A postharvest preservation remote intelligent control system and control method for vegetables

Abstract

The present application relates to the field of postharvest preservation control technology of vegetables, in particular to a kind of postharvest preservation remote intelligent control system and control method of vegetables, comprising the following steps: S1, the physiological information of vegetables in each local area is collected;S2, normalized physiological characteristic vector is generated;S3, physiological activity index distribution map is generated;S4, corruption propagation potential is generated;S5, risk level diagram is generated;S6, risk level diagram and goods value information are input into net economic value decision model, and the optimal intervention action vector is generated;S7, the optimal intervention action vector is analyzed, and hardware control instruction is generated;S8, the hardware control instruction is executed by driving bottom hardware;S9, after intervention execution, control effect is quantified, and net economic value decision model is updated online using control effect.The present application realizes accurate perception and evaluation of the state of vegetables, and solves the problem that environment control is disconnected with the actual physiological state of vegetables in traditional preservation technology.

Description

Technical Field

[0001] This invention relates to the field of postharvest preservation control technology for vegetables, specifically a remote intelligent control system and control method for postharvest preservation of vegetables. Background Technology

[0002] In the current post-harvest supply chain management of vegetables, maintaining product quality and extending shelf life are of paramount importance. To ensure the freshness of vegetables during storage or transportation, the industry generally relies on the control of macroscopic physical quantities of the environment, such as temperature and humidity.

[0003] However, traditional preservation technologies often separate environmental control from the actual, localized physiological state of vegetables, resulting in a lack of targeted control strategies. This extensive management approach struggles to address the heterogeneity of conditions within storage units, particularly failing to effectively identify and intervene in spoilage processes that originate locally and spread cascadingly, leading to significant economic losses. While some existing technologies incorporate sensors to monitor environmental or physiological indicators of vegetables, they generally suffer from insufficient predictive capabilities for future spoilage risks. These methods largely remain at a passive assessment of the current state, failing to establish a dynamic predictive model that integrates the current state, temporal trends, and spatial diffusion effects, thus hindering early warnings and preventative interventions. Furthermore, existing control decision-making systems often prioritize optimizing technical indicators, neglecting the execution costs of intervention measures, such as energy consumption, and the economic trade-off between these costs and the loss of product value due to preservation failures. This makes it difficult for the system to find the optimal balance between preservation effectiveness and operating costs, limiting the maximization of commercial value.

[0004] Therefore, how to provide a post-harvest preservation method for vegetables that can integrate multi-dimensional physiological information of vegetables, accurately predict the risk of spoilage and spread, and carry out closed-loop adaptive control with the goal of maximizing net economic value is a problem that urgently needs to be solved by those skilled in the art.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To solve the above-mentioned technical problems, this invention discloses a remote intelligent control system and control method for post-harvest preservation of vegetables. Specifically, the technical solution of this invention is as follows:

[0007] A remote intelligent control method for post-harvest preservation of vegetables includes the following steps:

[0008] S1. Collect vegetable physiological information from various local areas, including hyperspectral data, ethylene concentration, and carbon dioxide concentration;

[0009] S2. Based on the physiological information of vegetables, generate normalized physiological feature vectors;

[0010] S3. Input the physiological feature vector into the physiological activity index model to generate a physiological activity index distribution map;

[0011] S4. Input the distribution map of physiological activity indicators into the putrefaction propagation potential prediction model to generate the putrefaction propagation potential.

[0012] S5. Based on the preset risk threshold, assess the level of corruption transmission potential and generate a risk level map.

[0013] S6. Input the risk level map and cargo value information into the net economic value decision model to generate the optimal intervention action vector;

[0014] S7. Analyze the optimal intervention action vector and generate hardware control instructions;

[0015] S8 drives the underlying hardware to execute hardware control instructions;

[0016] S9. After the intervention is implemented, quantify the control effect and use the control effect to update the net economic value decision model online.

[0017] Preferably, S3 specifically includes:

[0018] S31. Retrieve weighting coefficients from the preset expert database;

[0019] S32. The physiological feature vectors are weighted and summed based on the weight coefficients to calculate the dimensionless physiological activity index.

[0020] S33. Combine the physiological activity indicators of each local area to generate a distribution map of physiological activity indicators.

[0021] Preferably, S4 specifically includes:

[0022] S41. Obtain the current physiological activity of the region;

[0023] S42. Calculate the rate of change of physiological activity over time;

[0024] S43, Accumulate the spatial diffusion effect of neighboring areas;

[0025] S44. Based on the preset static weights, dynamic weights, and spatial diffusion effect weights, the physiological activity, time change rate, and spatial diffusion effect are weighted and fused to calculate the putrefaction propagation potential.

[0026] Preferably, S5 specifically includes:

[0027] S51. Set early warning thresholds and danger thresholds as risk thresholds;

[0028] S52. Areas where the potential for corruption to spread is lower than or equal to the warning threshold shall be classified as safety levels.

[0029] S53. Areas where the potential for corruption to spread is higher than the warning threshold but lower than or equal to the danger threshold shall be classified as warning levels.

[0030] S54. Areas where the potential for corruption to spread exceeds the danger threshold are classified as danger levels.

