A fresh produce transportation whole-process tracking method and system based on RFID

Abstract

The application provides a fresh food transportation whole-process tracking method and system based on RFID. Specifically, a coupling state vector is constructed, which contains a three-dimensional position of a vehicle, a speed, an internal temperature and humidity of goods, and a remaining shelf life of the goods based on an Arrhenius equation; a parallel system model covering possible states of the whole transportation process is constructed; an initial state for the current prediction is generated for each model; for each mixed model, Kalman filtering state prediction is performed, and a process noise covariance matrix Q is dynamically configured according to information including a road type corresponding to a current position of the vehicle and real-time traffic flow; new measurement data is received, effective data is used to update state estimation and likelihood of each model after validity inspection through a verification gate based on Mahalanobis distance, and posterior probability of each model is calculated; and according to the updated model posterior probability, posterior state estimation of all parallel models is weighted and fused to output optimal state estimation at the current time.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of RFID, and particularly relates to a fresh produce transportation whole-process tracking method and system based on RFID. BACKGROUND

[0002] Fresh agricultural products and food are extremely sensitive to environmental temperature, humidity and other conditions in the transportation process from the production place to the consumer due to their biological characteristics, and any improper link may cause the quality of goods to decline, spoil and cause huge economic losses. Realizing whole-process, real-time and accurate monitoring of the in-transit state of fresh goods is the core of the cold-chain logistics field. At present, the commonly used technical solution is to obtain the geographic position information of the transportation vehicle by using the global positioning system (GPS), and to monitor the temperature and humidity in the goods box by combining the radio frequency identification (RFID) technology or various wireless sensors. This method can realize double monitoring of position and environment to a certain extent, but the data systems are often independently operated, and the information island phenomenon is serious, and the dynamic running state of the vehicle and the internal quality state of the goods cannot be deeply coupled and analyzed.

[0003] Moreover, fresh produce transportation is a complex dynamic process with multi-modal characteristics, and will experience normal driving, abnormal stop caused by traffic congestion, sudden failure of refrigeration equipment, opening of the box for loading and unloading operations and other different physical states. The traditional single model filtering method is difficult to accurately describe and adapt to the mutation of such states, and when the transportation state switches, the filtering accuracy will decrease significantly, and even the model may diverge. In addition, how to convert the discrete temperature and humidity measurement values into continuous and accurate prediction of the remaining shelf life of the core value of the goods, and how to effectively eliminate abnormal data generated by the sensor in the complex electromagnetic and physical environment, are still obvious short boards in the current technical solutions. Therefore, there is an urgent need for a comprehensive tracking method that can integrate multi-source information, cope with variable transportation conditions and accurately predict the internal quality state of the goods. SUMMARY

[0004] In view of the problems mentioned in the background, first, the application provides a fresh produce transportation whole-process tracking method based on RFID, comprising the following steps:

[0005] acquiring the fresh goods identification, temperature and humidity collected by the RFID tag, and the GPS position and vehicle running state data sent by the vehicle terminal; and constructing a coupled state vector containing the three-dimensional position of the vehicle, the speed, the internal temperature and humidity of the goods and the remaining shelf life of the goods based on the Arrhenius equation;

[0006] A parallel system model covering possible states of the whole transportation process is constructed, which includes a normal transportation model, an abnormal stop model, a temperature rise model caused by refrigeration failure, and a loading and unloading operation model; based on the posterior probability of each model at the previous time, the state estimation of each model is interactively mixed to generate an initial state for each model for this round of prediction;

[0007] For each model after the interactive mixing, Kalman filter state prediction is performed, wherein the evolution of the remaining shelf life state follows a dynamic equation coupled with the temperature state, and the process noise covariance matrix Q is dynamically configured according to information including the road type corresponding to the current position of the vehicle and the real-time traffic flow;

[0008] After receiving new measurement data, performing data validity test through a verification gate based on Mahalanobis distance, updating the state estimation and likelihood of each model using the valid data, and calculating the posterior probability of each model, the posterior state estimation of all parallel models is weighted and fused according to the updated model posterior probability to output the optimal state estimation at the current time.

