Agricultural product cold chain distribution intelligent monitoring method, equipment and medium

Abstract

The invention discloses an agricultural product cold chain distribution intelligent monitoring method and device and a medium, and relates to the technical field of intelligent monitoring, and the method comprises the steps: inputting a spatial-temporal feature matrix into a spatial-temporal attention prediction model, carrying out the local feature extraction of a sensor embedding layer, carrying out the time sequence modeling of a time sequence coding layer, and generating early warning information; performing multi-scene rehearsal on the early warning information to generate a virtual environment state set, and performing optimization solution on the virtual environment state set to obtain an optimal control strategy; based on the optimal control strategy, encryption transmission is carried out through 5G communication to generate an execution instruction set, and effect verification is carried out to generate a verification report; and performing abnormal rule analysis on the verification report, constructing an agricultural product cold chain abnormity early warning knowledge base, performing accurate risk prediction based on the agricultural product cold chain abnormity early warning knowledge base, and generating an intelligent monitoring report. Through a space-time attention prediction model and a multi-scene rehearsal optimization strategy, the robustness and reliability of cold chain distribution are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and in particular to an intelligent monitoring method, equipment and medium for cold chain distribution of agricultural products. Background Technology

[0002] As a core link in ensuring the quality and safety of fresh food, the development of intelligent monitoring technology is of great significance for reducing losses and improving supply chain efficiency in the cold chain distribution of agricultural products. Early cold chain monitoring mainly relied on manual inspections and single-point temperature sensors, which suffered from problems such as scattered data collection, poor real-time performance, and delayed risk warnings. With the rapid development of Internet of Things (IoT), big data, and artificial intelligence (AI) technologies, cold chain monitoring is gradually evolving towards multi-source sensor fusion, spatiotemporal feature mining, and intelligent decision-making. Among existing methods, cold chain monitoring based on multi-sensor networks can achieve the collection of multiple parameters such as temperature, humidity, and vibration.

[0003] However, existing technologies still have some shortcomings. On the one hand, the depth of multimodal data processing and spatiotemporal feature fusion is insufficient. In particular, the fusion of time-domain statistical features (such as temperature fluctuation variance) and spatial correlation features (such as temperature and humidity gradients between different containers) often adopts linear splicing or simple weighting, without fully considering the nonlinear coupling relationship between features, resulting in a weak ability to identify complex risks such as "sudden temperature rise + abnormal vibration". On the other hand, the scenario adaptability of strategy optimization is insufficient. Traditional optimization methods mostly generate control strategies based on fixed constraints (such as the upper limit of cooling power) and do not verify the generalization of the strategy through multi-scenario simulations (such as normal operating conditions, equipment hidden faults, and sudden external temperature drops), which easily leads to the phenomenon of "laboratory optimization, field failure". Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent monitoring method for cold chain distribution of agricultural products to solve the problems of insufficient depth of multimodal spatiotemporal feature fusion and poor adaptability of dynamic scenario strategy optimization.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent monitoring method for cold chain distribution of agricultural products, comprising, Collect and preprocess multi-source sensor data, and construct a spatiotemporal feature matrix based on the preprocessed multi-source sensor data; The spatiotemporal feature matrix is ​​input into the spatiotemporal attention prediction model, the sensor embedding layer performs local feature extraction, the temporal coding layer performs time series modeling, and the early warning information is generated. The system performs multi-scenario simulations of early warning information, generates a set of virtual environment states, and optimizes the solution of the virtual environment state set to obtain the optimal control strategy. Based on the optimal control strategy, encrypted transmission is performed via 5G communication to generate an execution instruction set, and the effect is verified to generate a verification report. Anomaly patterns are analyzed in the verification reports to construct an early warning knowledge base for agricultural product cold chain anomalies. Based on this knowledge base, accurate risk prediction is performed, and intelligent monitoring reports are generated.

[0007] As a preferred embodiment of the intelligent monitoring method for cold chain distribution of agricultural products described in this invention, the preprocessing includes data cleaning, format conversion, deduplication, normalization, and outlier handling.

[0008] As a preferred embodiment of the intelligent monitoring method for cold chain distribution of agricultural products described in this invention, the construction of the spatiotemporal feature matrix refers to extracting the temporal statistical features and spatial correlation features of the preprocessed multi-source sensor data and performing spatiotemporal feature fusion.