[0031] Preferably, S6 specifically includes:

[0032] S61. For areas under warning or danger, iterate through the set of preset intervention actions;

[0033] S62. For any intervention action in the set, calculate the execution cost;

[0034] S63. Call a pre-trained machine learning model to predict the expected residual loss cost after taking intervention actions;

[0035] S64. Subtract execution costs and expected residual loss costs from the current total market value of the goods to obtain the expected net economic value;

[0036] S65. Select the intervention action that maximizes the expected net economic value as the optimal intervention action vector.

[0037] Preferably, the hardware control instructions include intervention type parameters, intervention intensity parameters, duration parameters, and area of ​​action identification parameters.

[0038] Preferably, S9 specifically includes:

[0039] S91. After the intervention is completed, regenerate the risk level map;

[0040] S92. Compare the risk level charts before and after the intervention, and quantify the risk changes as a measure of control effectiveness;

[0041] S93. Use the control effect as feedback to update the machine learning model in the net economic value decision model used to predict expected residual loss costs.

[0042] A remote intelligent control system for post-harvest preservation of vegetables includes:

[0043] The physiological information processing module is used to collect the physiological information of vegetables in various local areas and generate normalized physiological feature vectors.

[0044] The physiological state modeling module is used to receive physiological feature vectors and generate a distribution map of physiological activity indicators.

[0045] The risk prediction and assessment module is used to receive the distribution map of physiological activity indicators, generate the putrefaction transmission potential, and assess and generate a risk level map based on the preset risk threshold.

[0046] The economic decision-making module receives the risk level map and combines it with cargo value information to generate the optimal intervention action vector.

[0047] The intervention execution module is used to parse the optimal intervention action vector into hardware control instructions and drive the underlying hardware to execute them.

[0048] The adaptive optimization module is used to quantify the control effect after the intervention is implemented and to update the economic decision-making module online using the control effect.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. This invention enables precise perception and assessment of the state of vegetables, solving the problem of disconnect between environmental control and the actual physiological state of vegetables in traditional preservation technology; by directly collecting multi-dimensional physiological information such as hyperspectral data and ethylene concentration, and shifting the control focus from indirect environmental physical quantities to direct vegetable physiological state, the control strategy can more realistically reflect the vital signs of vegetables;

[0051] 2. This invention enables dynamic prediction and early warning of corruption risks, solving the problem that existing technologies cannot accurately predict and intervene in cascading corruption; by constructing a corruption propagation potential prediction model, it innovatively integrates the current state of physiological activity, temporal change trends, and spatial diffusion effects, achieving a leap from static assessment to dynamic trend prediction, and gaining a time window for preventive intervention.

[0052] 3. This invention achieves the optimization of the economic value of preservation decisions and solves the contradiction between energy saving and preservation effect in traditional control. By introducing a net economic value decision model with the ultimate goal of maximizing commercial value, it intelligently weighs the implementation cost of intervention measures against the potential loss of goods, thereby effectively controlling the risk of spoilage while avoiding energy waste caused by excessive intervention.

[0053] 4. This invention achieves self-adaptation and self-optimization of the control system, solving the problem of traditional system models being fixed and unable to adapt to dynamic changes. By constructing a closed loop for quantifying the effects of intervention and online learning, the system can continuously update its internal decision-making model using actual control effects, enabling its decision-making level to continuously improve over long-term operation to adapt to the ever-changing cargo status and external environment. Attached Figure Description

[0054] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0055] Figure 1This is a flowchart of the method of the present invention;

[0056] Figure 2 This is a structural block diagram of the system of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0058] Example 1:

[0059] Please see Figure 1 A remote intelligent control method for post-harvest preservation of vegetables, characterized by the following steps:

[0060] S1. Collect vegetable physiological information from various local areas, including hyperspectral data, ethylene concentration, and carbon dioxide concentration;

[0061] S2. Based on the physiological information of vegetables, generate normalized physiological feature vectors;

[0062] S3. Input the physiological feature vector into the physiological activity index model to generate a physiological activity index distribution map;

[0063] S4. Input the distribution map of physiological activity indicators into the putrefaction propagation potential prediction model to generate the putrefaction propagation potential.

[0064] S5. Based on the preset risk threshold, assess the level of corruption transmission potential and generate a risk level map.

[0065] S6. Input the risk level map and cargo value information into the net economic value decision model to generate the optimal intervention action vector;

[0066] S7. Analyze the optimal intervention action vector and generate hardware control instructions;

[0067] S8 drives the underlying hardware to execute hardware control instructions;

[0068] S9. After the intervention is implemented, quantify the control effect and use the control effect to update the net economic value decision model online.

[0069] This embodiment provides a remote intelligent control method for post-harvest preservation of vegetables. The method constructs a complete closed-loop control system from physiological state perception to optimal economic value intervention, aiming to solve the problems in traditional preservation technology such as the disconnect between environmental control and the actual physiological state of vegetables, the inability to accurately predict and intervene in cascading spoilage, and the difficulty in balancing energy saving and preservation effect.