[0009] Optionally, the coupled state vector is defined as X k =[x k ,y k ,z k ,vx k ,vy k ,vz k ,T k ,H k ,SL k ] T , wherein [x k ,y k ,z k ] is the three-dimensional position coordinates of the vehicle at time k, [vx k ,vy k ,vz k ] is the three-dimensional speed of the vehicle at time k, T k and H k are the temperature and humidity inside the goods at time k, and SL k is the remaining shelf life of the goods at time k.

[0010] Optionally, the state evolution of the remaining shelf life of the goods SL k follows a dynamic model based on the Arrhenius equation: SL k =SL k-1 -Δt*k ref *exp[(Ea / R)*(1 / T ref -1 / T k )], wherein Δt is the sampling time interval, k ref is the reference temperature T refwhere Ea is the quality degradation rate constant, Ea is the activation energy of the degradation reaction, R is the ideal gas constant, T is the temperature k is the internal temperature state of the goods at time k.

[0011] Optionally, the parallel system model is specifically defined as follows:

[0012] Normal transportation model: the vehicle position and speed follow the constant speed kinematics model, and the temperature and humidity are stable around the set value;

[0013] Abnormal stop model: the vehicle speed is zero, the position remains unchanged, and the temperature and humidity remain stable;

[0014] Refrigeration failure temperature rise model: the vehicle motion state is consistent with the normal transportation model, and the temperature state follows the kinetic model representing continuous temperature rise;

[0015] Loading and unloading operation model: the vehicle speed is zero, and the temperature and humidity state model includes a function for responding to the temperature and humidity step change caused by the opening and closing of the box door.

[0016] Optionally, the process noise covariance matrix Q is dynamically configured according to information including the road type corresponding to the current position of the vehicle and the real-time traffic flow, comprising:

[0017] When it is determined that the vehicle is in a stable driving state, the process noise parameters corresponding to the position and speed components in Q are set to a set of preset low values;

[0018] When it is determined that the vehicle is in a non-stable driving state, the process noise parameters corresponding to the position and speed components in Q are set to a set of preset high values.

[0019] Optionally, the data validity test is performed through the Mahalanobis distance-based verification gate, comprising:

[0020] Calculate the residual v between the new measurement value and the predicted measurement value generated based on the state prediction k and the covariance matrix S k of the residual.

[0021] Calculate the square of the Mahalanobis distance d

[0022] Compare d 2 with the threshold value γ determined from the chi-square distribution according to the dimension of the measurement vector and the preset confidence level, when d 2 < γ, it is determined that the measurement data is valid.

[0023] Optionally, the posterior state estimates of all parallel models are weighted and fused, comprising:

[0024] The optimal state estimate after fusion is calculated by the formula where is the state estimate at time k, is the posterior state estimate of the jth model at time k, is the posterior probability of the jth model at time k, and the summation is taken over all models j.

[0025] Secondly, the application further provides a fresh food transportation whole-process tracking system based on RFID, comprising the following units:

[0026] The acquisition unit is used for acquiring fresh food product identification, temperature and humidity collected by the RFID tag, and GPS position and vehicle operation state data sent by the vehicle terminal; and a coupling state vector is constructed, which comprises vehicle three-dimensional position, speed, internal temperature and humidity of goods, and residual shelf life of goods based on the Arrhenius equation.

[0027] The model construction unit is used for constructing a parallel system model covering possible states in the whole transportation process, and the model comprises a normal transportation model, an abnormal stop model, a refrigeration failure temperature rise model and a loading and unloading operation model; based on the posterior probability of each model at the last time, the state estimates of each model are interactively mixed to generate an initial state for each model for this round of prediction;

[0028] The prediction unit is used for performing Kalman filter state prediction for each interactively mixed model, wherein the evolution of the residual shelf life state follows a dynamic equation coupled with the temperature state, and a process noise covariance matrix Q is dynamically configured according to information including the road type corresponding to the current position of the vehicle and real-time traffic flow;

[0029] The estimation unit is used for receiving new measurement data, performing data validity inspection through a verification gate based on Mahalanobis distance, updating the state estimates and likelihoods of each model using the valid data, and calculating the posterior probability of each model; according to the updated model posterior probability, the posterior state estimates of all parallel models are weighted and fused to output the optimal state estimate at the current time.

