain source with environmental status data to conduct a source tracing analysis of the heterogeneity risk of the grain pile, including the following specific steps: S21. Obtain information from the warehouse management information system, such as... Figure 5The data shown represents the initial inspection data of the purchased grain source at each monitoring point inside the grain warehouse, including initial moisture content, impurity content, imperfect grain rate, and the uniquely thermally coded identifier of the purchaser. Environmental status data, including temperature and relative humidity, for each monitoring point within the same time window are acquired from a multi-source sensor network deployed within the grain warehouse. It should be noted that the initial inspection data of the purchased grain source corresponding to each pre-set monitoring point inside the grain warehouse is systematically obtained from the warehouse management information system. This initial inspection data is crucial for assessing the basic quality of the grain source and tracing the source of heterogeneity. Specifically, it includes: initial moisture content, impurity content, and imperfect grain rate measured and recorded by standardized inspection instruments upon purchase and storage. Furthermore, to achieve accurate mapping between risk and source in subsequent analysis, this embodiment assigns a unique identifier to each independent purchaser and processes it using uniquely thermally coded technology. For example, if there are three buyers, A, B, and C, then "A" is coded as [1, 0, 0], "B" as [0, 1, 0], and "C" as [0, 0, 1]. This transforms the categorical data into numerical features suitable for machine learning models and avoids introducing misunderstandings related to size order. Simultaneously, from a multi-source sensor network pre-deployed three-dimensionally within the grain pile, this embodiment synchronously collects environmental status data corresponding to the aforementioned monitoring point locations within the same time window. This sensor network consists of numerous high-precision digital temperature and humidity sensor nodes embedded in the grain pile in a grid pattern. Through low-power wireless IoT technology, it collects and uploads temperature and relative humidity data from the monitoring points according to a unified timestamp. Based on the above, this embodiment constructs a spatiotemporal benchmark framework for risk analysis, precisely binding the static, pre-determined grain source identity file with the dynamic, real-time changing storage environment through spatial location and temporal synchronization. This solves the problem of the disconnect between grain condition data and grain source information in traditional storage management, enabling any subsequent environmental anomalies to be directly linked back to a specific grain batch and buyer.

[0020] S22. Based on the time window length within the current monitoring period, historical time windows are divided to obtain the initial moisture content, impurity content, imperfect grain rate, uniquely thermally coded buyer identifier, temperature, relative humidity, and mold marker gas concentration measured by a gas sensor at various monitoring points inside the grain warehouse over multiple historical monitoring periods. This data serves as a heterogeneity risk prediction dataset, which is then divided into a heterogeneity risk prediction training set and a heterogeneity risk prediction validation set. In this embodiment, the data is measured by a high-sensitivity gas sensor (e.g., a photoionization detector or a metal-oxide-semiconductor sensor array) at each monitoring point. The concentration of mold biomarkers at the monitoring points; common mold biomarkers include volatile organic compounds from microbial metabolism such as 3-methyl-1-butanol and decenol. The process for acquiring their concentration data is as follows: a sensor probe is embedded in the grain pile, and its sensing element comes into contact with the gas between the grain particles. When biomarker molecules are adsorbed onto the surface of the sensing material, they cause specific changes in conductivity or ion current. This analog signal is preliminarily processed by analog-to-digital conversion and a built-in algorithm to output a digital concentration value, which is then transmitted to the data center via the Internet of Things. All the above records from different monitoring points and different historical periods are aggregated into a structured heterogeneous risk prediction dataset.