[0070] The specific process of this method is as follows:

[0071] S1. Collecting vegetable physiological information from various local areas; vegetable physiological information refers to a multi-dimensional data set that can directly or indirectly reflect the state of vegetable life activities, and its role is to provide original and objective input for subsequent physiological state modeling; in this embodiment, information is collected through a sensor network deployed in the internal partitions of the preservation unit, such as a cold storage or container; specifically, a hyperspectral camera is used to collect the reflectance spectrum data of the vegetable surface, which is used to reflect the external state of the vegetable skin, such as color, damage or early lesions; an electrochemical gas sensor is used to monitor the ethylene concentration in the local environment in real time, which is a key indicator for measuring the vegetable ripening catalytic process; at the same time, an infrared gas sensor is used to monitor the carbon dioxide concentration, which is used to characterize the intensity of vegetable respiration; these sensors together constitute a three-dimensional, multi-source perception of the vital signs of vegetables in various local areas;

[0072] Before normalization, the system also includes a data verification sub-step, which is responsible for checking whether the raw data collected by the sensor is within a reasonable physical range and smoothing, interpolating or removing obvious outliers or missing values ​​to ensure the validity and reliability of the physiological information input into the subsequent model and enhance the robustness of the entire system.

[0073] S2. Based on the physiological information of vegetables, a normalized physiological feature vector is generated. This step converts raw data from different sensors, with different dimensions and physical meanings, into a unified and standardized format that can be mathematically modeled. In this embodiment, the hyperspectral data, ethylene concentration, and carbon dioxide concentration collected in S1 are normalized. The normalization adopts the maximum-minimum difference method, that is, based on the physiological limit values ​​that a specific vegetable variety may encounter during the entire typical shelf life, including the maximum and minimum values, the real-time readings of each sensor are linearly mapped to the [0,1] interval. Thus, the physiological information of each local region is integrated into a normalized physiological feature vector, for example... ,in , , These represent the normalized ethylene concentration, carbon dioxide concentration, and hyperspectral anomaly characteristic value, respectively.

[0074] S3. Input the physiological feature vector into the physiological activity index model to generate a physiological activity index distribution map; this step integrates multidimensional and heterogeneous physiological features into a single scalar index that can directly and macroscopically characterize the physiological activity of vegetables.

[0075] In this embodiment, the physiological activity index model is a linear weighted fusion model; through this model, the physiological feature vector generated by S2 is calculated into a dimensionless physiological activity index. Physiological activity index refers to a comprehensive value that quantifies the activity level of vegetable vital signs. The higher the value, the more intense the metabolic or decay process of the vegetable in that area. This value is calculated from all local areas within the preservation unit. By combining these values, a visual spatial distribution map of physiological activity indicators can be generated, which intuitively shows the distribution of the strength of vegetable life activities throughout the entire preservation space.

[0076] S4. Input the distribution map of physiological activity indicators into the putrefaction propagation potential prediction model to generate the putrefaction propagation potential. This step not only satisfies the assessment of the current state, but also aims to predict the dynamic evolution trend of putrefaction risk over a future period, especially the diffusion impact of highly active areas on surrounding areas. In this embodiment, the putrefaction propagation potential prediction model is a comprehensive prediction model that integrates the current state, temporal change trend, and spatial diffusion effect. This model uses the distribution map of physiological activity indicators generated in the previous step as the core input to calculate the potential of each local area over a future time step. The spread of corruption afterward The potential for corruption to spread is a forward-looking, dimensionless risk indicator that quantifies the likelihood and intensity of a region becoming a source of corruption and spreading outwards.

[0077] S5. Assess the potential for corruption transmission based on preset risk thresholds and generate a risk level map; in order to continuously... The numerical values ​​are converted into discrete risk levels that facilitate decision-making. This step involves threshold judgment. In this embodiment, two key risk thresholds are preset: a warning threshold and a warning threshold. and a danger threshold Based on these two thresholds, the system will analyze all local areas. The risk levels are assessed and classified into three categories: safe, warning, and dangerous. The assessment results of all areas ultimately form a risk level map, which serves as the direct basis for subsequent control decision-making systems to make intervention judgments.

[0078] S6. Input the risk level map and cargo value information into the net economic value decision model to generate the optimal intervention action vector. This step transforms the control decision from the traditional pursuit of optimal physical indicators to maximizing commercial economic value as the optimization objective. In this embodiment, the net economic value decision model is a cost-benefit analysis model. It receives the risk level map generated in S5 and combines it with pre-input cargo value information, such as the procurement cost of goods in each region and the current market price, to systematically evaluate a series of preset intervention actions for all regions in a warning or dangerous state. By calculating the execution cost of each alternative action and the expected residual loss cost after execution, the model finally outputs an optimal intervention action vector that maximizes the expected net economic value. ;

[0079] S7. Analyze the optimal intervention action vector and generate hardware control instructions; this step translates the high-level decision output in S6 into specific commands that the underlying hardware can understand and execute; in this embodiment, the optimal intervention action vector... It is itself a structured data packet containing all the parameters required to perform the intervention; the parsing process is to extract specific information such as intervention type, intensity, duration and area of ​​effect from the vector and map it into hardware control instructions that conform to the hardware communication protocol.