[0030] Optionally, the coupling state vector is defined as X k = [x k , y k , z k , vx k , vy k , vz k , T k , H k , SL k ] T , wherein [x k , y k , z k ] is the three-dimensional position coordinate of the vehicle at time k, [vx k , vy kvz k is the three-dimensional velocity of the vehicle at time k, T k is the temperature inside the cargo at time k, and H k is the humidity inside the cargo at time k, SL k is the remaining shelf life of the cargo at time k.

[0031] Optionally, the state evolution of the remaining shelf life SL k of the cargo follows a kinetic model based on the Arrhenius equation: SL k = SL k-1 - Δt * k ref * exp[(Ea / R)*(1 / T ref - 1 / T k )], where Δt is the sampling time interval, k ref is the quality deterioration rate constant at a reference temperature T ref , Ea is the activation energy of the deterioration reaction, R is the ideal gas constant, and T k is the temperature state inside the cargo at time k.

[0032] Optionally, the parallel system model is specifically defined as follows:

[0033] Normal transportation model: the vehicle position and velocity follow the constant speed kinematics model, and the temperature and humidity are stable around the set values;

[0034] Abnormal stop model: the vehicle speed is zero, the position remains unchanged, and the temperature and humidity remain stable;

[0035] Refrigeration failure temperature rise model: the vehicle motion state is consistent with the normal transportation model, and the temperature state follows a kinetic model representing continuous temperature rise;

[0036] Loading and unloading operation model: the vehicle speed is zero, and the temperature and humidity state model includes a function for responding to the temperature and humidity step changes caused by the opening and closing of the box door.

[0037] Optionally, the process noise covariance matrix Q is dynamically configured according to information including the road type corresponding to the current position of the vehicle and the real-time traffic flow, comprising:

[0038] When it is determined that the vehicle is in a stable driving state, the process noise parameters corresponding to the position and velocity components in Q are set to a set of preset low values;

[0039] When it is determined that the vehicle is in a non-stable driving state, the process noise parameters corresponding to the position and velocity components in Q are set to a set of preset high values.

[0040] Optionally, the data validity test is performed through a verification gate based on the Mahalanobis distance, comprising:

[0041] calculating the residual v between the new measurement and the predicted measurement generated based on the state prediction k and its covariance matrix S k ;

[0042] calculating the square of Mahalanobis distance

[0043] comparing d 2 with the threshold value γ determined from the chi-square distribution according to the dimension of the measurement vector and the preset confidence level, when d 2 < γ, determining that the measurement data is valid.

[0044] Optionally, the weighted fusion of the posterior state estimates of all parallel models comprises:

[0045] calculating the optimal state estimate after fusion by the formula , wherein is the state estimate at time k, is the posterior state estimate of the jth model at time k, is the posterior probability of the jth model at time k, and the summation is performed for all models j.

[0046] Compared with the prior art, the present application has at least the following advantages:

[0047] 1. By constructing a coupled state vector containing the vehicle position, speed, cargo temperature and humidity, and remaining shelf life, and closely combining the vehicle running state with the cargo quality state, the quantitative continuous prediction of the remaining shelf life of the goods is realized.

[0048] 2. By using the interactive multiple model filtering framework, through the preset normal transportation, abnormal stoppage, refrigeration failure and other parallel models, the possible switching of various working conditions during transportation can be accurately dealt with, and the defect of single model in complex scenes is overcome.

[0049] 3. By introducing the data verification mechanism based on Mahalanobis distance, abnormal measurement data is effectively filtered out, and the filtering process is adjusted combined with road type, traffic flow and other information, the state estimates of all parallel models are fused, and more smooth and accurate comprehensive state information is output, which improves the tracking accuracy and reliability of the fresh goods state in complex transportation environment. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 is the flowchart of Example 1;

[0051] Figure 2 is a schematic diagram for verifying the validity of data;

[0052] Figure 3 is a schematic diagram for identifying different motion modes;

[0053] Figure 4 An illustration of the weighted fusion state estimation. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the scope of protection of the present application.