[0009] As a preferred embodiment of the intelligent monitoring method for cold chain distribution of agricultural products according to the present invention, the specific steps for generating early warning information are as follows: A spatiotemporal attention prediction model is constructed by using multimodal feature fusion to perform cross-domain feature alignment and dynamic weight allocation between the sensor embedding layer and the temporal coding layer. The spatiotemporal feature matrix is ​​input into the spatiotemporal attention prediction model, and the sensor embedding layer extracts local features through multi-scale convolutional kernels to form multi-dimensional monitoring features. The temporal coding layer uses a bidirectional gated recurrent network to model time series data and capture the dynamic trend characteristics of temperature and humidity changes. Multidimensional monitoring features and dynamic trend features of temperature and humidity changes are integrated through a feature fusion channel to generate early warning information.

[0010] As a preferred embodiment of the intelligent monitoring method for cold chain distribution of agricultural products according to the present invention, the specific steps for obtaining the optimal control strategy are as follows: Extract multi-dimensional environmental state variables from early warning information, conduct multi-scenario pre-simulation to construct a virtual environment state set, and use the constrained optimization method to adjust parameters based on the virtual environment state set to generate an initial control strategy set; The initial set of control strategies is validated to generate a set of feasible control schemes, and the results are ranked to generate a set of candidate optimization strategies. The optimal control strategy is obtained by crossover and mutation optimization of the candidate optimization strategy set through genetic algorithm.

[0011] As a preferred embodiment of the intelligent monitoring method for cold chain distribution of agricultural products according to the present invention, the specific steps for generating the verification report are as follows: Based on the optimal control strategy, the command perturbation is performed through 5G key negotiation to generate an encrypted command set, which is then input into the edge node for decryption and verification to generate an execution command set; The execution instruction set is format-validated to generate an instruction set to be verified, and the validity of the instruction set to be verified is validated to generate a verification report.

[0012] As a preferred embodiment of the intelligent monitoring method for cold chain distribution of agricultural products according to the present invention, the specific steps for generating the intelligent monitoring report are as follows: Extract abnormal patterns from the verification report, generate an abnormal feature set, and perform feature association and rule extraction based on the abnormal feature set to build an early warning knowledge base for agricultural product cold chain anomalies; Based on the knowledge base for early warning of abnormalities in the cold chain of agricultural products, risk factor association analysis is performed using association rule analysis to generate a risk association feature set; Assess the risk probability of a risk-related feature set and generate a risk prediction result set; The system filters key risk indicators from the risk prediction result set, generates an indicator dataset, transforms the indicator dataset into trend charts, and integrates them to generate an intelligent monitoring report.

[0013] Secondly, the present invention provides an intelligent monitoring system for cold chain distribution of agricultural products, including a data acquisition module, a risk prediction module, a strategy optimization module, an effect verification module, and a report generation module; The data acquisition module is used to collect multi-source sensor data and perform preprocessing, and to construct a spatiotemporal feature matrix based on the preprocessed multi-source sensor data; The risk prediction module is used to input the spatiotemporal feature matrix into the spatiotemporal attention prediction model, the sensor embedding layer extracts local features, the time-series coding layer performs time series modeling, and generates early warning information. The strategy optimization module is used to perform multi-scenario pre-simulation of early warning information, generate a virtual environment state set, and optimize the virtual environment state set to obtain the optimal control strategy. The effect verification module is used to generate an execution instruction set through encrypted transmission via 5G communication based on the optimal control strategy, and to perform effect verification and generate a verification report. The report generation module is used to analyze anomalies in verification reports, build a knowledge base for early warning of anomalies in the cold chain of agricultural products, and make accurate risk predictions based on the knowledge base to generate intelligent monitoring reports.

[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent monitoring method for cold chain distribution of agricultural products as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent monitoring method for cold chain distribution of agricultural products as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by using a spatiotemporal attention prediction model for local feature extraction and time series modeling, it achieves accurate identification of complex risk scenarios such as "sudden temperature rise + abnormal vibration", improving the accuracy and timeliness of risk warning. The multi-scenario pre-simulation optimization strategy used simultaneously can dynamically verify the control strategy through a virtual environment state set, improving the robustness and reliability of cold chain distribution. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for an intelligent monitoring method for cold chain distribution of agricultural products.

[0019] Figure 2 A schematic diagram of an intelligent monitoring system for cold chain distribution of agricultural products.

[0020] Figure 3 A flowchart generated for the optimal control strategy.