[0080] S8. Drive the underlying hardware to execute hardware control instructions; this step is the physical execution link of the control closed loop; in this embodiment, the hardware control instructions generated in S7 are sent to the hardware controller in the preservation unit, such as a programmable logic controller; the controller precisely drives the corresponding actuators according to the instructions, such as the ventilation fan in a specific area, the solenoid valve of the refrigeration coil, or the ethylene removal device, so as to achieve precise physical intervention in the risk area.

[0081] S9. After the intervention is implemented, the control effect is quantified, and the net economic value decision model is updated online using the control effect. To enable the system to have adaptive and self-optimizing capabilities, a feedback mechanism is constructed in this step. After the intervention action in S8 is completed, the system control flow automatically returns to S1, re-collects data, and generates a new risk level map after the intervention. By comparing the changes in the risk level map before and after the intervention, the system can quantify the control effect. This effect data will be used as a new training sample to update the machine learning model in S6 used to predict the expected residual loss cost online, thereby making the system's decisions increasingly accurate.

[0082] Through the aforementioned steps, this invention constructs a complete intelligent control method that integrates physiological information perception with closed-loop intervention for optimal economic value. Compared to existing technologies, it shifts the control focus from indirect environmental physical quantities to the direct physiological state of vegetables, enabling early and accurate prediction of spoilage risks. More importantly, by introducing a net economic value decision-making model, it can minimize operational energy consumption and cargo loss while ensuring preservation effects, thus optimizing commercial value. Its adaptive optimization mechanism ensures long-term, stable, and efficient system operation, thereby improving the overall economic benefits and intelligent management level of the post-harvest vegetable supply chain.

[0083] Example 2:

[0084] S3 specifically includes:

[0085] S31. Retrieve weighting coefficients from the preset expert database;

[0086] S32. The physiological feature vectors are weighted and summed based on the weight coefficients to calculate the dimensionless physiological activity index.

[0087] S33. Combine the physiological activity indicators of each local area to generate a distribution map of physiological activity indicators.

[0088] This embodiment is a specific implementation of step S3 in embodiment 1; based on embodiment 1, the step of inputting the physiological feature vector into the physiological activity index model to generate a physiological activity index distribution map is specifically implemented through the following sub-steps:

[0089] S31. Retrieve weighting coefficients from a pre-defined expert database. To enable the physiological activity index model to adapt to different types of vegetables, such as the physiological differences between climacteric vegetables (tomatoes and melons) and non-climacteric vegetables, this embodiment constructs an expert database. The expert database refers to a set of pre-defined parameters storing the physiological characteristics of different preservation targets. Its data source is established by performing regression analysis on historical experimental data and combining it with domain expert knowledge. Before calculation, the system will automatically retrieve a set of optimal weighting coefficients from this database based on the current vegetable category to be preserved. ;

[0090] S32. The physiological feature vectors are weighted and summed based on weighting coefficients to calculate the dimensionless physiological activity index; this step is the core calculation step in generating the physiological activity index; the physiological activity index model used in this embodiment is specifically mathematically expressed as follows:

[0091]

[0092] in, Physiological activity index is a dimensionless comprehensive score that quantifies the activity level of vital signs in vegetables.

[0093] The normalized physiological characteristics are the normalized ethylene concentration, carbon dioxide concentration, and hyperspectral anomaly characteristic value, which are obtained from the normalization calculation in the previous step S2.

[0094] The weighting coefficients are three dimensionless parameters that balance the importance of different physiological characteristics in the comprehensive assessment. They are retrieved from the expert database in step S31.

[0095] This formula allows for the transformation of multidimensional physiological feature vectors. To merge into a single entity that can directly characterize physiological activity. value;

[0096] S33. Combine the physiological activity indicators of each local region to generate a distribution map of physiological activity indicators; after performing the calculation in S32 on each monitored local region within the preservation unit, a series of values ​​corresponding to different spatial locations will be obtained. Value; these By combining the values ​​with their spatial coordinates, a two-dimensional or three-dimensional spatial distribution map of physiological activity indicators can be generated.

[0097] The technical solution in this embodiment enables the calculation of physiological activity indicators to be highly adaptive. By retrieving weight coefficients for specific varieties from the expert database, the universality and accuracy of the model are ensured, avoiding the huge deviations that may occur when using a fixed-weight model to deal with vegetables with different physiological characteristics. This weighted fusion method can more scientifically and accurately quantify the true physiological state of vegetables, providing more reliable basic data for subsequent risk prediction, thereby improving the accuracy of the entire system assessment.

[0098] Example 3:

[0099] S4 specifically includes:

[0100] S41. Obtain the current physiological activity of the region;

[0101] S42. Calculate the rate of change of physiological activity over time;

[0102] S43, Accumulate the spatial diffusion effect of neighboring areas;

[0103] S44. Based on the preset static weights, dynamic weights, and spatial diffusion effect weights, the physiological activity, time change rate, and spatial diffusion effect are weighted and fused to calculate the putrefaction propagation potential.