[0021] S23. Construct a random forest model based on a heterogeneous risk prediction training set. Use the initial moisture content, impurity content, imperfect grain rate, uniquely thermally encoded buyer identifier, temperature, and relative humidity of each monitoring point inside the grain warehouse within each historical monitoring period as input features of the random forest model. Use the concentration of mold marker gas at each monitoring point in the corresponding historical monitoring period within the heterogeneous risk prediction training set as the output target of the random forest model. Train the random forest model to obtain an initial heterogeneous risk prediction model. Validate the initial heterogeneous risk prediction model using a heterogeneous risk prediction validation set. The output should be greater than or equal to the preset first model accuracy. The initial heterogeneous risk prediction model is used as the heterogeneous risk prediction model. It should be noted that random forest is an ensemble learning algorithm that improves prediction accuracy and stability by constructing a large number of independent decision trees and voting or averaging them. It is very suitable for handling high-dimensional, nonlinear data with categorical features in the scenario involved in this embodiment. In this embodiment, the specific construction and training process of the heterogeneous risk prediction model is as follows: First, the model parameters are initialized, including setting the number of decision trees in the forest, the maximum number of features considered when splitting each tree, the maximum depth of the tree, etc. These hyperparameters can be optimized by combining grid search with cross-validation. The operation is based on a heterogeneous risk prediction training set. For each historical record in the training set, the initial moisture content, impurity content, imperfect particle rate, uniquely thermally encoded buyer identifier, temperature, and relative humidity are used as the input feature vector. The concentration of mold marker gas, actually measured by the sensor, corresponding to that record is set as the output target value that the model needs to learn and approximate. In this embodiment, multiple sample subsets with replacement are extracted from the training set using a bootstrap sampling method to generate slightly different training data for each decision tree. During the growth process of each decision tree, based on criteria such as Gini impurity or information gain, the above six... The optimal split point is recursively selected from the features to generate a series of prediction rules from features to target concentrations. This solidifies the complex patterns hidden in historical data into a reusable and efficient prediction engine. Furthermore, this embodiment achieves the ability to predict future risks by training and validating a heterogeneous risk prediction model. At the same time, since the model input includes the purchaser's identifier, the model will automatically capture the different patterns of mold risk of grains from different sources under the same environment during the learning process. This provides direct algorithmic support for subsequent accurate traceability and is the core technology for addressing the impact of different purchasers on the quality of raw grains.

[0022] In this embodiment, step S2 combines the initial inspection data of the acquired grain source with environmental status data to conduct a source tracing analysis of the heterogeneity risk of the grain pile, which specifically includes the following steps: S24. Input the initial moisture content, impurity content, imperfect grain rate, uniquely thermally coded purchaser identifier, temperature, and relative humidity of each monitoring point within the current monitoring period into the heterogeneity risk prediction model, and output the predicted mold marker gas concentration for each monitoring point. This achieves spatial visualization and quantitative diagnosis of risk, transforming passive monitoring into proactive prediction. Traditional storage safety relies on periodic manual spot checks or waiting for sensors to detect actual gas exceedances. This method is not only inefficient and has limited coverage, but also suffers from severe early warning lag, often resulting in irreversible deterioration of grain quality by the time problems are discovered. In contrast, this embodiment, through model prediction, can identify high-risk locations before mold actually proliferates and gas concentrations significantly increase, identifying weak links most likely to experience problems first due to poor-quality sources and unfavorable conditions. This provides managers with a clear risk map, enabling control measures to shift from aimless general prevention to precise intervention targeting high-risk points, greatly improving the scientific nature and timeliness of storage management.