[0055] In a first embodiment, as shown in Figure 1 A method for tracking fresh food transportation based on RFID is provided, comprising the following steps:

[0056] S1, obtaining fresh food product identification, temperature and humidity collected by RFID tags, and GPS position and vehicle running state data sent by a vehicle terminal; constructing a coupled state vector including vehicle three-dimensional position, speed, internal temperature and humidity of goods, and remaining shelf life of goods based on the Arrhenius equation;

[0057] Active RFID tags with temperature and humidity sensors are deployed inside the transport vehicle's cargo box. The tags periodically broadcast product identification, temperature and humidity data, which are received and aggregated by the RFID reader inside the cargo box. At the same time, the vehicle T-Box terminal obtains three-dimensional position information of longitude, latitude and elevation through the built-in GPS module, and obtains running data such as speed, engine speed, and door switch state through the connection of vehicle CAN bus. All data are uploaded to the cloud server in real time through 4G or 5G wireless network for unified processing. In a more specific embodiment, a nine-dimensional state vector X is defined, whose components are the vehicle's position in the east, north and sky directions, the vehicle's speed in the east, north and sky directions, the temperature inside the goods, the humidity inside the goods, and the remaining shelf life of the goods. The remaining shelf life is preferably a normalized value, for example, initially 1, and the rate of change over time is related to the temperature T in the state vector through the Arrhenius equation, i.e. the loss rate of the remaining shelf life is proportional to the exponential function of the temperature.

[0058] S2, constructing a parallel system model covering possible states of the entire transportation process, the model including: a normal transportation model, an abnormal stop model, a refrigeration failure temperature rise model, and a loading and unloading operation model; based on the posterior probability of each model at the last time, the state estimation of each model is interactively mixed to generate an initial state for each model for this round of prediction;

[0059] Four parallel system models are initialized for the interactive multi-model Kalman filter. The normal transportation model adopts constant velocity kinematic model to describe the vehicle dynamics, and assumes that the temperature and humidity are stable within a small range. The abnormal stop model constrains the vehicle speed around zero, and keeps the temperature and humidity stable. The refrigeration failure temperature rise model introduces a specific dynamic equation describing the linear or exponential rise of temperature over time, in addition to the vehicle kinematics. The loading and unloading model sets the vehicle speed to zero, and significantly increases the process noise of the temperature and humidity states to reflect the dramatic disturbance of the environment caused by the opening of the box. According to the predefined inter-model transition probability matrix and the posterior probability of each model at the previous time k-1, the mixing probability of transition from model i to model j is calculated. Using these mixing probabilities, the weighted sum of the posterior state estimates and covariance matrices of all models at time k-1 is performed to generate an initial state estimate and an initial covariance matrix for each model j at time k.

[0060] S3, for each interactive mixed model, perform Kalman filter state prediction, wherein the evolution of the remaining shelf life state follows the dynamics equation coupled with the temperature state, and the process noise covariance matrix Q is dynamically configured according to information including the road type corresponding to the current location of the vehicle and the real-time traffic flow;

[0061] For each model j, the prediction equation of the extended Kalman filter is applied. The state transition matrix F is set according to the definition of model j, for example, in the constant velocity model, the position is equal to the position at the previous time plus the predicted speed multiplied by the time. In particular, for the remaining shelf life component, its predicted value is calculated by decreasing according to the temperature state at the previous time and the Arrhenius dynamics equation. The process noise covariance matrix Q is dynamically set, and the real-time map service API is called. When it is detected that the vehicle is driving on a congested road section, the Q matrix elements related to speed are increased; when driving on a bumpy rural road, the Q matrix elements related to position are increased.

[0062] S4, receive new measurement data, after verifying the validity of the data through the Mahalanobis distance-based validation gate, update the state estimate and likelihood of each model using the valid data, and calculate the posterior probability of each model; according to the updated model posterior probability, the posterior state estimates of all parallel models are weighted and fused to output the optimal state estimate at the current time.

[0063] When new GPS and RFID measurement data arrive, the Mahalanobis distance between the new measurement value and the predicted measurement value of each model is calculated. The distance is compared with the preset threshold based on the chi-square distribution. If the Mahalanobis distance exceeds the threshold, the measurement data is determined to be an outlier and is discarded, such as Figure 2The update step of the extended Kalman filter is performed for valid data passed by the test, and the Kalman gain is calculated for each model, and the state estimation and covariance matrix are updated. At the same time, the likelihood function value of each model is calculated according to the innovation and innovation covariance, and the posterior probability of each model at the current time is updated by combining the Bayesian formula.