[0104] This embodiment is a specific implementation of step S4 in embodiment 1, and its core lies in the construction and calculation of the corruption propagation potential prediction model. Based on embodiment 1, this model calculates the corruption propagation potential through the following sub-steps:

[0105] S41. Obtain the current physiological activity of the region; this step directly utilizes the physiological activity indicators generated in the preceding step S3. As the basic input to the model; Refers to the area At the present moment The physiological activity index value; it represents the static benchmark for risk assessment;

[0106] S42. Calculate the time-varying rate of physiological activity; to capture the dynamic development trend of risk, this embodiment introduces a time dimension; the time-varying rate of physiological activity is approximated by a first-order backward difference, specifically calculated as follows: This value reflects the region. At the most recent time step The rate of deterioration or improvement in endogenous physiological activity represents a dynamic component of risk assessment.

[0107] S43. Accumulate the spatial diffusion effect of neighboring areas; considering that putrefaction is usually infectious, a high-risk area will affect its surrounding areas; this embodiment quantifies this impact by calculating the spatial diffusion effect of neighboring areas; the specific calculation is as follows: ,in It is related to the region Adjacent region sets It is the neighboring area The current physiological activity, and It is a description area For the region The dimensionless spatial weights affecting intensity; this term represents the spatial components of risk assessment.

[0108] S44. Based on preset static weights, dynamic weights, and spatial diffusion effect weights, physiological activity, time-varying rate, and spatial diffusion effect are weighted and fused to calculate the putrefaction propagation potential. This step integrates the above three components into the final prediction index. To ensure the consistency of the formula's dimensions, a characteristic time scale is introduced to make the time-varying rate term dimensionless. The mathematical expression of this model is as follows:

[0109] in, The potential for corruption to spread is a dimensionless risk prediction indicator; its function is to quantify regional risks. In the future The risks of corruption and its potential for spread;

[0110] and Physiological activity indicators, which are obtained from the calculation in the preceding step S3;

[0111] The preset static weight, dynamic weight, and spatial diffusion effect weight are three dimensionless coefficients that sum to 1. Their function is to balance the importance of the current state, the rate of change, and the spatial impact. They are set based on risk evolution patterns identified from a large amount of historical data.

[0112] : Characteristic time scale, its physical meaning lies in the typical time constant that characterizes the evolution of physiological activity. It is derived from the pre-calibration based on the putrefaction dynamics of a specific vegetable variety. Its dimension is time, which is used to ensure that all terms in the formula are dimensionless.

[0113] The putrefaction propagation potential prediction model proposed in this embodiment, compared to one that relies solely on current physiological activity, offers a more advanced approach. The proposed method for making judgments has significant advantages. By innovatively integrating the temporal rate of change of physiological activity and the spatial diffusion effect, the model can provide early warnings of rapidly developing corruption cores with the risk of spreading, achieving a leap from state assessment to trend prediction. This greatly enhances the system's early warning capability, providing a valuable time window for taking preventive intervention measures, thereby more effectively suppressing the occurrence of cascading corruption.

[0114] Example 4:

[0115] S5 specifically includes:

[0116] S51. Set early warning thresholds and danger thresholds as risk thresholds;

[0117] S52. Areas where the potential for corruption to spread is lower than or equal to the warning threshold shall be classified as safety levels.

[0118] S53. Areas where the potential for corruption to spread is higher than the warning threshold but lower than or equal to the danger threshold shall be classified as warning levels.

[0119] S54. Areas where the potential for corruption to spread exceeds the danger threshold are classified as danger levels.

[0120] This embodiment is a specific implementation of step S5 in embodiment 1, which details how to assess the level of corruption propagation potential based on a preset risk threshold;

[0121] S51. Set warning thresholds and danger thresholds as risk thresholds; in order to achieve refined risk management, this embodiment sets two key custom threshold parameters;

[0122] Warning threshold Its function is to distinguish between a safe state and a warning state that requires attention; the setting of this threshold is based on rigorous statistical evidence. It is determined by statistical analysis of massive historical preservation data as the inflection point of the Weber distribution where the physiological state of vegetables deviates from the ideal dormancy state.

[0123] Danger threshold Its function is to distinguish between a warning state and a dangerous state that requires immediate intervention; the determination of this threshold combines accelerated aging experiments with historical business loss data; once the potential for corruption to spread in a region exceeds this value, the probability of irreversible business value loss within 24 hours will exceed 95% if no intervention is taken;

[0124] S52. Areas with a corruption propagation potential lower than or equal to the warning threshold are classified into security levels; for the calculated corruption propagation potential... The system marks the area with the specified security level.

[0125] S53. Areas where the potential for corruption to spread is higher than the warning threshold but lower than or equal to the danger threshold are classified as warning levels; for areas that meet the following criteria... The system marks the area as a warning level;

[0126] S54. Areas where the potential for corruption to spread exceeds the danger threshold are classified as danger levels; for The system marks the area as a danger level;

[0127] This embodiment achieves dynamic and precise risk level classification by setting early warning and danger thresholds with clear physical and commercial significance. This classification method is not only logically clear and easy to implement, but its threshold setting is based on big data statistics and experimental calibration, which has a high degree of scientificity and reliability. It enables the subsequent decision-making system to take intervention measures of different intensities and costs according to different risk levels, avoiding a one-size-fits-all approach to extensive management, and laying a solid foundation for achieving a graded, precise and economical control strategy.

