A grain storage whole-process intelligent management method and system

Abstract

The present application relates to the technical field of grain storage management, and particularly relates to a kind of intelligent management method and system for the whole process of grain storage, comprising: collecting motor operating parameter of conveying equipment to extract reference current fingerprint data in the stage of warehousing, obtaining intermediate current fingerprint data in the same way during storage period operation and comparing to obtain current characteristic deviation, inputting current characteristic deviation into first level proxy model to output physical flow state change index, inputting measured moisture and grain temperature data into second level proxy model to output quality index prediction value and deterioration rate and updating dynamic file, comparing with early warning threshold to determine quality state to generate discharge priority adjustment and shunt path control instruction, generating comprehensive discharge sequence instruction to drive discharge equipment to execute according to dynamic file when multiple batches coexist, realizing the automation closed loop of grain quality prediction and discharge scheduling.

Description

Technical Field

[0001] This invention relates to the field of grain storage management technology, and in particular to an intelligent management method and system for the entire process of grain harvesting and storage. Background Technology

[0002] Grain storage and procurement is a core component of ensuring national food security. Throughout the entire process of grain storage and distribution, dynamic monitoring of grain quality and rational decision-making regarding outbound dispatch are crucial for ensuring the continuous replenishment of reserves and minimizing post-harvest losses. Currently, the grain storage industry has widely deployed grain condition monitoring and control systems, using equipment such as grain temperature and humidity sensors to monitor the storage environment online. Simultaneously, based on laboratory sampling and testing or near-infrared spectroscopy, key quality indicators such as fatty acid value and moisture content of grains can be regularly tested. Furthermore, some grain depots utilize machine learning models, combined with parameters such as effective accumulated temperature, moisture content upon entry, and storage duration, to predict long-term trends in grain quality. These technological means provide vital support for safe grain storage.

[0003] However, existing technologies still face a core problem in practical applications: a disconnect between the automated connection between grain quality forecasting and outbound scheduling execution. Specifically, the output of current quality inspection and forecasting processes is typically provided to managers in the form of reports or early warning information. Subsequent outbound decisions—including which warehouses and batches of grain should be prioritized for outbound processing, and whether specific batches need to be diverted—remain highly dependent on human experience and manual operation. This separation of "detection-forecasting" from "decision-execution" leads to a lag between the discovery of quality deterioration trends and the implementation of scheduling measures. By the time managers make decisions based on offline inspection data or interim forecast reports, some grain may have already suffered irreversible quality loss, missing the optimal time for "replacing old with new." Therefore, how to automatically connect dynamic grain quality forecasting with outbound scheduling execution to achieve data-driven, precise scheduling has become a technical problem that needs to be solved in this field. Summary of the Invention

[0004] To overcome the above deficiencies, this invention provides an intelligent management method and system for the entire process of grain storage and procurement, aiming to improve the problem in existing grain storage and procurement management where the automatic connection between grain quality prediction and outbound scheduling is broken due to their independent operation, resulting in a lag in response from the detection of quality deterioration trends to the implementation of scheduling measures.

[0005] In a first aspect, the present invention provides the following technical solution: an intelligent management method for the entire process of grain harvesting and storage, comprising:

[0006] S1. During the storage stage of the target batch of grain, the motor operating parameters of the conveying equipment are collected. After feature extraction, the reference current fingerprint data characterizing the initial physical load characteristics of the batch of grain is obtained.

[0007] S2. During the storage period, when the batch of grain passes through the conveying equipment due to storage management operations, its motor operating parameters are collected, and intermediate current fingerprint data is obtained by using the same feature extraction method as S1.

[0008] S3. Compare the intermediate current fingerprint data with the reference current fingerprint data to obtain the current characteristic deviation.

[0009] S4. Input the current characteristic deviation into the first-level proxy model and output the physical rheological state change index of the batch of grain compared with the time of storage; input the physical rheological state change index, combined with the current measured moisture and storage temperature data, into the second-level proxy model and output the predicted quality index and deterioration rate of the batch of grain, generate quality evolution trend data, and update the dynamic file of the batch of grain accordingly.

[0010] S5. Compare the quality evolution trend data with the preset quality warning threshold to determine the quality status of the batch of grain, and generate an outbound priority adjustment instruction and / or diversion path control instruction based on the quality status.

[0011] S6. When multiple batches coexist, generate a comprehensive outbound sequence instruction based on the quality evolution trend data in the dynamic archives of each batch to drive the outbound equipment to execute.

[0012] Secondly, the present invention provides the following technical solution: an intelligent management system for the entire process of grain harvesting and storage, the system comprising:

[0013] The benchmark fingerprint acquisition module is used to collect the motor operating parameters of the target batch of grain when it passes through the conveying equipment during the storage stage. After feature extraction, the benchmark current fingerprint data characterizing the initial physical load characteristics of the batch of grain is obtained.

[0014] The intermediate fingerprint acquisition module is used to collect the motor operating parameters of the batch of grain when it passes through the conveying equipment during the storage period due to storage management operations, and to obtain intermediate current fingerprint data by using the same feature extraction method as S1.

[0015] The deviation calculation module is used to compare the intermediate current fingerprint data with the reference current fingerprint data to obtain the current characteristic deviation.

[0016] The multi-level agent inference module is used to input the deviation of the current characteristics into the first-level agent model and output the physical rheological state change index of the batch of grain compared with the time of storage; the physical rheological state change index, combined with the current measured moisture and storage temperature data, is input into the second-level agent model and outputs the predicted quality index and deterioration rate of the batch of grain, generating quality evolution trend data, and updating the dynamic file of the batch of grain accordingly.

[0017] The quality status determination and scheduling module is used to compare the quality evolution trend data with the preset quality warning threshold, determine the quality status of the batch of grain, and generate outbound priority adjustment instructions and / or diversion path control instructions based on the quality status.

[0018] The multi-batch collaborative scheduling module is used to generate a comprehensive outbound sequence instruction based on the quality evolution trend data in the dynamic archives of each batch when multiple batches coexist, so as to drive the outbound equipment to execute.