[0023] S25. The monitoring points with predicted mold marker gas concentrations greater than the mold marker gas concentration threshold are identified as quality abnormality monitoring points, and the monitoring points with predicted mold marker gas concentrations less than or equal to the mold marker gas concentration threshold are identified as quality normal monitoring points. All monitoring points identified as quality abnormality monitoring points are statistically clustered based on the corresponding spatial location and the associated purchaser identifiers. The grain source corresponding to the purchaser identifier with the highest frequency is identified as the traceability direction of the batch's storage heterogeneity risk within the current monitoring period. S26. Obtain the variation range of predicted mold characteristic gas concentrations at all monitoring points, calculate the difference between the average and minimum values ​​of predicted mold characteristic gas concentrations at all monitoring points, and use the ratio of this difference to the variation range of predicted mold characteristic gas concentrations at all monitoring points as the heterogeneity risk of the grain pile within the current monitoring period. This embodiment has two core tasks: first, to objectively determine and cluster the sources of risk points; and second, to calculate the global heterogeneity risk level. In the specific implementation of this embodiment, an objective standard for determination is first required, namely, the threshold for mold marker gas concentration. This threshold is obtained based on a large amount of historical safe storage data. This embodiment collects the predicted gas concentration data of all monitoring points that have been under a generally accepted safe state (in this embodiment, the generally accepted safe state means that the quality of the grain leaving the warehouse meets the standards) for a long period (e.g., more than one year), calculates its distribution, and takes the 95th percentile as the threshold. Furthermore, after the determination is completed in this embodiment, an in-depth source tracing analysis is performed on all quality anomaly monitoring points. Each anomaly carries its spatial location information and a corresponding buyer identifier. This embodiment performs a statistical clustering operation: iterating through all anomalies and counting the frequency of each different buyer identifier. For example, if 100 anomalies are found, 65 belong to buyer A, 20 to buyer B, and 15 to buyer C. By comparison, the grain sources corresponding to the buyer identifier with the highest frequency are automatically identified as the source of the batch's storage heterogeneity risk within the current monitoring period. This achieves precise risk identification, directly attributing the general localized risk of grain piles to a specific buyer's grain source. This completely changes the traditional storage management dilemma of finding problems but being unable to assign responsibility or improve acquisition standards in a targeted manner. It reduces quality fluctuations and storage risks caused by differences in buyers from the source, and is a key step in forming a closed-loop management system.

[0024] In this embodiment, as Figure 4 As shown, step S3 combines chemical gas data and environmental condition data of the grain pile to conduct a dynamic simulation analysis of the risk of respiration heat accumulation in the grain pile, including the following specific steps: S31. Chemical gas data and environmental status data are acquired from a multi-source sensor network deployed within the grain silo, including carbon dioxide concentration and temperature and relative humidity at various monitoring points. The respiration rate of the grain at each monitoring point is calculated based on the rate of change of carbon dioxide concentration per unit time. In this embodiment, two types of real-time data are simultaneously read from the multi-source sensor nodes deployed within the grain silo, forming a three-dimensional sensing network. The first type is chemical gas data, specifically carbon dioxide concentration. Its acquisition relies on an NDIR carbon dioxide sensor installed inside the grain pile. In this embodiment, the NDIR carbon dioxide sensor works as follows: an infrared light source inside the sensor emits infrared light of a specific wavelength, which passes through the gas sample in the grain pile pores and reaches the detector. Carbon dioxide molecules have characteristic absorption of this wavelength of infrared light; the higher the concentration, the stronger the absorption, and the weaker the light intensity received by the detector. By measuring the attenuation of the light intensity and performing calculations based on temperature, pressure compensation, and a built-in algorithm, the sensor outputs the real-time carbon dioxide concentration value in digital signal form. The second type is environmental status data, including temperature and relative humidity, which is obtained from data integrated into the same probe. A digital temperature and humidity sensor continuously measures the temperature and humidity. Based on the direct measurements mentioned above, the most crucial step in this embodiment is to calculate the grain respiration rate at each monitoring point. Specifically, grain respiration is a life metabolic process that consumes oxygen and produces carbon dioxide and heat, and its rate is a direct indicator for quantifying the root cause of heat accumulation. The calculation is based on the rate of change of carbon dioxide concentration per unit time. The specific process is as follows: In this embodiment, the carbon dioxide concentration value at each monitoring point is recorded at two consecutive sampling times. The difference between the carbon dioxide concentration values ​​at two consecutive sampling times is divided by the sampling time interval to obtain the grain respiration rate, which directly reflects the increase in carbon dioxide produced by the grain at that point per unit time due to respiration, i.e., the respiration intensity. In practical applications, this embodiment can also be modified by combining parameters such as grain pile porosity and grain density to obtain a more accurate respiration rate per unit mass of grain. Based on the above, this embodiment achieves quantitative and refined monitoring of grain piles. Unlike traditional methods that rely solely on temperature to determine heat accumulation, this method monitors heat from its physiological source. By calculating the respiration rate, the metabolic activity of the grain itself can be directly assessed, solving the problem in complex storage structures where it is difficult to distinguish whether heat originates from the grain's own respiration or from external environmental heat conduction based solely on the temperature field. Especially in the deeper layers of grain piles, where air circulation is poor, the heat generated by respiration easily accumulates, forming a stuffy pile. The respiration rate data provided in this embodiment is the most direct and sensitive front-end indicator for early warning of this internal biochemical heat generation risk, providing indispensable input parameters for subsequent simulation of the dynamic processes of heat generation, conduction, and accumulation.