[0128] Example 5:

[0129] S6 specifically includes:

[0130] S61. For areas under warning or danger, iterate through the set of preset intervention actions;

[0131] S62. For any intervention action in the set, calculate the execution cost;

[0132] S63. Call a pre-trained machine learning model to predict the expected residual loss cost after taking intervention actions;

[0133] S64. Subtract execution costs and expected residual loss costs from the current total market value of the goods to obtain the expected net economic value;

[0134] S65. Select the intervention action that maximizes the expected net economic value as the optimal intervention action vector.

[0135] This embodiment is a specific implementation of step S6 described in embodiment 1, and elaborates in detail how the net economic value decision model generates the optimal intervention action vector;

[0136] S61. For areas under warning or danger, iterate through the set of preset intervention actions; the system maintains a set of preset intervention actions. The preset intervention action set refers to a library containing all technically feasible control operations, such as {ventilating area A for 5 minutes, lowering the overall temperature by 1 degree Celsius, and performing targeted ethylene removal in area B}. When a warning or hazardous area is detected, the decision-making process is initiated and begins iterating through each alternative action in this set. ;

[0137] S62. For any intervention action in the set, calculate the execution cost; the system will calculate the execution cost. Required execution costs The cost is in monetary units and is calculated based on a preset equipment energy consumption model and electricity prices, consumable prices, etc. For example, the cost of ventilation is related to its duration and fan power.

[0138] S63. Call the pre-trained machine learning model to predict the expected residual loss cost after taking intervention actions; this is the core step in the decision-making process; for each alternative action... The system utilizes a pre-trained machine learning model to predict risks in a given risk state. Next, implement intervention actions. Expected residual loss cost after This model is built by learning from massive amounts of historical risk-intervention-outcome-loss data, and can predict the mitigation effects of different intervention measures on different risk states and quantify them as potential future economic losses.

[0139] S64. Subtract execution costs and expected residual loss costs from the current total market value of the goods to obtain the expected net economic value; this step involves cost-benefit calculation; expected net economic value. The calculation formula is as follows:

[0140]

[0141] in: The current total market value is derived from known parameters input from external sources.

[0142] : Expected residual loss cost, which is predicted by the machine learning model in step S63;

[0143] : Execution cost, which is derived from the calculation based on the equipment energy consumption model in step S62;

[0144] S65. Select the intervention action that maximizes the expected net economic value as the optimal intervention action vector; the system will traverse the set. All actions in the process, and calculate a value for each action. Value; ultimately, choose the one that makes The maximized action is the optimal intervention action vector. Formalized as:

[0145]

[0146] This embodiment changes the traditional decision-making paradigm of food preservation control by establishing a net economic value decision-making model. It no longer pursues the optimization of a certain technical indicator in isolation, but takes the maximization of commercial value as the ultimate goal, and makes a dynamic and intelligent trade-off between intervention costs and potential losses. This method can accurately identify the intervention strategy with the highest cost-effectiveness ratio, effectively control the risk of spoilage, and avoid energy waste caused by excessive intervention, thereby bringing direct and significant economic benefits to users.

[0147] Example 6:

[0148] The hardware control instructions include intervention type parameters, intervention intensity parameters, duration parameters, and area of ​​action identification parameters.

[0149] This embodiment is a specific description of hardware control instructions; based on embodiment 1, it uses the optimal intervention action vector. The parsed hardware control instructions are a structured instruction package that explicitly contains all the information required to perform physical intervention. Its characteristics include:

[0150] Intervention type parameters This parameter defines the type of operation to be performed, for example... Indicates ventilation. Indicates a drop in temperature, or Indicates ethylene removal;

[0151] Intervention intensity parameters This parameter quantifies the severity of the operation; for example, in ventilation interventions, This can represent the target speed of the fan; in cooling intervention, It can represent the power percentage of a refrigeration system;

[0152] Duration parameter This parameter defines the duration for which the intervention action needs to be maintained, in seconds or minutes.

[0153] Scope Identifier Parameters This parameter specifies the target physical space for the intervention action, for example... This indicates that the command only applies to zone 3 within the preservation unit;

[0154] A complete Examples can be: The vector will be parsed into an instruction sent to the ventilation fan controller of Zone 3, causing it to run at 1500 RPM for 300 seconds.

[0155] By structuring hardware control commands into parameterized vectors that include intervention type, intensity, duration, and area of ​​effect, this embodiment ensures that information transmission from upper-level economic decision-making to lower-level physical execution is lossless and unambiguous. This design enables control commands to have high precision and flexibility, supporting complex and refined regional and intensity-based intervention strategies, ensuring that optimal decisions can be executed accurately, and is an important guarantee for achieving precise control and system reliability.

[0156] Example 7:

[0157] S9 specifically includes:

[0158] S91. After the intervention is completed, regenerate the risk level map;

[0159] S92. Compare the risk level charts before and after the intervention, and quantify the risk changes as a measure of control effectiveness;

[0160] S93. Use the control effect as feedback to update the machine learning model in the net economic value decision model used to predict expected residual loss costs.