[0021] Figure 4 A flowchart for generating intelligent monitoring reports. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides an intelligent monitoring method for cold chain distribution of agricultural products, including the following steps: S1. Collect multi-source sensor data and preprocess it, and construct a spatiotemporal feature matrix based on the preprocessed multi-source sensor data; S1.1, Multi-source sensor data includes temperature data, humidity data, vibration data, positioning data, door status data, power supply voltage data, cooling power data, and airflow velocity data; It should be noted that temperature sensors are deployed to collect temperature data in real time in each area, humidity sensors monitor ambient humidity, triaxial accelerometers detect the X / Y / Z acceleration of the box vibration during transportation to obtain vibration data, GPS / BeiDou dual-mode positioning devices obtain real-time positioning data, door magnetic sensors monitor the door opening and closing status to obtain door status data, voltage sensors collect power supply voltage data, power meters measure the cooling power data of the refrigeration unit, and wind speed sensors monitor the airflow speed data inside the box.

[0026] S1.2 Preprocessing includes data cleaning, format conversion, deduplication, normalization, and outlier handling; It should be noted that, firstly, the collected multi-source sensor data, including temperature, humidity, vibration, positioning, door status, power supply voltage, cooling power, and airflow velocity data, undergoes data cleaning to remove incomplete, erroneous, or irrelevant data records. Next, format conversion is performed to ensure all data sources adhere to a unified format standard for subsequent processing. Then, deduplication is performed to eliminate duplicate data entries, ensuring the uniqueness of the dataset. A normalization step maps data values ​​from different sources to the same scale, preventing certain features from unduly affecting the results due to excessively large differences in magnitude. Finally, outlier handling identifies and corrects data points that deviate from the normal range, ensuring the accuracy and reliability of the dataset.

[0027] S1.3 Extract the temporal statistical features and spatial correlation features of the preprocessed multi-source sensor data, and perform spatiotemporal feature fusion to generate a spatiotemporal feature matrix; It should be noted that by acquiring seven types of statistics—mean, variance, maximum, minimum, root mean square, skewness, and kurtosis—from the preprocessed multi-source sensor data, a temporal feature vector reflecting local fluctuations and long-term trends is generated. Subsequently, based on the spatial layout of the sensors, the spatial gradient of similar multi-source sensor data at different locations and the spatial correlation of multiple sources at the same location, as well as regional extreme value differences, are obtained to generate a spatial correlation feature matrix reflecting spatial heterogeneity. Finally, the temporal feature vector and the spatial correlation feature matrix are concatenated column-wise through spatiotemporal alignment to generate a spatiotemporal feature matrix.

[0028] S2. Input the spatiotemporal feature matrix into the spatiotemporal attention prediction model, the sensor embedding layer performs local feature extraction, the time-series coding layer performs time series modeling, and generates early warning information. S2.1. Utilize multimodal feature fusion to perform cross-domain feature alignment and dynamic weight allocation on the sensor embedding layer and temporal coding layer to construct a spatiotemporal attention prediction model; It should be noted that the sensor embedding layer is constructed using the `tf.keras.layers.MultiHeadAttention` directive, and a self-attention mechanism is configured for the sensor embedding layer. This involves setting up 8 attention heads, each with a 64-dimensional dimension. A positional encoding layer is also added to the sensor embedding layer to capture the spatial location information of the sensor data, thus completing the construction of the sensor embedding layer. A temporal encoding layer is constructed by calling a bidirectional gated recurrent network using the `tf.keras.layers.RNN` directive, and a Dropout layer is configured for the temporal encoding layer to prevent overfitting. The number of bidirectional gated recurrent networks is set to 128, and the Dropout ratio is 0.2, thus completing the construction of the temporal encoding layer. Multimodal feature fusion is used to concatenate tensors between the sensor embedding layer and the temporal coding layer to generate a similarity score matrix. Bilinear interpolation is used to perform weighted fusion of the similarity score matrix, and the feature dimensions are stacked using the torch.stack parameter. A gating mechanism is used simultaneously to perform cross-layer residual connections, and GeLU activation function is added after the sensor embedding layer and the temporal coding layer to achieve nonlinear transformation. The spatiotemporal attention prediction model is constructed using the output of the Dropout layer. Next, the constructed spatiotemporal attention prediction model is trained. Specifically, the spatiotemporal feature matrix is ​​first divided into training, validation, and test sets in a 7:2:1 ratio. On the training set, the Adam optimizer is used to optimize the parameters of the spatiotemporal attention prediction model, with an initial learning rate of 0.001 and a learning rate decay strategy applied, decreasing the learning rate every 10 epochs. Early stopping is also used to prevent overfitting, with a patience value of 10. On the validation set, MAE and RMSE are used as evaluation metrics to monitor the performance of the spatiotemporal attention prediction model. When the validation loss no longer improves after 5 consecutive epochs, training is stopped, and the trained spatiotemporal attention prediction model is output using the `model.save_weights` method.