[0019] The present invention has the following beneficial effects:

[0020] 1. In this invention, by constructing a fully automated closed loop from quality perception and trend prediction to scheduling execution, the quality evolution trend data output by the multi-level agent model is directly linked with the outbound scheduling system, automatically generating outbound priority adjustment instructions and diversion path control instructions. This eliminates the response lag between grain quality prediction and outbound scheduling execution due to reliance on human experience in the prior art, and significantly improves the timeliness of promoting new grain storage and management efficiency.

[0021] 2. In this invention, the operating parameters of the drive motor of the conveying equipment are used as in-situ sensing signals of the grain quality status. By calculating the deviation of the current characteristics and inverting the physical rheological state, non-destructive online continuous monitoring of the changes in physical properties of grain during storage is realized. There is no need to add additional dedicated quality detection sensors, which reduces the system deployment and maintenance costs.

[0022] 3. In this invention, a gray box model that integrates the mechanism of granular mechanics and data-driven approaches is used as the first-level proxy model. The load calculation formula of the conveying equipment is used as the physical skeleton and Gaussian process regression is used to compensate for the residuals. This not only ensures the interpretability of physical rheological state reasoning and small sample generalization ability, but also has the adaptive fitting accuracy of data-driven approaches, thus overcoming the deficiency of insufficient interpretability of pure black box models in real warehouse applications.

[0023] 4. In this invention, a time-series prediction model based on an attention mechanism is used to extract the whole process physical rheological state change index sequence and grain temperature sequence from the dynamic archives since the grain was put into storage. The influence weight of each historical moment on the current quality deterioration is automatically learned. Compared with the static prediction method that only uses the data at the current moment, the prediction accuracy of the quality index prediction value and the deterioration rate is significantly improved, and the prediction confidence interval is output to provide richer information for subsequent decision-making. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating an intelligent management method for the entire grain storage process proposed in this invention.

[0025] Figure 2 This is a schematic diagram of the current fingerprint acquisition and current characteristic deviation calculation process of an intelligent management method for the entire process of grain storage proposed in this invention.

[0026] Figure 3 This is a schematic diagram of the multi-level agent reasoning process of an intelligent management method for the entire process of grain storage proposed in this invention;

[0027] Figure 4 This is a schematic diagram of the quality status determination and scheduling process of an intelligent management method for the entire process of grain harvesting and storage proposed in this invention.

[0028] Figure 5 This is a schematic diagram of the architecture of an intelligent management system for the entire process of grain storage proposed in this invention. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1

[0031] In the first embodiment of the present invention, the present invention provides an intelligent management method for the entire process of grain harvesting and storage, such as... Figures 1-4 As shown, it includes the following steps:

[0032] S1. During the storage stage of the target batch of grain, the motor operating parameters of the conveying equipment are collected. After feature extraction, the reference current fingerprint data characterizing the initial physical load characteristics of the batch of grain is obtained.

[0033] Furthermore, in S1, the step of obtaining the reference current fingerprint data characterizing the initial physical load characteristics of this batch of grain specifically includes:

[0034] During the storage phase of the target batch of grain, collect the three-phase current waveform data of the motor when the batch of grain passes through the conveying equipment.

[0035] The system automatically identifies the no-load and load segments within the same operating cycle for the collected three-phase current waveform data of the motor. The current signal of the no-load segment is used as a real-time background noise reference, and the current signal of the load segment is differentially processed with the current signal of the adjacent no-load segment to eliminate common-mode interference.

[0036] From the differentially processed load segment current signal, extract the effective value of the current in the stable load segment and at least one waveform characteristic parameter as the reference current fingerprint data associated with the storage of this batch of grain.

[0037] Specifically, during the grain storage stage of the target batch, the conveying equipment is a scraper conveyor or a bucket elevator, driven by a three-phase asynchronous motor. High-precision current sensors deployed at the motor terminals collect real-time three-phase current waveform data of the motor as the grain passes through the conveying equipment at a sampling rate of 1000Hz. For the collected three-phase current waveform data, automatic identification of the no-load and load segments is first performed. The identification method is as follows: a current amplitude threshold is set to 1.2 times the motor's no-load rated current. A segment where the current amplitude exceeds this threshold for 200 consecutive sampling points is identified as a load segment, and a segment where the current amplitude does not exceed this threshold for 200 consecutive sampling points is identified as a no-load segment. For each identified load segment, the immediately preceding no-load segment on the time axis is selected as the real-time background noise reference for that load segment.

[0038] The current signal of each load segment is differentially processed point-by-point with the corresponding current signal of the adjacent unloaded segment. This involves subtracting the unloaded segment current signal from the load segment current signal to obtain the differentially processed load segment current signal, thus eliminating common-mode interference caused by grid voltage fluctuations and equipment vibration. From the differentially processed load segment current signal, the portion with fluctuation amplitude less than a preset fluctuation threshold is extracted as the load stable segment. The preset fluctuation threshold is 0.5 times the standard deviation of the load segment current signal. The RMS current value, crest factor, and impulse factor are extracted from the load stable segment to form the baseline current fingerprint data associated with the storage of this batch of grain. The calculation methods for each parameter are as follows: RMS current value... The calculation formula is:

[0039] ;

[0040] Where T is the time length of the load stabilization period, and i(t) is the instantaneous current value at the t-th sampling point within the load stabilization period after differential processing. This represents the summation of all sampling points within the load stability period.

[0041] Peak factor The calculation formula is:

[0042] ;

[0043] in, This represents the maximum absolute value of the instantaneous current during the load's steady-state period. This represents the effective value of the current during the stable load phase.

[0044] Pulse factor The calculation formula is:

[0045] ;

[0046] in, This represents the maximum absolute value of the instantaneous current during the load's steady-state period. It is the sum of the absolute values ​​of the instantaneous current at each sampling point within the load stabilization period, and T is the time length of the load stabilization period.

[0047] The above effective value of current Crest factor C and impulse factor As a reference current fingerprint, it is stored in association with the batch number, storage date, initial moisture content, and initial fatty acid value of the batch of grain, forming a dynamic file for the batch of grain.

[0048] S2. During the storage period, when the batch of grain passes through the conveying equipment due to storage management operations, collect its motor operating parameters and use the same feature extraction method as S1 to obtain intermediate current fingerprint data.