[0025] S32. Obtain the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days the raw grain has been stored at each monitoring point within multiple historical monitoring periods, as well as the actual temperature deviation calculated after the end of the corresponding historical monitoring period, as a respiration heat accumulation risk prediction dataset. Divide the respiration heat accumulation risk prediction dataset into a respiration heat accumulation risk prediction training set and a respiration heat accumulation risk prediction validation set. The actual temperature deviation is the difference between the average temperature of the monitoring point and the average of the average temperatures of all monitoring points within the monitoring period. In this embodiment, multiple complete period data are extracted from the archived historical database according to the same duration and start time offset as the current monitoring period. For example, if the current period is the hottest part of summer, data periods from the historical database that are also in summer are extracted first to match similar climatic backgrounds. For each monitoring point within each historical period, a multivariate time series is constructed as input features, including respiration rate, temperature, and relative humidity arranged in chronological order. Furthermore, the time segment features employ cyclic encoding, converting a 24-hour day into sine and cosine values ​​to smoothly represent the periodicity of time. The number of days the raw grain has been stored is presented as a monotonically increasing feature, reflecting the degree of grain aging. Temperature deviation directly reflects whether a point is a heat source or a cold source relative to the entire storage area. This embodiment enables the model to learn to extract patterns from dynamic changes that ultimately lead to abnormal heat accumulation. For example, the model may learn to identify a pattern of continuously rising respiration rates coupled with persistently high relative humidity, which is more predictive of future positive temperature deviations than a single high temperature. By training with a large amount of such historical paired data, the model obtained in this embodiment internalizes an understanding of the thermodynamic behavior of the grain storage ecosystem, making it far more profound and accurate than predictions based solely on current instantaneous values, thus providing the possibility for subsequent high-precision dynamic simulation predictions.

[0026] S33. Construct a Long Short-Term Memory (LSTM) network model, and train and validate it based on a respiratory heat accumulation risk prediction training set. Use the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days the raw grain has been stored at each monitoring point within each historical monitoring period as input features to the LTM network model. Use the actual temperature deviation at the end of the historical monitoring period corresponding to the input features as the output target of the LTM network model, train the model, and obtain an initial respiratory heat accumulation risk prediction model. Validate the initial respiratory heat accumulation risk prediction model using a respiratory heat accumulation risk prediction validation set. The model with an accuracy greater than or equal to the preset model accuracy is taken as the final respiratory heat accumulation risk prediction model. Figure 6As shown; specifically, Long Short-Term Memory (LSTM) networks are a variant of recurrent neural networks, particularly adept at processing and predicting time-series data because they can memorize long-term dependencies through internal gating mechanisms, making them very suitable for inferring the final state based on trends across multiple consecutive time windows in the scenario described in this embodiment; specifically, an LSTM model structure is initialized, which includes defining the number of network layers, the number of LSTM units per layer, and the final fully connected output layer. After the model is built, it is trained using the respiratory heat accumulation risk prediction training set prepared in step S32. During training, for each historical monitoring period in the dataset, the model uses the time series of five features arranged in time window order—grain respiration rate, temperature, relative humidity, time segment features, and number of days the raw grain has been stored—as the model's input features. The actual temperature deviation after the end of the historical monitoring period, corresponding to the input features, is used as the output target that the model needs to fit. Existing technologies can only take action after a significant temperature increase is detected, by which time heat may have already accumulated and caused damage. The LSTM model trained in this embodiment can dynamically simulate and predict the future distribution of heat in the grain pile based on the current respiratory intensity and environmental change sequence. In particular, it can identify in advance which points may become heat accumulation centers, and predict the location and degree of risk before the heat accumulation actually forms a large scale, so as to implement advanced and precise intervention.