[0029] It should also be noted that multimodal feature fusion refers to a method of integrating and processing feature information from different sensors, modalities, or data sources. This involves extracting key features from each modality (such as temporal fluctuation features of temperature and frequency spectrum features of vibration), resolving spatiotemporal or semantic alignment issues between modalities, and employing methods such as weighted fusion, attention mechanisms, or multi-tensor joint modeling to transform scattered single-modal features into more complementary comprehensive features. The core objective is to overcome the limitations of single-modal information, fully utilize the synergistic correlation of different modal data, and improve the perception capability and representation accuracy of complex scenes.

[0030] S2.2 Input the spatiotemporal feature matrix into the spatiotemporal attention prediction model. The sensor embedding layer extracts local features through multi-scale convolution kernels to form multi-dimensional monitoring features. It should be noted that three different sizes of convolution kernels (such as 3×1, 5×1, and 7×1 convolution kernels in the time-feature plane dimension) are applied to the input spatiotemporal feature matrix for local feature extraction: each convolution kernel performs a weighted summation of the local regions of adjacent time steps and corresponding spatial locations within the current time window to capture local change patterns at short, medium, and long time scales; then, the output of each convolution kernel is subjected to nonlinear activation and feature standardization to eliminate dimensional differences and enhance feature stability; finally, the features after convolution at the three scales are concatenated along the feature dimension to form a composite feature containing local information at multiple time scales, and the composite feature is dimensionality reduced and integrated through pointwise convolution to form a multidimensional monitoring feature.

[0031] S2.3 The time-series coding layer uses a bidirectional gated recurrent network to model time series data and capture the dynamic trend characteristics of temperature and humidity changes. It should be noted that, from the input spatiotemporal feature matrix, temperature and humidity features of all spatial locations are integrated in time step order to extract time-dimensional sequence data and form a time series input. Then, a bidirectional gated recurrent network is initialized, containing two independent GRUs (gated recurrent networks) for forward and backward directions. The forward GRU processes the time series in time step order, capturing the dynamic evolution of temperature and humidity over time through a gating mechanism. The backward GRU processes the time series in reverse order, capturing potential changes in the reverse direction of time (such as the recent cooling rate and the initial features of sudden humidity changes). The two GRUs output hidden states at each time step, and the forward and backward hidden states are concatenated and fused to comprehensively reflect the changes in temperature and humidity in the forward and reverse time dimensions. Finally, a linear transformation is used to map the concatenated hidden states into a low-dimensional vector, generating dynamic trend features containing the dynamic trends of temperature and humidity.

[0032] S2.4 Integrate multi-dimensional monitoring features and dynamic trend features of temperature and humidity changes through the feature fusion channel to generate early warning information; It should be noted that in the feature fusion channel, a unified time reference protocol is used to perform bidirectional linear resampling alignment of the time dimensions of multidimensional monitoring features and dynamic trend features to eliminate temporal phase differences. At the same time, spatial positions are unified to the same coordinate system through coordinate mapping to eliminate spatial deviations and ensure strict correspondence between spatiotemporal dimensions. Next, a dual-branch feature fusion mechanism is used for integration: a spatial attention layer is applied to the multi-dimensional monitoring features to extract the feature responses of key spatial regions (such as areas with high incidence of temperature anomalies), and a temporal convolution kernel is used to extract local temporal patterns for dynamic trend features. Then, these are concatenated along the channel dimension to generate a hybrid feature tensor. Finally, the hybrid feature tensor is subjected to element-wise nonlinear transformation using the GeLU activation function to enhance feature expression. Then, the dimension is increased to the warning category dimension through a fully connected layer to output the original score of each warning level. Subsequently, the original score is exponentially normalized using the Softmax function to generate the warning level probability distribution. The probability distribution of warning levels is divided based on the preset three-level warning thresholds: when the probability distribution of warning levels falls within the first-level threshold range, it is marked as the highest warning level; when the probability distribution of warning levels falls within the second-level threshold range, it is marked as the medium warning level; when the probability distribution of warning levels falls within the third-level threshold range, it is marked as the low warning level; finally, the different warning levels are integrated according to priority to generate complete warning information.

[0033] It should also be noted that the three-level early warning thresholds are defined based on the degree of impact of business risks and historical data statistics. For example, the values ​​are: Level 1 threshold range [0.8, 1.0], Level 2 threshold range [0.5, 0.8], and Level 3 threshold range [0, 0.5].