[0049] Furthermore, in S2, the steps for obtaining the intermediate current fingerprint data specifically include:

[0050] During the storage period, when the batch of grain passes through the conveying equipment due to storage management operations, the three-phase current waveform data of the motor of the batch of grain passing through the conveying equipment is collected.

[0051] For the collected three-phase current waveform data of the motor, perform the same no-load segment and load segment identification and differential processing as in step S1 to extract the current signal segment corresponding to the load stable segment in the process of transporting this batch of grain.

[0052] Extract the effective value of the current and at least one waveform characteristic parameter from the current signal segment of the load stabilization segment as intermediate current fingerprint data, and store them in association with the dynamic archive of the batch of grain.

[0053] Specifically, during grain storage, warehousing management operations include transferring grain between warehouses, ventilation and temperature equalization, or outbound operations. When the same batch of grain passes through the same conveyor equipment as in step S1 again due to the aforementioned warehousing management operations, a high-precision current sensor deployed at the drive motor terminal of the conveyor equipment collects the three-phase current waveform data of the motor in real time at a sampling rate of 1000Hz. The collected three-phase current waveform data of the motor is processed using the same procedure as in step S1. Automatic identification of the no-load segment and the load segment is performed sequentially, with the current amplitude threshold and the number of continuous sampling points remaining consistent with S1, namely 1.2 times the rated no-load current of the motor and 200 consecutive sampling points. After identification, the current signal of each load segment is differentially processed with the corresponding adjacent no-load segment current signal, sample by sample, to obtain the differentially processed load segment current signal. The load stability segment is extracted from the differentially processed load segment current signal, with the fluctuation threshold being the same as in S1, which is 0.5 times the standard deviation of the load segment current signal.

[0054] The effective current value, crest factor, and impulse factor are extracted from the load steady-state section. The calculation formulas for each parameter are exactly the same as those for S1, and the effective current value is calculated. Crest factor C and impulse factor Together, they constitute the intermediate current fingerprint data of this batch of grain.

[0055] The intermediate current fingerprint data is associated with the batch number and collection timestamp of the grain batch and stored in the dynamic archive of that batch. Since this operation uses the same conveying equipment, the same signal processing method, and the same feature extraction method as the warehousing stage, the intermediate current fingerprint data and the reference current fingerprint data have a basis for direct comparison, providing data support for subsequent calculation of current characteristic deviation.

[0056] S3. Compare the intermediate current fingerprint data with the reference current fingerprint data to obtain the current characteristic deviation.

[0057] Furthermore, in S3, the step of obtaining the current characteristic deviation specifically includes:

[0058] The various feature parameters in the intermediate current fingerprint data and the reference current fingerprint data are used to construct the reference feature vector and the current feature vector, respectively.

[0059] Calculate the Mahalanobis distance between the baseline feature vector and the current feature vector in the multidimensional feature space;

[0060] The calculated Mahalanobis distance is used as the current characteristic deviation, which characterizes the overall deviation of the current physical load characteristics of the batch of grain from the initial physical load characteristics, and is stored in the dynamic archive of the batch of grain.

[0061] Specifically, the reference current fingerprint data stored in stage S1 and the intermediate current fingerprint data corresponding to this operation stored in stage S2 are retrieved from the dynamic archive of this batch of grain. The reference current fingerprint data includes the effective value of the current. Crest factor C and impulse factor The intermediate current fingerprint data also includes the effective value of the current collected in this study. Crest factor C and impulse factor The three feature parameters from the reference current fingerprint data are used to construct a reference feature vector. The three feature parameters from the intermediate current fingerprint data are used to construct the current feature vector. Since the load stabilization time lengths in the baseline current fingerprint data and the intermediate current fingerprint data may differ, linear interpolation is used to align the feature parameter sequences corresponding to the two load stabilization segments to the same number of sampling points before calculating the Mahalanobis distance. In this embodiment, the sequence is uniformly aligned to 200 sampling points before constructing the feature vector. The baseline feature vector represents the initial physical load characteristics of the batch of grain when it enters the warehouse, while the current feature vector represents the current physical load characteristics of the batch of grain during this storage management operation.

[0062] The Mahalanobis distance between the current eigenvector and the reference eigenvector in the multidimensional feature space is calculated and used as the deviation of the current feature. The formula for calculating the Mahalanobis distance is:

[0063] ;

[0064] in, For the current feature vector, As the baseline feature vector, It is the inverse of the covariance matrix. The calculated Mahalanobis distance is the deviation of the current characteristic.

[0065] The covariance matrix SS is constructed during the system initialization phase by repeatedly collecting current characteristic parameters under no-load and standard load conditions on the transmission equipment and calculating the covariance between these characteristic parameters. The covariance matrix S is a 3×3 symmetric matrix, and its element in the i-th row and j-th column is... The calculation formula is:

[0066] ;

[0067] Where K is the total number of samples collected during the initialization phase. Let i be the value of the i-th feature parameter in the k-th sample. Let K be the mean of the i-th feature parameter across K samples. In this embodiment, the total number of samples K collected during the initialization phase is set to 100. To ensure the adaptability of the covariance matrix to the slow drift of the device state during long-term monitoring, a sliding window method is used to dynamically update the covariance matrix. Whenever a new intermediate current fingerprint data that has been manually verified as reliable is added, the oldest historical sample is removed from the window, the new sample is added to the window, and the covariance matrix is ​​recalculated. The sliding window size is set to 100 samples. In the automatic operation mode without manual verification, each new data is only added to the sliding window if the deviation of the current feature does not trigger an abnormal alarm, in order to avoid abnormal data contaminating the covariance matrix.

[0068] The calculated Mahalanobis distance As a current characteristic deviation, it is associated with the batch number of the grain and the timestamp of this collection, and stored in the dynamic file of that batch of grain. Current characteristic deviation The larger the value, the greater the deviation of the current physical load characteristics of the batch of grain from the initial physical load characteristics when it was put into storage, that is, the more significant the changes in physical rheological states such as scattering and internal friction angle have occurred.