[0027] S34. Input the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days of raw grain storage at each monitoring point within the current monitoring period into the trained respiration heat accumulation risk prediction model, and output the predicted value of temperature deviation at each monitoring point at the end of the current monitoring period.

[0028] In this embodiment, step S3, which combines chemical gas data and environmental condition data of the grain pile to perform dynamic simulation analysis of the risk of respiration heat accumulation in the grain pile, also includes the following specific steps: S35. The monitoring points with predicted temperature deviation values ​​greater than the temperature deviation threshold are identified as abnormal heat accumulation monitoring points, and the monitoring points with predicted temperature deviation values ​​less than or equal to the temperature deviation threshold are identified as normal heat accumulation monitoring points. In this embodiment, the temperature deviation threshold is optimized through historical data backtracking analysis. S36. Obtain the variation range of the predicted temperature deviation values ​​of all monitoring points, calculate the difference between the average and minimum values ​​of the predicted temperature deviation values ​​of all monitoring points, and use the ratio of this difference to the variation range of the predicted temperature deviation values ​​of all monitoring points as the respiratory heat accumulation risk of the grain pile within the current monitoring period. This embodiment essentially measures the relative relationship between the overall trend and internal fluctuations. A high respiratory heat accumulation risk has two possibilities: one is that the overall average value is very high (significant heat accumulation), even if the internal fluctuations are not large; the other is that the overall average value is not high, but the internal fluctuations are extremely large (extreme hot spots exist). Both of these situations mean high risk and are effectively captured by the respiratory heat accumulation risk. The types, quantities, and storage times of grains stored in different warehouses may differ, and their absolute temperature values ​​are not comparable. However, the respiratory heat accumulation risk provided in this embodiment integrates the risk index of trends and unevenness, making horizontal comparison possible and facilitating the identification of the warehouse with the highest risk by the storage center.

[0029] In this embodiment, step S4 integrates the results of heterogeneity risk tracing analysis of grain piles and the results of dynamic simulation analysis of respiratory heat accumulation risk to assess the storage risk of stored raw grains, including the following specific steps: S41. Extract the heterogeneity risk and respiratory heat accumulation risk of the grain pile within the current monitoring period; S42. The storage risk of the grain pile is obtained by weighted summation of the heterogeneity risk and the respiratory heat accumulation risk within the current monitoring period.

[0030] In this embodiment, step S5 involves making a disposal decision on the stored raw grain based on the storage risk assessment results, including the following specific details: S51. Obtain the storage risk of raw grain in the current monitoring period, and extract the source of grain and heat accumulation anomaly monitoring point for the traceability of heterogeneity risk of this batch of stored grain in the current monitoring period. S52. A preset grain storage risk threshold is established. When the storage risk of the stored raw grain in the current monitoring period exceeds the storage risk threshold, an emergency release procedure is executed for the grain source with the traceability direction of the heterogeneity risk in this batch of stored grain in the current monitoring period, and an emergency quality inspection is conducted on the grain piles at the heat accumulation anomaly monitoring points. When the storage risk of the stored raw grain in the current monitoring period is less than or equal to the storage risk threshold, a priority release procedure is executed for the grain source with the traceability direction of the heterogeneity risk in this batch of stored grain in the current monitoring period, and random inspections are conducted on the grain piles at the heat accumulation anomaly monitoring points. It should be noted that the weights and thresholds in this embodiment are determined as follows: 3000 sets of initial inspection data of the purchased grain source are obtained from the historical raw grain storage database, along with the corresponding chemical gas data and environmental status data of the grain piles, to calculate the storage risk of the stored raw grain. Simultaneously, the judgment results of whether the quality of the 3000 sets of grain piles has deteriorated are obtained. The storage risks of 3,000 sets of raw grains and the corresponding judgment results of whether the grain pile quality has deteriorated were imported into the fitting software for fitting, and the corresponding weights and threshold values ​​that meet the highest coefficient of determination were output.