[0034] S3. Perform multi-scenario pre-simulation of the early warning information to generate a virtual environment state set, and optimize the generated virtual environment state set to obtain the optimal control strategy; S3.1 Extract multi-dimensional environmental state variables from early warning information, conduct multi-scenario pre-drills to construct a virtual environment state set, and use the constraint optimization method to adjust parameters based on the virtual environment state set to generate an initial control strategy set; It should be noted that a comprehensive simulation is conducted for typical scenarios that may occur during cold chain transportation, such as normal transportation status, equipment failure status, and extreme environmental interference, to construct a virtual environment state set that can cover various risk conditions. Subsequently, pre-processed multi-source sensor data and early warning information are integrated to extract multi-dimensional environmental state variables, including the current temperature value, humidity gradient, vibration amplitude, positioning offset, time-dimension characteristics, and spatial location. Spatiotemporal alignment and standardization methods are used to merge the virtual environment state and multi-dimensional environmental state variables into a unified multi-dimensional virtual environment state set. Finally, with the objective function of reducing the risk of cargo loss, the physical performance limitations of cold chain equipment (such as the power limit and temperature regulation rate of refrigeration equipment) and the special requirements of cargo transportation (such as the temperature control sensitivity of different types of goods) are comprehensively considered. Constraint optimization methods are used to search for control combinations suitable for different scenarios to generate an initial control strategy set.

[0035] S3.2. Perform rule verification on the initial control strategy set to generate a set of feasible control schemes, and sort the results to generate a set of candidate optimization strategies. It should be noted that, firstly, the initial control strategy set is traversed, and illegal control strategies are screened out through logical judgments (such as "adjusted power ≤ equipment upper limit" and "temperature ≥ minimum safe value after execution") to generate a set of feasible control schemes that meet the rules; then, the effect of the set of feasible control schemes is evaluated, and the risk reduction, energy consumption increment and environmental stability of the set of feasible control schemes are obtained based on simulation. The schemes are then weighted and sorted according to multiple indicators of "risk-energy consumption-stability" to generate a set of candidate optimization strategies.

[0036] It should also be noted that the effect evaluation of the set of feasible control schemes specifically refers to: applying the control parameters (such as cooling power and fan start-up and shutdown time) of each feasible scheme to historical transportation data (such as the same route and temperature and humidity fluctuation records), simulating the execution, and statistically analyzing the actual corrosion rate, equipment energy consumption, and temperature and humidity fluctuation range; comparing the baseline situation without adopting the scheme (such as the corrosion rate and energy consumption controlled according to the original rules), obtaining the risk reduction, energy consumption increment, and environmental stability, and quantifying the actual effect of each scheme.

[0037] S3.3. The optimal control strategy is obtained by performing crossover and mutation optimization on the candidate optimization strategy set through genetic algorithm; It should be noted that the candidate optimization strategy set is initialized as the initial population of the genetic algorithm, with each candidate optimization strategy as an individual, and the strategy parameters are represented by real numbers. Then, the fitness of each individual is obtained, and the fitness is a comprehensive score of "risk reduction rate - energy consumption increment". Based on the fitness, a selection operation (such as roulette wheel selection or tournament selection) is performed, and individuals with high fitness are retained as parents. A crossover operation is performed on the parent individuals, exchanging some strategy parameters to generate offspring strategies. A mutation operation is performed on the offspring and parent individuals (such as applying a small-range random perturbation to the cooling power adjustment). After mutation, it is checked whether the candidate optimization strategy meets the equipment constraints and safety constraints. If not, the parameters are adjusted through a repair function. The "selection-crossover-mutation-evaluation-constraint check" process is repeated until the termination condition is reached (such as the number of iterations reaching 100). Finally, the individual with the highest fitness is selected as the optimal control strategy.

[0038] It should also be noted that equipment constraints are defined based on the physical performance parameters (such as the upper limit of refrigeration power) and operational limitations (such as the time interval between single start-stop operations) of cold chain equipment (such as refrigeration units). Safety constraints are defined based on cargo quality assurance (such as upper humidity limits) and operational safety (such as vibration amplitude limits).