[0161] This embodiment is a specific implementation of step S9 in embodiment 1. It details how the system quantifies the control effect and uses the control effect to update the net economic value decision model online, thereby forming an adaptive closed loop.

[0162] S91. After the intervention is completed, regenerate the risk level map; when an intervention action is performed by... Once the system is executed, it waits for a preset observation period, such as 30 minutes, and then automatically triggers a completely new and complete data collection and analysis process, that is, repeating steps S1 to S5 to generate a new risk level map representing the post-intervention status.

[0163] S92. Compare the risk level maps before and after the intervention to quantify the risk change as the control effect; the system compares the two risk level maps before and after the intervention by region; the control effect is quantified as the migration status of the risk level, for example, recording how many dangerous areas have turned into warning or safe areas, and how many warning areas have turned into safe areas; this positive change in risk level is defined as a successful control.

[0164] S93. Use the control effect as feedback to update the machine learning model in the net economic value decision model used to predict expected residual loss costs; this step is crucial for achieving online learning and self-optimization; the control effect quantified in step S92 is compared with the pre-intervention risk state that led to this effect. The intervention actions taken Together, they form a new, high-quality training data sample tuple. This new sample is input into step S63 for prediction. In the machine learning model, trigger an online incremental training;

[0165] This embodiment constructs a complete closed loop of intervention-evaluation-feedback-learning, endowing the entire control system with adaptive optimization capabilities. The system is no longer a static executor based on fixed rules, but an intelligent agent that can continuously learn and evolve in actual operation. By continuously utilizing the latest control effects to update its internal decision-making model, the system's prediction of the effectiveness of various intervention measures under different risk states becomes increasingly accurate, thereby continuously improving its long-term decision-making level and ensuring that it can always make the closest to optimal economic decisions when facing constantly changing cargo conditions and external environments.

[0166] Example 8:

[0167] Please see Figure 2 A remote intelligent control system for post-harvest preservation of vegetables, comprising:

[0168] The physiological information processing module is used to collect the physiological information of vegetables in various local areas and generate normalized physiological feature vectors.

[0169] The physiological state modeling module is used to receive physiological feature vectors and generate a distribution map of physiological activity indicators.

[0170] The risk prediction and assessment module is used to receive the distribution map of physiological activity indicators, generate the putrefaction transmission potential, and assess and generate a risk level map based on the preset risk threshold.

[0171] The economic decision-making module receives the risk level map and combines it with cargo value information to generate the optimal intervention action vector.

[0172] The intervention execution module is used to parse the optimal intervention action vector into hardware control instructions and drive the underlying hardware to execute them.

[0173] The adaptive optimization module is used to quantify the control effect after the intervention is implemented and to update the economic decision-making module online using the control effect.

[0174] This embodiment provides a remote intelligent control system for post-harvest preservation of vegetables. This system is a physical or virtual device designed to implement the above-described method; it logically or physically includes the following cooperative modules:

[0175] Physiological information processing module: As the data input end of the system, its function is to collect raw physiological data from the physical world and standardize it; in this embodiment, it integrates hardware such as hyperspectral camera, gas sensor and temperature and humidity sensor, and has built-in data acquisition and normalization algorithm; it is responsible for executing S1 and S2 in the aforementioned method to collect vegetable physiological information in various local areas and generate normalized physiological feature vectors, and its output is passed to the physiological state modeling module.

[0176] Physiological state modeling module: As the system's state assessment unit, its function is to integrate multi-dimensional physiological characteristics into a single, macroscopic activity index. In this embodiment, it has a built-in linear weighted fusion model and expert database. It is responsible for executing S3, which receives physiological feature vectors, generates a distribution map of physiological activity indexes, and outputs the distribution map to the risk prediction and assessment module.

[0177] Risk prediction and assessment module: As the risk early warning unit of the system, its function is to predict the evolution trend of future risks from the current state and classify them into levels; in this embodiment, it implements the corruption propagation potential prediction model and risk level assessment logic; it is responsible for executing S4 and S5 to receive the distribution map of physiological activity indicators, generate the corruption propagation potential, and evaluate and generate a risk level map based on the preset risk threshold; its output risk level map is the input of the economic decision-making module.

[0178] Economic Decision Module: As the core decision-making unit of the system, its function is to generate the optimal control strategy with the goal of maximizing business value. In this embodiment, it has a built-in net economic value decision model, which includes a machine learning model for predicting residual losses. It is responsible for executing S6, which receives the risk level map, combines it with the cargo value information to generate the optimal intervention action vector, and passes the vector to the intervention execution module.

[0179] Intervention Execution Module: As the physical execution unit of the system, its function is to accurately translate high-level decisions into low-level physical operations. In this embodiment, it includes an instruction parser and an interface for communication with low-level hardware, such as a programmable logic controller. It is responsible for executing S7 and S8 to parse the optimal intervention action vector into hardware control instructions and drive the low-level hardware to execute them.