[0069] S4. Input the deviation of the current characteristics into the first-level surrogate model and output the physical rheological state change index of the batch of grain compared with the time of entry into the warehouse. Input the physical rheological state change index, combined with the current measured moisture and storage temperature data, into the second-level surrogate model and output the predicted quality index and deterioration rate of the batch of grain, generate quality evolution trend data, and update the dynamic file of the batch of grain accordingly.

[0070] Furthermore, in S4, the step of outputting the index of the change in physical rheological state of this batch of grain compared to when it entered the warehouse specifically includes:

[0071] The deviation of the current characteristic is used as the input parameter of the first-level surrogate model and loaded into the pre-trained first-level surrogate model.

[0072] The first-level proxy model is a gray box model that integrates the mechanism of granular mechanics and data-driven approach. It uses the load calculation formula of conveying equipment in granular mechanics as the physical skeleton and uses a Gaussian process regression model to compensate for the residual between the output of the physical skeleton and the actual measured value.

[0073] The first-level proxy model outputs the physical rheological state change index and prediction uncertainty measure of the batch of grain compared with the time of entry into the warehouse based on the deviation of the current characteristics.

[0074] The physical rheological state change index and the measurement of prediction uncertainty are stored in the dynamic archive of this batch of grain.

[0075] Furthermore, in S4, the steps of generating quality evolution trend data and updating the dynamic profile of this batch of grain specifically include:

[0076] Obtain the current measured moisture content of this batch of grain;

[0077] From the dynamic archives of this batch of grain, extract the physical rheological state change index sequence and grain temperature sequence recorded at all historical collection moments since it entered the warehouse;

[0078] The current measured moisture value, the physical rheological state change index sequence, and the grain temperature sequence together constitute the characteristic time series data of the whole process;

[0079] The entire process of feature time series data is input into the time series prediction model based on the attention mechanism. The model learns the influence weight of the features at each historical moment on the current quality deterioration through the self-attention mechanism, and makes predictions based on the weighted time series features.

[0080] The time-series forecasting model outputs the predicted quality indicators, deterioration rate, and prediction confidence interval for this batch of grain.

[0081] Based on the predicted values ​​of quality indicators, the rate of deterioration, and the prediction confidence interval, the quality evolution trend data of this batch of grain is generated.

[0082] The quality evolution trend data is stored in the dynamic archive of this batch of grain, replacing or supplementing the original quality evolution trend data in the dynamic archive, thus completing the update of the dynamic archive.

[0083] Specifically, step S4 is divided into two sub-steps: first-level proxy model reasoning and second-level proxy model reasoning. The first-level proxy model is a gray-box model that integrates the granular mechanical mechanism with data-driven approaches. First, the deviation of the current characteristic calculated in step S3 is retrieved from the dynamic archive of this batch of grain. The physical framework of the first-level proxy model adopts the load calculation formula for conveying equipment in granular mechanics. For scraper conveyors, the load torque of the drive motor is positively correlated with the internal friction coefficient of the conveyed material, and the deviation of the current characteristic is... The physical skeleton outputs the initial physical rheological state change index, which satisfies a linear mapping relationship with the change in the material's internal friction coefficient Δμ. Calculated using the following formula:

[0084] ;

[0085] Where 'a' is the proportionality coefficient and 'b' is the bias constant. In this embodiment, 'a' is 0.85 and 'b' is 0.01.

[0086] There is a residual between the physical skeleton output and the actual physical rheological state change index. To compensate for this residual, a Gaussian process regression model is used. The training process of the Gaussian process regression model is as follows: The deviation of current characteristics and the corresponding true values ​​of the physical rheological state change index of multiple historical batches of grain during actual storage are collected. The true values ​​of the physical rheological state change index are obtained by measuring the rate of change of the grain's free-flow angle in the laboratory. The total number of training samples is no less than 200 groups. The residual between the physical skeleton output value and the true value is used as the fitting target of the Gaussian process regression, and the deviation of current characteristics is used as the fitting target. As input, a Gaussian process regression model is trained. The kernel function uses radial basis functions, and the hyperparameters are automatically determined by maximizing the logarithmic marginal likelihood. During online inference, the deviation of the current feature is... Input the physical skeleton to obtain the initial physical rheological state change index. At the same time, Input the trained Gaussian process regression model to obtain the residual prediction mean. and residual prediction standard deviation The index of change in the physical rheological state of this batch of grain compared to when it entered the warehouse. The calculation formula is:

[0087] ;

[0088] The measurement of forecast uncertainty is taken as the standard deviation of the residual forecast, i.e. The physical rheological state change index and measurement of forecast uncertainty Link the current timestamp and store it in the dynamic archive of this batch of grain.

[0089] The second-level proxy model is a time-series prediction model based on an attention mechanism. First, it obtains the current measured moisture value of the batch of grain. The moisture content is obtained through online detection using a microwave moisture meter installed on the conveying equipment, as described in this embodiment. The value was 14.2%. The physical rheological state change index sequence, recorded at all historical collection points since the grain entered storage, was extracted from the dynamic archives of this batch of grain. Grain temperature sequence Where T represents the total number of data collections at the current moment, and the grain temperature data is taken from the average temperature of the grain pile recorded by the grain condition monitoring and control system. A full-process characteristic time-series data is constructed. The current measured moisture value is then used. As a constant feature, it is concatenated with the physical rheological state change index and grain temperature at each historical moment to form the input feature matrix. The eigenvector at time step t is .

[0090] The time series prediction model employs an LSTM network structure based on an attention mechanism. The model consists of one LSTM layer, one self-attention layer, and one fully connected output layer. The LSTM layer has a hidden layer dimension of 128, encoding the input feature matrix into a sequence of hidden states. The attention layer calculates the weights of the impact of each historical moment on the current quality degradation, and the attention weights are used for this purpose. The calculation formula is:

[0091] ;

[0092] Where W is the trainable weight matrix, and v is the trainable weight vector. The formula for calculating the weighted context vector c is:

[0093] ;

[0094] The fully connected output layer takes the context vector c as input and outputs the predicted value of the quality index. Deterioration rate r and prediction variance Predicted values ​​of quality indicators The predicted fatty acid value at the current moment is represented by mg KOH / 100g; the deterioration rate r represents the predicted daily increase in fatty acid value, in mg KOH / 100g / day; the prediction confidence interval is taken as... .