[0039] S4. Based on the optimal control strategy, generate an execution instruction set through encrypted transmission using 5G communication, and verify the effect to generate a verification report; S4.1 Based on the optimal control strategy, the command perturbation is performed through 5G key negotiation to generate an encrypted command set, which is then input into the edge node for decryption and verification to generate an execution command set; It should be noted that the optimal control strategy is encapsulated into structured instructions according to the device communication protocol, specifying the instruction type, target device, parameter value, and execution timestamp. Then, the AES-256 symmetric encryption algorithm is used with a 256-bit key to perform multiple rounds of encryption transformations on the structured instructions, including byte substitution, row shifting, column obfuscation, and round key addition, generating an encrypted instruction set. Next, the encrypted instruction set is encapsulated into encrypted data packets carrying device identifiers, timestamps, and other information via 5G communication, and transmitted to the edge node using the low latency characteristics of the 5G network. After receiving the encrypted data packets via the 5G network, the edge node uses a shared key to decrypt them, verifying the integrity and legality of the structured instructions. Finally, the decrypted structured instructions are parsed to extract key information such as device ID, operation type, and parameter values, and converted into corresponding binary or hexadecimal execution instructions according to the device communication protocol. Simultaneously, the required checksum is added, ultimately generating the execution instruction set.

[0040] S4.2 Perform format validation on the execution instruction set to generate an instruction set to be verified, and perform validity validation on the instruction set to be verified to generate a validation report; It should be noted that, in accordance with the format specifications of the device communication protocol, the structural integrity, data type legality, and parameter validity of each instruction set are checked, and instruction sets with format errors or non-compliance are filtered out to generate a set of instructions to be verified. Subsequently, the set of instructions to be verified is sent to the simulation environment for execution, and the execution feedback, the measured values ​​of temperature and humidity after execution, the abnormal alarm codes, and the execution time are recorded in real time. Finally, the expected goals of the executed instructions are compared with the execution feedback to obtain the validity, latency, and stability of the executed instructions, and a verification report is generated.

[0041] It should also be noted that the format specifications of the device communication protocol include the required fields "Device ID", "Operation Type", parameter value range, and data type; The expected goal of executing instructions is based on the objective function definition of the initial control strategy.

[0042] S5. Analyze the abnormal patterns in the verification report to build an agricultural product cold chain anomaly early warning knowledge base, and generate intelligent monitoring reports based on the generated agricultural product cold chain anomaly early warning knowledge base for accurate risk prediction.

[0043] S5.1 Extract abnormal patterns from the verification report, generate an abnormal feature set, and perform feature association and rule extraction based on the abnormal feature set to build an agricultural product cold chain abnormal early warning knowledge base; It should be noted that statistical analysis is performed on the execution feedback in the verification report. Anomaly features such as anomaly type (e.g., "temperature exceeds limit"), anomaly parameters (e.g., "temperature 32℃"), and associated environmental variables (e.g., "humidity 85%) are extracted through anomaly classification rules to generate anomaly feature set. Subsequently, potential correlations between anomaly features are explored through association rule mining, and rule filtering is performed in combination with preset equipment operating parameters to generate rule set. Finally, the anomaly feature set and rule set are integrated into an agricultural product cold chain anomaly early warning knowledge base.

[0044] It should also be noted that the process of identifying potential associations between abnormal features through association rule mining is as follows: construct a co-occurrence matrix of abnormal feature set and preprocessed multi-source sensor data based on transactional data format; by setting a minimum support threshold, filter out frequent itemsets that meet the minimum support (occurrence frequency) and generate association rules (such as "temperature > 30℃ → fan stop probability increases by 2.8 times" and "humidity > 85% and door open for > 5 minutes → condensate alarm probability reaches 92%"). The transactional data format is based on the time sensitivity of key monitoring parameters such as temperature, humidity, vibration, and equipment status in cold chain transportation scenarios, as well as the definition of equipment operation events. The minimum support threshold is defined based on the co-occurrence frequency in the cold chain transportation scenario, with an exemplary value range of 3% to 15%. The anomaly classification rules are based on the statistical definition of historical anomaly cases; The preset equipment operating parameters are defined based on the equipment's rated performance and critical conditions for safe operation.

[0045] S5.2 Based on the knowledge base for early warning of abnormalities in the cold chain of agricultural products, risk factor association analysis is performed using association rule analysis to generate a risk association feature set; It should be noted that, based on the knowledge base for early warning of abnormalities in the cold chain of agricultural products, multi-dimensional risk factor data such as environmental parameters, equipment status, operation records and abnormal tags are extracted, and after cleaning and standardization, a structured dataset is formed. Then, the Apriori algorithm is used to mine association rules, output risk factor combinations, and generate a risk association feature set.