[0031] Example 2 like Figure 2 As shown, this embodiment provides an online detection system for static storage of raw grain based on multi-sensor fusion, including a multi-source acquisition module, a grain source heterogeneity detection module, a grain accumulation heat detection module, a storage risk assessment module, and a disposal decision module; Among them, the multi-source acquisition module is used to collect the initial inspection data of the purchased grain pile, and through the multi-source sensor network deployed in the grain warehouse, it simultaneously collects the chemical gas data and environmental status data of the grain pile. The grain source heterogeneity detection module is used to combine the initial inspection data of the purchased grain source with environmental status data to conduct source tracing analysis of the heterogeneity risk of the grain pile; The grain accumulation heat detection module is used to dynamically simulate and analyze the risk of respiratory heat accumulation in grain piles by combining chemical gas data and environmental condition data. The storage risk assessment module is used to integrate the results of heterogeneity risk tracing analysis of grain piles and the results of dynamic simulation analysis of respiratory heat accumulation risk to assess the storage risk of stored raw grains. The disposal decision module is used to make disposal decisions for stored raw grains based on the results of the grain storage risk assessment.

[0032] The parameters and steps for implementing the corresponding functions of each unit module in the online detection system for static storage of raw grain based on multi-sensor fusion of the present invention can be referred to the parameters and steps in the embodiments of the online detection method for static storage of raw grain based on multi-sensor fusion described above, and will not be repeated here.

[0033] Example 3 An electronic device according to an embodiment of the present invention includes a processor and a memory. The memory stores a computer program that can be called by the processor. The processor executes an online detection method for static storage of raw grain based on multi-sensor fusion by calling the computer program stored in the memory. It should be noted that all computer programs for the online detection method for static storage of raw grain based on multi-sensor fusion are implemented in C language. The multi-source acquisition module, the grain source heterogeneity detection module, the grain accumulation heat detection module, the storage risk assessment module, and the disposal decision module are all controlled by a remote server.

[0034] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0035] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-sensor fusion-based online monitoring system for static grain storage, characterized in that, It includes a multi-source data acquisition module, a grain source heterogeneity detection module, a grain accumulation heat detection module, a storage risk assessment module, and a disposal decision module; Among them, the multi-source acquisition module is used to collect the initial inspection data of the purchased grain pile, and through the multi-source sensor network deployed in the grain warehouse, it simultaneously collects the chemical gas data and environmental status data of the grain pile. The grain source heterogeneity detection module is used to combine the initial inspection data of the purchased grain source with environmental status data to conduct source tracing analysis of the heterogeneity risk of the grain pile; The grain accumulation heat detection module is used to dynamically simulate and analyze the risk of respiratory heat accumulation in grain piles by combining chemical gas data and environmental condition data. The storage risk assessment module is used to integrate the results of heterogeneity risk tracing analysis of grain piles and the results of dynamic simulation analysis of respiratory heat accumulation risk to assess the storage risk of stored raw grains. The disposal decision module is used to make disposal decisions for stored raw grains based on the results of the grain storage risk assessment.