[0095] The training process of the time-series prediction model is as follows: Collect time-series data of the entire process of grain from storage to shipment across multiple historical batches. The label is the true fatty acid value obtained from laboratory testing at the time of shipment for the corresponding batch of grain. The deterioration rate label is approximated by the difference between two adjacent quality test values. The loss function consists of a weighted sum of the mean squared error of the predicted values ​​and the negative log-likelihood of the variance. The Adam optimizer is used with an initial learning rate of 0.001, a batch size of 32, and 300 training epochs.

[0096] During online inference, the constructed full-process feature time-series data is input into the trained time-series prediction model, which outputs predicted quality index values. Deterioration rate r and prediction standard deviation The predicted values ​​of quality indicators, the rate of deterioration, and the predicted confidence interval are combined into quality evolution trend data, which is stored in the dynamic file of this batch of grain. The original quality evolution trend data in the dynamic file is replaced or supplemented to complete the update of the dynamic file.

[0097] S5. Compare the quality evolution trend data with the preset quality early warning threshold to determine the quality status of the batch of grain, and generate outbound priority adjustment instructions and / or diversion path control instructions based on the quality status.

[0098] Furthermore, in S5, the steps for determining the quality status of this batch of grain specifically include:

[0099] The latest quality evolution trend data and predictive uncertainty measures are read from the dynamic archives of this batch of grain. The quality evolution trend data includes at least the predicted values ​​of quality indicators and the rate of deterioration at the current moment.

[0100] Calculate the remaining safe storage days for this batch of grain based on the current predicted quality indicators, the rate of deterioration, and the preset quality warning threshold.

[0101] The remaining safe storage days are compared with a preset warning day threshold, which includes at least a first warning day threshold and a second warning day threshold.

[0102] If the remaining safe storage days are greater than the first warning days threshold, or the prediction uncertainty measure is higher than the preset confidence threshold, then the quality status of the batch of grain is determined to be safe.

[0103] If the remaining safe storage days are between the second warning days threshold and the first warning days threshold, and the prediction uncertainty measure is not higher than the preset confidence threshold, then the quality status of this batch of grain is determined to be of concern.

[0104] If the remaining safe storage days are less than the second warning days threshold, and the prediction uncertainty measure is not higher than the preset confidence threshold, then the quality status of this batch of grain is determined to be urgent.

[0105] Furthermore, in S5, the steps for generating outbound priority adjustment instructions and / or diversion path control instructions based on quality status specifically include:

[0106] When the quality status is safe, an instruction is generated to maintain the current outbound priority, without triggering the diversion path control;

[0107] When the quality status is "concerned", a first adjustment instruction is generated to increase the outbound priority of this batch of grain, indicating that the outbound order of this batch of grain is brought forward to before the normal batch, but without triggering diversion path control.

[0108] When the quality status is urgent, a second adjustment instruction is generated to increase the priority of the batch of grain leaving the warehouse. At the same time, a diversion path control instruction is generated to guide the batch of grain to a preset downgraded treatment or feed grain dedicated channel.

[0109] Specifically, step S5 consists of two sub-steps: quality status determination and scheduling instruction generation. In the quality status determination sub-step, the latest quality evolution trend data and prediction uncertainty measure stored in step S4 are read from the dynamic archive of this batch of grain. The quality evolution trend data includes the predicted values ​​of quality indicators at the current moment. Deterioration rate r and prediction standard deviation The measurement of prediction uncertainty is the standard deviation of the residual predictions output by the first-level surrogate model. Predicted values ​​of quality indicators The current value represents the predicted fatty acid value, in mg KOH / 100g; the deterioration rate r represents the predicted daily increase in fatty acid value, in mg KOH / 100g / day.

[0110] Based on the current quality index prediction value Deterioration rate r and preset quality warning threshold Calculate the remaining safe storage days for this batch of grain. In application scenarios where the degradation rate is relatively stable and quality indicators increase approximately linearly over time, a linear extrapolation method is used for calculation. The calculation formula is as follows:

[0111] ;

[0112] in, The preset quality warning threshold, Let be the predicted quality index value at the current moment, and 'r' be the rate of deterioration. In this embodiment, the quality warning threshold is... The value is set at 25 mg KOH / 100g, which is the upper limit of the recommended storage standard for rice fatty acid value. It should be noted that when the rate of deterioration is unstable or the quality deterioration exhibits a significantly non-linear characteristic, the linear extrapolation formula may produce large deviations. In this case, the quality warning threshold can be adjusted. As the target value, the time-series data of the entire process are input into the time-series prediction model, and the quality index is directly estimated through daily iterative prediction. The required remaining days are used as a supplementary calculation method for the remaining safe storage days. Meanwhile, if the prediction confidence interval output by the time series prediction model exceeds a preset width threshold, a manual review process is triggered to ensure the reliability of critical decisions.

[0113] If the rate of deterioration r is less than or equal to 0, it indicates that the quality index does not show an increasing trend, and the remaining safe storage days will be... Set to a preset maximum value. Calculate the remaining safe storage days. It is compared with a preset warning day threshold. The preset warning day threshold includes a first warning day threshold. Second warning days threshold ,and Greater than In this embodiment, The value is 90 days. The value is 30 days.

[0114] The judgment rules for the quality status are as follows: If the remaining safe storage days are greater than the first warning day threshold , or the prediction uncertainty measure is higher than the preset confidence threshold , the quality status of this batch of grain is determined to be safe. The preset confidence threshold has a value of 0.15, which is determined through statistical analysis of historical data. When the standard deviation of the residual prediction exceeds 0.15, it indicates that the confidence of the physical rheological state inference result of the first-level proxy model is insufficient.

[0115] If the remaining safe storage days fall between the second warning day threshold and the first warning day threshold , that is , and the prediction uncertainty measure is not higher than the preset confidence threshold , the quality status of this batch of grain is determined to be of concern.