[0180] Adaptive optimization module: As the system's self-learning and evolution unit, its function is to evaluate the control effect and use it to optimize future decisions; in this embodiment, it quantifies the effect by comparing data before and after the intervention and has the ability to update the machine learning model online; it is responsible for executing S9, which is used to quantify the control effect after the intervention is implemented and to update the economic decision-making module online using the control effect;

[0181] This embodiment, through modular system design, clearly defines the various functional units and their interrelationships for implementing the method of the present invention, ensuring the feasibility and reliability of the technical solution. Each module performs its own function and is tightly coupled, together forming a complete intelligent system from data acquisition, status assessment, risk prediction, economic decision-making to closed-loop optimization. This system architecture is not only logically clear and easy to develop and maintain, but also fully functional, capable of stably and efficiently executing the aforementioned methods. In practical applications, it enables intelligent, refined, and economical management of the vegetable preservation process. Furthermore, through adaptive optimization, the prediction becomes increasingly accurate, ensuring that the system can continuously adapt to changes in the condition of goods and the environment during long-term operation, continuously improving the decision-making level, and guaranteeing the long-term stability and efficiency of the entire control system.

[0182] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention; any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0183] 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.

Claims

1. A remote intelligent control method for post-harvest preservation of vegetables, characterized in that, Includes the following steps: S1. Collect vegetable physiological information from various local areas, including hyperspectral data, ethylene concentration, and carbon dioxide concentration; S2. Based on the physiological information of vegetables, generate normalized physiological feature vectors; S3. Input the physiological feature vector into the physiological activity index model to generate a physiological activity index distribution map; S4. Input the distribution map of physiological activity indicators into the putrefaction propagation potential prediction model to generate the putrefaction propagation potential. S5. Based on the preset risk threshold, assess the level of corruption transmission potential and generate a risk level map. S6. Input the risk level map and cargo value information into the net economic value decision model to generate the optimal intervention action vector; S7. Analyze the optimal intervention action vector and generate hardware control instructions; S8 drives the underlying hardware to execute hardware control instructions; S9. After the intervention is implemented, quantify the control effect and use the control effect to update the net economic value decision model online; Specifically, S4 includes: S41, obtaining the current physiological activity of the region. S42. Calculate the rate of change of physiological activity over time. S43. Accumulate the spatial diffusion effect of neighboring areas. S44. Based on the preset static weights, dynamic weights, and spatial diffusion effect weights, the physiological activity, time change rate, and spatial diffusion effect are weighted and fused to calculate the putrefaction propagation potential. S6 specifically includes: S61, for areas in a warning or dangerous state, iterating through the set of preset intervention actions. S62. For any intervention action in the set, calculate the execution cost. S63. Call the pre-trained machine learning model to predict the expected residual loss cost after taking intervention actions. S64. Subtract execution costs and expected residual loss costs from the current total market value of the goods to obtain the expected net economic value. S65. Select the intervention action that maximizes the expected net economic value as the optimal intervention action vector.

2. The remote intelligent control method for post-harvest preservation of vegetables according to claim 1, characterized in that, S3 specifically includes: S31. Retrieve weighting coefficients from the preset expert database; S32. The physiological feature vectors are weighted and summed based on the weight coefficients to calculate the dimensionless physiological activity index. S33. Combine the physiological activity indicators of each local area to generate a distribution map of physiological activity indicators.

3. The remote intelligent control method for post-harvest preservation of vegetables according to claim 1, characterized in that, S5 specifically includes: S51. Set early warning thresholds and danger thresholds as risk thresholds; S52. Areas where the potential for corruption to spread is lower than or equal to the warning threshold shall be classified as safety levels. S53. Areas where the potential for corruption to spread is higher than the warning threshold but lower than or equal to the danger threshold shall be classified as warning levels. S54. Areas where the potential for corruption to spread exceeds the danger threshold are classified as danger levels.

4. The remote intelligent control method for post-harvest preservation of vegetables according to claim 1, characterized in that, The hardware control instructions include intervention type parameters, intervention intensity parameters, duration parameters, and area of ​​action identification parameters.

5. The remote intelligent control method for post-harvest preservation of vegetables according to claim 1, characterized in that, S9 specifically includes: S91. After the intervention is completed, regenerate the risk level map; S92. Compare the risk level charts before and after the intervention, and quantify the risk changes as a measure of control effectiveness; S93. Use the control effect as feedback to update the machine learning model in the net economic value decision model used to predict expected residual loss costs.

6. A remote intelligent control system for post-harvest preservation of vegetables, based on the remote intelligent control method for post-harvest preservation of vegetables according to any one of claims 1-5, characterized in that, include: The physiological information processing module is used to collect the physiological information of vegetables in various local areas and generate normalized physiological feature vectors. The physiological state modeling module is used to receive physiological feature vectors and generate a distribution map of physiological activity indicators. The risk prediction and assessment module is used to receive the distribution map of physiological activity indicators, generate the putrefaction transmission potential, and assess and generate a risk level map based on the preset risk threshold. The economic decision-making module receives the risk level map and combines it with cargo value information to generate the optimal intervention action vector. The intervention execution module is used to parse the optimal intervention action vector into hardware control instructions and drive the underlying hardware to execute them. The adaptive optimization module is used to quantify the control effect after the intervention is implemented and to update the economic decision-making module online using the control effect.

Images