2. The online detection system for static storage of raw grain based on multi-sensor fusion according to claim 1, characterized in that, The method of combining initial inspection data of the acquired grain sources with environmental status data to conduct source tracing analysis of heterogeneity risks in grain piles includes the following specific steps: S21. Obtain the initial inspection data of the purchased grain source corresponding to each monitoring point inside the grain warehouse from the warehouse management information system, including the initial moisture content, impurity content, imperfect grain rate, and the purchaser identification with unique thermal coding; obtain the environmental status data of each monitoring point within the same time window from the multi-source sensor network deployed in the grain warehouse, including temperature and relative humidity. S22. Based on the length of the time window in the current monitoring period, divide the historical time window and obtain the initial moisture content, impurity content, imperfect grain rate, buyer identification with unique thermal coding, temperature, relative humidity, and the concentration of mold marker gas measured by the gas sensor at each monitoring point inside the grain warehouse in multiple historical monitoring periods as a heterogeneity risk prediction dataset. Divide the heterogeneity risk prediction dataset into a heterogeneity risk prediction training set and a heterogeneity risk prediction validation set. S23. Construct a random forest model. Based on the heterogeneous risk prediction training set, use the initial moisture content, impurity content, imperfect grain rate, uniquely thermally encoded purchaser identifier, temperature, and relative humidity of each monitoring point inside the grain warehouse in each historical monitoring period as the input features of the random forest model. Use the concentration of mold marker gas at each monitoring point in the heterogeneous risk prediction training set in the corresponding historical monitoring period as the output target of the random forest model. Train the random forest model to obtain the initial heterogeneous risk prediction model. Validate the initial heterogeneous risk prediction model through the heterogeneous risk prediction validation set. The initial heterogeneous risk prediction model with an accuracy greater than or equal to the preset first model is taken as the heterogeneous risk prediction model.

3. The online detection system for static storage of raw grain based on multi-sensor fusion according to claim 2, characterized in that, The method of combining initial inspection data of the acquired grain sources with environmental status data to conduct source tracing analysis of heterogeneity risks in grain piles also includes the following steps: S24. Input the initial moisture content, impurity content, imperfect particle rate, buyer identification with unique thermal coding, temperature and relative humidity of each monitoring point in the current monitoring cycle into the heterogeneity risk prediction model, and output the predicted concentration of mold marker gas at each monitoring point. S25. The monitoring points with predicted mold marker gas concentrations greater than the mold marker gas concentration threshold are identified as quality abnormality monitoring points, and the monitoring points with predicted mold marker gas concentrations less than or equal to the mold marker gas concentration threshold are identified as quality normal monitoring points. All monitoring points identified as quality abnormality monitoring points are statistically clustered based on the corresponding spatial location and the associated purchaser identifiers. The grain source corresponding to the purchaser identifier with the highest frequency is identified as the traceability direction of the batch's storage heterogeneity risk within the current monitoring period. S26. Obtain the variation range of predicted mold characteristic gas concentrations at all monitoring points, calculate the difference between the average and minimum values ​​of predicted mold characteristic gas concentrations at all monitoring points, and use the ratio of the difference to the variation range of predicted mold characteristic gas concentrations at all monitoring points as the heterogeneity risk of the grain pile in the current monitoring period.

4. The online detection system for static storage of raw grain based on multi-sensor fusion according to claim 3, characterized in that, The dynamic simulation analysis of the respiration heat accumulation risk of the grain pile, combining chemical gas data and environmental condition data, includes the following specific steps: S31. Obtain chemical gas data and environmental status data of each monitoring point inside the grain warehouse from the multi-source sensor network deployed inside the grain warehouse. The chemical gas data is carbon dioxide concentration, and the environmental status data includes temperature and relative humidity. Calculate the grain respiration rate at each monitoring point based on the rate of change of carbon dioxide concentration per unit time. S32. Obtain the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days the raw grain was stored at each monitoring point in multiple historical monitoring periods, as well as the actual temperature deviation calculated after the end of the corresponding historical monitoring period, as a dataset for predicting the risk of heat accumulation during respiration. The respiratory heat accumulation risk prediction dataset is divided into a respiratory heat accumulation risk prediction training set and a respiratory heat accumulation risk prediction validation set; wherein, the actual temperature deviation is the difference between the average temperature of the monitoring point and the average of the average temperatures of all monitoring points during the monitoring period. S33. Construct a long short-term memory network model and train and validate it based on the respiratory heat accumulation risk prediction training set. Use the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days of raw grain storage at each monitoring point in each historical monitoring period as input features of the long short-term memory network model. Use the actual temperature deviation after the end of the historical monitoring period corresponding to the input features as the output target of the long short-term memory network model. Train the model to obtain the initial respiratory heat accumulation risk prediction model. The initial respiratory fever risk prediction model was validated using a respiratory fever risk prediction validation set. The model with an accuracy greater than or equal to the preset model was used as the final respiratory fever risk prediction model. S34. Input the grain respiration rate, temperature, relative humidity, time segment characteristics, and number of days of raw grain storage at each monitoring point within the current monitoring period into the trained respiration heat accumulation risk prediction model, and output the predicted value of temperature deviation at each monitoring point at the end of the current monitoring period.