[0116] If the remaining safe storage days are less than the second warning day threshold , that is <D2, and the prediction uncertainty measure is not higher than the preset confidence threshold , the quality status of this batch of grain is determined to be urgent. Through the above arrangement, when the prediction uncertainty of the model is high, the system adopts a conservative strategy to preferentially determine the status as safe, so as to avoid making high-risk scheduling decisions based on low-confidence data, and improve the reliability of the system in complex environments.

[0117] In the dispatch instruction generation sub-step, based on the determined quality status, corresponding outbound priority adjustment instructions and diversion path control instructions are generated. When the quality status is safe, it indicates that the batch of grain has no risk of quality deterioration within the foreseeable storage period. An instruction to maintain the current outbound priority is generated, preserving the original sorting position of the batch of grain in the outbound queue, without triggering diversion path control. When the quality status is of concern, it indicates that the quality of the batch of grain has shown a deterioration trend. The remaining safe storage time is limited but has not yet reached an urgent level. A first adjustment instruction to increase the outbound priority of the batch of grain is generated, instructing that the outbound order of the batch of grain be advanced to before all batches in the safe status, but without triggering diversion path control. The batch of grain is still outbound through the normal outbound channel. When the quality status is urgent, it indicates that the quality of the batch of grain has approached or reached the warning threshold, and the remaining safe storage time is insufficient. A second adjustment instruction to increase the outbound priority of the batch of grain is generated, raising the outbound order of the batch of grain to the highest priority. At the same time, a diversion path control instruction is generated, instructing that the batch of grain be guided to the preset downgraded processing channel or the feed grain dedicated channel. The downgrading processing channel points to the special unloading port in the grain depot used to receive substandard grain, while the feed grain special channel points to the special outbound line that is directly loaded onto trucks and sent to feed processing enterprises.

[0118] The generated outbound priority adjustment instructions and diversion path control instructions are sent to the grain depot's outbound control system. Based on these instructions, the outbound control system adjusts the opening and closing status of the gates at each unloading port and the steering configuration of the conveying equipment to execute the outbound operation for the corresponding batch.

[0119] S6. When multiple batches coexist, generate a comprehensive outbound sequence instruction based on the quality evolution trend data in the dynamic files of each batch to drive the outbound equipment to execute.

[0120] Furthermore, in S6, the specific steps for generating a comprehensive outbound sequence instruction to drive the outbound equipment to execute include:

[0121] Read the dynamic records of each batch of grain separately to obtain the remaining safe storage days and deterioration rate in the latest record of each batch;

[0122] The batches are sorted in ascending order of remaining safe storage days. For batches with the same remaining safe storage days, those with the fastest deterioration rate are given priority, and a comprehensive outbound sequence instruction is generated.

[0123] The integrated outbound sequence command is sent to the outbound control system to control the outbound equipment to perform outbound operations according to the integrated outbound sequence command.

[0124] Specifically, when multiple batches of grain are simultaneously awaiting release from the grain depot, the host computer management platform initiates a multi-batch collaborative scheduling process. First, it reads the dynamic records of each batch of grain and retrieves the remaining safe storage days from the latest records of each batch. And the rate of degradation r. Remaining safe storage days. The deterioration rate r is calculated by step S5 and stored in the dynamic archives of each batch. The deterioration rate r is output by the second-level agent model in step S4 and stored in the dynamic archives of each batch.

[0125] Secondly, the batches are sorted in ascending order of remaining safe storage days. The sorting rule is: remaining safe storage days. The shorter the batch, the higher its urgency for release, and the higher its ranking in the release sequence. For batches with the same remaining safe storage days, those with the faster deterioration rate (r) are prioritized. A faster deterioration rate (r) indicates a faster rate of quality deterioration in the batch of grain, and should be released first given the same remaining safe storage days.

[0126] ;

[0127] in, The remaining safe storage days are in days; r is the rate of deterioration, in mg KOH / 100g / day. It is a very small positive number, used to prevent the denominator from being zero, and takes the value of ; This is the degradation rate weighting coefficient, used to adjust the contribution of degradation rate in the ranking process. The value ranges from 0 to 1, and the specific value is determined based on statistical analysis of historical outbound data. When it is necessary to emphasize the impact of the degradation rate on the sorting, the value is increased. Reduce when there is a greater emphasis on the remaining safe storage days. In this embodiment The value is 0.5. The higher the priority score P value, the higher the urgency of releasing this batch of grain.

[0128] The batches are sorted in descending order of their overall outbound priority score P, generating an overall outbound sequence instruction. The overall outbound sequence instruction is a list of batch numbers arranged in descending order of outbound priority, with the first batch in the list being the highest priority outbound batch and the last batch being the lowest priority outbound batch.

[0129] The generated integrated outbound sequence instruction is sent to the grain depot's outbound control system. The outbound control system is a programmable logic controller or a distributed control system. After receiving the integrated outbound sequence instruction, it controls the opening and closing of the outbound gates of each warehouse and the steering configuration of the conveying equipment in sequence according to the batch order specified in the instruction, so as to transport the corresponding batch of grain to the designated unloading port or outbound channel, and complete the coordinated outbound operation of multiple batches.

[0130] Example 2

[0131] In existing grain storage and management, the grain quality detection and prediction stages and the outbound scheduling decision-making stages operate independently. The output of quality detection and prediction is typically provided to managers in the form of reports or early warning information. Subsequent outbound scheduling decisions, including which batches of grain should be prioritized for outbound processing and whether specific batches need to be downgraded and diverted, still rely on human experience and manual operation. This separation of prediction and execution leads to a significant response lag between the discovery of a quality deterioration trend and the actual implementation of scheduling measures. By the time managers make outbound decisions based on offline detection data or interim prediction reports, some grain may have already suffered irreversible quality loss, missing the optimal time for storage. To address these problems, this invention provides an intelligent management system for the entire grain storage and procurement process, the structure of which is as follows: Figure 5 As shown. The specific implementation process of this system is as follows:

[0132] The benchmark fingerprint acquisition module is used to collect the motor operating parameters of the target batch of grain when it passes through the conveying equipment during the storage stage. After feature extraction, the benchmark current fingerprint data characterizing the initial physical load characteristics of the batch of grain is obtained.