[0046] It should also be noted that using the Apriori algorithm to mine association rules in a structured dataset requires scanning the structured dataset, obtaining the support of each individual item (such as "high humidity" and "door open for more than 5 minutes"), filtering out frequent itemsets that meet the minimum support, and generating basic risk factor combinations. Then, iterate through all frequent itemsets to generate candidate association rules and obtain the confidence of the candidate association rules (the conditional probability of the rule being true). Retain the candidate association rules that meet the minimum confidence, and finally output strongly associated risk factor combinations (such as "high humidity + door open for more than 5 minutes") and candidate association rules, integrating them to generate a risk association feature set.

[0047] S5.3 Evaluate the risk probability of the risk-related feature set and generate a risk prediction result set; It should be noted that the number of times each risk factor combination in the risk association feature set has triggered a warning in history is extracted, and the risk probability is obtained by the ratio of the number of historical warnings to the total number of occurrences; the risk probability, warning information and risk association feature set are integrated to generate a risk prediction result set.

[0048] S5.4 Filter the key risk indicators in the risk prediction result set, generate an indicator dataset, and transform the indicator dataset into trend charts to generate an intelligent monitoring report.

[0049] It should be noted that, based on preset statistical characteristics, quantitative indicators such as risk probability, warning level, and frequency of occurrence of risk factor combinations are extracted from the risk prediction results set to generate an indicator dataset. Subsequently, the indicators are classified and processed according to their types: time series indicators (such as risk probability) are organized into "time-value" series, frequency distribution indicators (such as factor combination frequency) are organized into "factor combination-frequency" grouped data, and correlation indicators (such as correlation strength) are organized into "factor A-factor B-correlation coefficient" matrices. Finally, visualization methods are selected according to the classification results: time series indicators are displayed using line charts to show fluctuations along the time axis, frequency distribution indicators are displayed using bar charts to show frequency by factor combination, and correlation indicators are displayed using heatmaps with varying color intensity to show the strength of the correlation, generating a set of charts that intuitively reflect risk trends, and trend charts are generated. Finally, the indicator dataset and trend charts are integrated into an intelligent monitoring report.

[0050] It should also be noted that the preset statistical characteristics are based on the actual distribution definitions of the quantified fields, such as the mean and standard deviation.

[0051] This embodiment also provides an intelligent monitoring system for cold chain distribution of agricultural products, including: a data acquisition module, a risk prediction module, a strategy optimization module, an effect verification module, and a report generation module; The data acquisition module is used to collect multi-source sensor data and perform preprocessing, and to construct a spatiotemporal feature matrix based on the preprocessed multi-source sensor data; The risk prediction module is used to input the spatiotemporal feature matrix into the spatiotemporal attention prediction model, the sensor embedding layer extracts local features, the time-series coding layer performs time series modeling, and generates early warning information. The strategy optimization module is used to perform multi-scenario pre-simulation of early warning information, generate a virtual environment state set, and optimize the virtual environment state set to obtain the optimal control strategy. The effect verification module is used to generate an execution instruction set through encrypted transmission via 5G communication based on the optimal control strategy, and to perform effect verification and generate a verification report. The report generation module is used to analyze anomalies in verification reports, build a knowledge base for early warning of anomalies in the cold chain of agricultural products, and make accurate risk predictions based on the knowledge base to generate intelligent monitoring reports.

[0052] This embodiment also provides a computer device applicable to the intelligent monitoring method for cold chain distribution of agricultural products, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent monitoring method for cold chain distribution of agricultural products as proposed in the above embodiment.

[0053] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0054] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent monitoring method for cold chain distribution of agricultural products as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0055] In summary, this invention achieves accurate identification of complex risk scenarios such as "sudden temperature rise + abnormal vibration" through local feature extraction and time series modeling using a spatiotemporal attention prediction model, thereby improving the accuracy and timeliness of risk warnings. The simultaneously used multi-scenario pre-simulation optimization strategy enables dynamic verification of control strategies through a virtual environment state set, enhancing the robustness and reliability of cold chain delivery.