5. The online detection system for static storage of raw grain based on multi-sensor fusion according to claim 4, characterized in that, The dynamic simulation analysis of the respiratory heat accumulation risk of grain piles, which combines chemical gas data and environmental condition data, also includes the following specific steps: S35. The monitoring points with predicted temperature deviation values ​​greater than the temperature deviation threshold are identified as abnormal heat accumulation monitoring points, and the monitoring points with predicted temperature deviation values ​​less than or equal to the temperature deviation threshold are identified as normal heat accumulation monitoring points. S36. Obtain the variation range of the predicted temperature deviation values ​​of all monitoring points, calculate the difference between the average and minimum values ​​of the predicted temperature deviation values ​​of all monitoring points, and use the ratio of the difference to the variation range of the predicted temperature deviation values ​​of all monitoring points as the risk of respiratory heat accumulation in the grain pile during the current monitoring period.

6. The online detection system for static storage of raw grain based on multi-sensor fusion according to claim 5, characterized in that, The combined analysis results of heterogeneity risk tracing and dynamic simulation analysis of respiratory heat accumulation risk of the integrated grain pile are used to assess the storage risk of the stored raw grain, including the following specific steps: S41. Extract the heterogeneity risk and respiratory heat accumulation risk of the grain pile within the current monitoring period; S42. The storage risk of the grain pile is obtained by weighted summation of the heterogeneity risk and the respiratory heat accumulation risk within the current monitoring period.

7. The online detection system for static storage of raw grain based on multi-sensor fusion according to claim 6, characterized in that, The decision to dispose of stored raw grains based on the risk assessment results includes the following specific details: S51. Obtain the storage risk of raw grain in the current monitoring period, and extract the source of grain and heat accumulation anomaly monitoring point for the traceability of heterogeneity risk of this batch of stored grain in the current monitoring period. S52. Preset grain storage risk threshold. When the grain storage risk of the stored raw grain in the current monitoring period is greater than the grain storage risk threshold, an emergency release procedure will be executed for the grain source with traceability direction of the heterogeneity risk of this batch in the current monitoring period, and an emergency quality inspection will be carried out on the grain pile at the heat accumulation anomaly monitoring point. When the grain storage risk of the stored raw grain in the current monitoring period is less than or equal to the grain storage risk threshold, a priority release procedure will be executed for the grain source with traceability direction of the heterogeneity risk of this batch in the current monitoring period, and a random inspection will be carried out on the grain pile at the heat accumulation anomaly monitoring point.

8. A method for online detection of static grain storage based on multi-sensor fusion, applied to the online detection system for static grain storage based on multi-sensor fusion as described in any one of claims 1-7, characterized in that, The specific steps include the following: The system collects preliminary inspection data of the purchased grain from the grain piles and simultaneously collects chemical gas data and environmental status data of the grain piles through a multi-source sensor network deployed in the grain warehouse. By combining preliminary inspection data of the acquired grain sources with environmental status data, a source tracing analysis was conducted on the heterogeneity risk of the grain piles. By combining chemical gas data and environmental condition data of grain piles, a dynamic simulation analysis of the risk of respiratory heat accumulation in grain piles is conducted. By integrating the results of heterogeneity risk tracing analysis of grain piles and dynamic simulation analysis of respiratory heat accumulation risk, the storage risk of stored raw grains is assessed. Based on the risk assessment results of the stored grain, a decision is made on the disposal of the stored grain.

9. An electronic device, comprising: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; characterized in that the processor executes the online detection method for static storage of raw grain based on multi-sensor fusion as described in claim 8 by calling the computer program stored in the memory.

Images