[0133] The intermediate fingerprint acquisition module is used to collect the motor operating parameters of the batch of grain when it passes through the conveying equipment during the storage period due to storage management operations. It uses the same feature extraction method as S1 to obtain intermediate current fingerprint data.

[0134] The deviation calculation module is used to compare the intermediate current fingerprint data with the reference current fingerprint data to obtain the current characteristic deviation.

[0135] The multi-level agent inference module is used to input the deviation of the current characteristics into the first-level agent model and output the physical rheological state change index of the batch of grain compared with the time of storage; the physical rheological state change index, combined with the current measured moisture and storage temperature data, is input into the second-level agent model and outputs the predicted quality index and deterioration rate of the batch of grain, generating quality evolution trend data, and updating the dynamic file of the batch of grain accordingly.

[0136] The quality status determination and scheduling module is used to compare the quality evolution trend data with the preset quality warning threshold to determine the quality status of the batch of grain, and generate outbound priority adjustment instructions and / or diversion path control instructions based on the quality status.

[0137] The multi-batch collaborative scheduling module is used to generate a comprehensive outbound sequence instruction based on the quality evolution trend data in the dynamic files of each batch when multiple batches coexist, so as to drive the outbound equipment to execute.

[0138] Specifically, the baseline fingerprint acquisition module collects motor operating parameters of the target batch of grain as it passes through the conveyor equipment during the warehousing stage. After feature extraction, it obtains baseline current fingerprint data characterizing the initial physical load characteristics of the batch of grain. The intermediate fingerprint acquisition module collects motor operating parameters of the batch of grain as it passes through the conveyor equipment during the storage period due to warehousing management operations. It uses the same feature extraction method as the baseline fingerprint acquisition module to obtain intermediate current fingerprint data. The deviation calculation module compares the intermediate current fingerprint data with the baseline current fingerprint data to obtain the current characteristic deviation. The multi-level surrogate inference module inputs the current characteristic deviation into the first-level surrogate model and outputs the physical rheological state change index of the batch of grain compared to when it entered the warehouse. It then inputs the physical rheological state change index, combined with the current measured moisture and storage temperature data, into the second-level surrogate model to output the predicted quality index and deterioration rate of the batch of grain, generating quality evolution trend data to update the dynamic file of the batch of grain. The quality status determination and scheduling module compares quality evolution trend data with preset quality warning thresholds to determine the quality status of the batch of grain, and generates outbound priority adjustment instructions and diversion path control instructions based on the quality status. The multi-batch collaborative scheduling module generates a comprehensive outbound sequence instruction based on the quality evolution trend data in the dynamic files of each batch when multiple batches coexist, thereby driving the outbound equipment to execute.

[0139] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent management of the entire process of grain harvesting and storage, characterized in that, include: S1. During the storage stage of the target batch of grain, the motor operating parameters of the conveying equipment are collected. After feature extraction, the reference current fingerprint data characterizing the initial physical load characteristics of the batch of grain is obtained. S2. During the storage period, when the batch of grain passes through the conveying equipment due to storage management operations, its motor operating parameters are collected, and intermediate current fingerprint data is obtained by using the same feature extraction method as S1. S3. Compare the intermediate current fingerprint data with the reference current fingerprint data to obtain the current characteristic deviation. S4. Input the current characteristic deviation into the first-level proxy model and output the physical rheological state change index of the batch of grain compared with the time of storage; input the physical rheological state change index, combined with the current measured moisture and storage temperature data, into the second-level proxy model and output the predicted quality index and deterioration rate of the batch of grain, generate quality evolution trend data, and update the dynamic file of the batch of grain accordingly. S5. Compare the quality evolution trend data with the preset quality warning threshold to determine the quality status of the batch of grain, and generate an outbound priority adjustment instruction and / or diversion path control instruction based on the quality status. S6. When multiple batches coexist, generate a comprehensive outbound sequence instruction based on the quality evolution trend data in the dynamic archives of each batch to drive the outbound equipment to execute.

2. The intelligent management method for the entire process of grain harvesting and storage according to claim 1, characterized in that, In step S1, the step of obtaining the reference current fingerprint data characterizing the initial physical load characteristics of the batch of grain specifically includes: During the storage phase of the target batch of grain, collect the three-phase current waveform data of the motor when the batch of grain passes through the conveying equipment. The system automatically identifies the no-load and load segments within the same operating cycle for the collected three-phase current waveform data of the motor. The current signal of the no-load segment is used as a real-time background noise reference, and the current signal of the load segment is differentially processed with the current signal of the adjacent no-load segment to eliminate common-mode interference. From the differentially processed load segment current signal, extract the effective value of the current in the stable load segment and at least one waveform characteristic parameter as the reference current fingerprint data associated with the storage of this batch of grain.

3. The intelligent management method for the entire process of grain harvesting and storage according to claim 1, characterized in that, In step S2, the step of obtaining intermediate current fingerprint data specifically includes: During the storage period, when the batch of grain passes through the conveying equipment due to storage management operations, the three-phase current waveform data of the motor when the batch of grain passes through the conveying equipment is collected. For the collected three-phase current waveform data of the motor, perform the same no-load segment and load segment identification and differential processing as in step S1 to extract the current signal segment corresponding to the load stable segment in the process of transporting this batch of grain. From the current signal segment of the load stabilization segment, extract the effective value of the current of the load stabilization segment and at least one waveform characteristic parameter as intermediate current fingerprint data, and store it in association with the dynamic archive of the batch of grain.

4. The intelligent management method for the entire process of grain harvesting and storage according to claim 1, characterized in that, In step S3, the step of obtaining the current characteristic deviation specifically includes: The intermediate current fingerprint data and the reference current fingerprint data are used to construct a reference feature vector and a current feature vector, respectively. Calculate the Mahalanobis distance between the baseline feature vector and the current feature vector in the multidimensional feature space; The calculated Mahalanobis distance is used as the current characteristic deviation, which characterizes the overall deviation of the current physical load characteristics of the batch of grain from the initial physical load characteristics, and is stored in the dynamic file of the batch of grain.