[0056] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent monitoring of cold chain distribution of agricultural products, characterized in that: The application relates to a cold chain risk prediction method based on a multi-source sensor data fusion model. The method comprises the following steps: collecting and preprocessing multi-source sensor data, constructing a space-time feature matrix based on the preprocessed multi-source sensor data, inputting the space-time feature matrix into a space-time attention prediction model, performing local feature extraction through a sensor embedding layer, performing time series modeling through a time sequence coding layer, generating early warning information, performing multi-scene simulation on the early warning information, generating a virtual environment state set, optimizing and solving the virtual environment state set to obtain an optimal control strategy, generating an execution instruction set through 5G communication based on the optimal control strategy, and generating a verification report through effect verification. The preprocessing comprises data cleaning, format conversion, deduplication, normalization and abnormal value processing. The construction of the space-time feature matrix refers to the extraction of time domain statistical features and space correlation features of the preprocessed multi-source sensor data, and space-time feature fusion. The generation of the early warning information comprises the following steps: A space-time attention prediction model is constructed by using multi-modal feature fusion to perform cross-domain feature alignment and dynamic weight distribution on the sensor embedding layer and the time sequence coding layer. 2.The intelligent monitoring method for agricultural product cold chain distribution according to claim 1, characterized in that: The space-time feature matrix is input into the space-time attention prediction model, the sensor embedding layer performs local feature extraction through a multi-scale convolution kernel, and multi-dimensional monitoring features are formed. 3.The intelligent monitoring method for agricultural product cold-chain distribution according to claim 2, characterized in that: The time sequence coding layer performs time series modeling through a bidirectional gated recurrent network to capture dynamic trend features of temperature and humidity changes.

4. The intelligent monitoring method for agricultural product cold chain distribution according to claim 3, characterized in that: The multi-dimensional monitoring features and the dynamic trend features of temperature and humidity changes are integrated through a feature fusion channel to generate early warning information. The optimal control strategy is obtained through the following steps: Multi-dimensional environment state variables of the early warning information are extracted, a virtual environment state set is constructed through multi-scene simulation, and an initial regulation and control strategy set is generated through parameter adjustment based on the virtual environment state set by using a constraint optimization method. Rule checking is performed on the initial regulation and control strategy set to generate a feasible control scheme set, and effect sorting is performed to generate a candidate optimization strategy set. The candidate optimization strategy set is optimized through genetic algorithm crossover and mutation to obtain the optimal control strategy.

5. The intelligent monitoring method for agricultural product cold chain distribution according to claim 4, characterized in that: The verification report is generated through the following steps: Based on the optimal control strategy, an encrypted instruction set is generated through 5G key negotiation for instruction disturbance, and the execution instruction set is generated by inputting the encrypted instruction set into an edge node for decryption verification. The format of the execution instruction set is checked to generate a to-be-verified instruction set, and the to-be-verified instruction set is subjected to validity verification to generate a verification report. The intelligent monitoring report is generated through the following steps:

6. The intelligent monitoring method for agricultural product cold chain distribution according to claim 5, characterized in that: Abnormal rules in the verification report are extracted to generate an abnormal feature set, and feature correlation and rule extraction are performed according to the abnormal feature set to construct an agricultural product cold chain abnormal early warning knowledge base. Based on the agricultural product cold chain abnormal early warning knowledge base, risk factor correlation analysis is performed through an association rule analysis method to generate a risk correlation feature set. The risk probability of the risk correlation feature set is evaluated to generate a risk prediction result set.

7. The intelligent monitoring method for agricultural product cold chain distribution according to claim 6, characterized in that: Key risk indicators of the risk prediction result set are screened to generate an index data set, the index data set is converted into a trend chart, and the intelligent monitoring report is integrated. ​ ​ ​ ​ 8. An intelligent monitoring system for agricultural product cold chain distribution, based on the intelligent monitoring method for agricultural product cold chain distribution according to any one of claims 1-7, characterized in that: The method comprises a data acquisition module, a risk prediction module, a strategy optimization module, an effect verification module, and a report generation module. The data acquisition module is configured to acquire multi-source sensor data and perform preprocessing, and construct a space-time feature matrix based on the preprocessed multi-source sensor data. The risk prediction module is configured to input the space-time feature matrix into a space-time attention prediction model, perform local feature extraction through a sensor embedding layer, and perform time series modeling through a time series encoding layer to generate early warning information. The strategy optimization module is configured to perform multi-scenario rehearsal on the early warning information, generate a virtual environment state set, and optimize and solve the virtual environment state set to obtain an optimal control strategy. The effect verification module is configured to generate an execution instruction set through encrypted transmission based on the optimal control strategy through 5G communication, and perform effect verification to generate a verification report. The report generation module is configured to perform abnormality rule analysis on the verification report, construct an agricultural product cold chain abnormality early warning knowledge base, and perform precise risk prediction based on the agricultural product cold chain abnormality early warning knowledge base to generate an intelligent monitoring report. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the agricultural product cold chain distribution intelligent monitoring method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the agricultural product cold chain distribution intelligent monitoring method of any one of claims 1-7.

Images