5. The intelligent management method for the entire process of grain harvesting and storage according to claim 1, characterized in that, In step S4, the step of outputting the index of change in the physical rheological state of the batch of grain compared to when it was stored specifically includes: The deviation of the current characteristic is used as the input parameter of the first-level surrogate model and loaded into the pre-trained first-level surrogate model; The first-level proxy model is a gray box model that integrates the mechanism of granular mechanics and data-driven approach. It uses the load calculation formula of conveying equipment in granular mechanics as the physical skeleton and uses a Gaussian process regression model to compensate for the residual between the output of the physical skeleton and the actual measured value. The first-level proxy model outputs the physical rheological state change index and prediction uncertainty measure of the batch of grain compared with the time of entry into the warehouse based on the deviation of the current characteristics. The physical rheological state change index and the measurement of prediction uncertainty are stored in the dynamic archive of this batch of grain.

6. The intelligent management method for the entire process of grain harvesting and storage according to claim 1, characterized in that, In step S4, the step of generating quality evolution trend data and updating the dynamic profile of the batch of grain specifically includes: Obtain the current measured moisture content of this batch of grain; From the dynamic archives of this batch of grain, extract the physical rheological state change index sequence and grain temperature sequence recorded at all historical collection moments since it entered the warehouse; The current measured moisture value, the physical rheological state change index sequence, and the grain temperature sequence together constitute the full-process characteristic time series data; The entire process feature time series data is input into a time series prediction model based on an attention mechanism. The model learns the influence weight of each historical moment feature on the current quality deterioration through a self-attention mechanism, and makes predictions based on the weighted time series features. The time-series prediction model outputs the predicted quality indicators, deterioration rate, and prediction confidence interval for this batch of grain. Based on the predicted values ​​of the quality indicators, the rate of deterioration, and the prediction confidence interval, the quality evolution trend data of this batch of grain is generated. The quality evolution trend data is stored in the dynamic archive of this batch of grain, replacing or supplementing the original quality evolution trend data in the dynamic archive, thus completing the update of the dynamic archive.

7. The intelligent management method for the entire process of grain harvesting and storage according to claim 1, characterized in that, In step S5, the step of determining the quality status of the batch of grain specifically includes: The latest quality evolution trend data and prediction uncertainty measure are read from the dynamic archive of this batch of grain. The quality evolution trend data includes at least the predicted value of the quality index and the rate of deterioration at the current moment. Based on the predicted quality index value at the current moment, the deterioration rate, and the preset quality warning threshold, calculate the remaining safe storage days for this batch of grain. The remaining safe storage days are compared with a preset warning day threshold, wherein the preset warning day threshold includes at least a first warning day threshold and a second warning day threshold. If the remaining safe storage days are greater than the first warning days threshold, or the prediction uncertainty measure is higher than the preset confidence threshold, then the quality status of the batch of grain is determined to be safe. If the remaining safe storage days are between the second warning days threshold and the first warning days threshold, and the prediction uncertainty measure is not higher than the preset confidence threshold, then the quality status of this batch of grain is determined to be of concern. If the remaining safe storage days are less than the second warning days threshold, and the prediction uncertainty measure is not higher than the preset confidence threshold, then the quality status of the batch of grain is determined to be urgent.

8. The intelligent management method for the entire process of grain harvesting and storage according to claim 7, characterized in that, In step S5, the step of generating an outbound priority adjustment instruction and / or a diversion path control instruction based on the quality status specifically includes: When the quality status is safe, an instruction is generated to maintain the current outbound priority, and the diversion path control is not triggered. When the quality status is "concerned", a first adjustment instruction is generated to increase the outbound priority of this batch of grain, indicating that the outbound order of this batch of grain is brought forward to before the normal batch, but without triggering diversion path control. When the quality status is urgent, a second adjustment instruction is generated to increase the priority of the batch of grain leaving the warehouse, and a diversion path control instruction is generated at the same time to guide the batch of grain to a preset downgraded treatment or feed grain dedicated channel.

9. The intelligent management method for the entire process of grain harvesting and storage according to claim 1, characterized in that, In step S6, the step of generating a comprehensive outbound sequence instruction to drive the outbound equipment to execute the instructions specifically includes: Read the dynamic records of each batch of grain separately to obtain the remaining safe storage days and deterioration rate in the latest record of each batch; The batches are sorted in ascending order of remaining safe storage days. For batches with the same remaining safe storage days, those with the fastest deterioration rate are given priority, and a comprehensive outbound sequence instruction is generated. The integrated outbound sequence instruction is sent to the outbound control system to control the outbound equipment to perform outbound operations according to the integrated outbound sequence instruction.

10. An intelligent management system for the entire process of grain harvesting and storage, characterized in that, The system for the intelligent management method of the entire process of grain acquisition and storage as described in any one of claims 1-9 comprises: The benchmark fingerprint acquisition module is used to collect the motor operating parameters of the target batch of grain when it passes through the conveying equipment during the storage stage. After feature extraction, the benchmark current fingerprint data characterizing the initial physical load characteristics of the batch of grain is obtained. The intermediate fingerprint acquisition module is used to collect the motor operating parameters of the batch of grain when it passes through the conveying equipment during the storage period due to storage management operations, and to obtain intermediate current fingerprint data by using the same feature extraction method as S1. The deviation calculation module is used to compare the intermediate current fingerprint data with the reference current fingerprint data to obtain the current characteristic deviation. The multi-level agent inference module is used to input the deviation of the current characteristics into the first-level agent model and output the physical rheological state change index of the batch of grain compared with the time of storage; the physical rheological state change index, combined with the current measured moisture and storage temperature data, is input into the second-level agent model and outputs the predicted quality index and deterioration rate of the batch of grain, generating quality evolution trend data, and updating the dynamic file of the batch of grain accordingly. The quality status determination and scheduling module is used to compare the quality evolution trend data with the preset quality warning threshold, determine the quality status of the batch of grain, and generate outbound priority adjustment instructions and / or diversion path control instructions based on the quality status. The multi-batch collaborative scheduling module is used to generate a comprehensive outbound sequence instruction based on the quality evolution trend data in the dynamic archives of each batch when multiple batches coexist, so as to drive the outbound equipment to execute.

Images