[0064] The posterior probability of each model calculated in the above step is taken as a weight to perform a weighted average on the updated posterior state estimation vectors of all parallel models, so as to obtain a combined state estimation that integrates information of all models. For example, the final output temperature estimation value is the sum of the temperature estimation of the normal transportation model, the temperature estimation of the refrigeration failure model, and the like multiplied by their respective posterior probabilities. The state vector after the weighted fusion is the final output optimal state estimation at the current time, which contains the most possible vehicle position, speed, cargo temperature and humidity, and remaining shelf life.

[0065] In a preferred embodiment, the coupling state vector is defined as X k =[x k ,y k ,z k ,vx k ,vy k ,vz k ,T k ,H k ,SL k ] T , wherein [x k ,y k ,z k ] is the three-dimensional position coordinate of the vehicle at time k, [vx k ,vy k ,vz k ] is the three-dimensional speed of the vehicle at time k, T k and H k are the temperature and humidity inside the cargo at time k, and SL k is the remaining shelf life of the cargo at time k.

[0066] The coupling state vector X k can express the complete state of the cold chain transportation system at a specific time k. For example, at 3:00 pm when a batch of vaccines is being transported, the state vector X k is instantiated as: the position coordinates x k , y k , z k are 116.3 degrees east longitude, 39.9 degrees north latitude, and 55 meters above sea level; the speed components vx k , vy k , vz k are 20 meters per second eastward, 5 meters per second northward, and the vertical speed is zero; the temperature Tk is 5 degrees Celsius; humidity H k is sixty percent; the shelf life SL of the goods calculated from historical temperature and humidity data k is 350 hours. Integrating the physical motion state of the vehicle with the quality state of the goods in the same vector enables the system to understand and quantify the mutual influence between the two. For example, when the vehicle is stuck in traffic and the speed vx k , vy k , vz k approaches zero for a long time, if the refrigeration system load increases, it can cause a slight fluctuation in the temperature T_k. Although the fluctuation is slight, it will directly affect the calculation of the shelf life SL k through the dynamic model, thereby achieving the linkage monitoring and accurate evaluation of the transportation event and the quality of the goods.

[0067] In a preferred embodiment, the state evolution of the shelf life SL k of the goods follows a dynamic model based on the Arrhenius equation: SL k = SL k-1 - Δt * k ref * exp[(Ea / R)*(1 / T ref - 1 / T k )], where Δt is the sampling time interval, k ref is the quality degradation rate constant at the reference temperature T ref , Ea is the activation energy of the degradation reaction, R is the ideal gas constant, and T k is the internal temperature state of the goods at time k.

[0068] Temperature change and the loss of goods and shelf life are closely related. Assuming that a batch of fresh milk is being transported, the reference shelf temperature Tref is set to 4 degrees Celsius, and the quality degradation rate constant kref at this temperature is 0.002 per hour. The activation energy Ea of the milk degradation reaction is 65,000 J / mol, and the ideal gas constant R is 8.314 J / (mol·K). The sampling time interval Δt is 1 minute, i.e., 1 / 60 hours. In the last minute, the calculated shelf life SL k-1 is 168 hours. If the temperature sensor reading Tk in the current minute is 6 degrees Celsius, which is higher than the reference temperature, the value of the exponential term will be greater than 1, resulting in an actual quality degradation rate higher than the reference rate. Through the formula calculation, the loss of shelf life in the current minute will be greater than the loss at the standard 4 degrees Celsius. The new shelf life SL k will be updated to a value slightly less than 168 hours minus the standard loss. Compared to a fixed shelf life, it can more truly reflect the additional quality decline caused by the suboptimal temperature control during transportation.

[0069] In a preferred embodiment, the parallel system model is defined as follows:

[0070] Normal transport model: vehicle position and velocity follow constant speed kinematic model, temperature and humidity are stable around set values;

[0071] Abnormal stop model: vehicle velocity is zero, position remains unchanged, temperature and humidity remain stable;

[0072] Refrigeration failure temperature rise model: vehicle motion state is consistent with the normal transport model, temperature state follows a dynamic model representing continuous temperature rise;

[0073] Loading and unloading operation model: vehicle velocity is zero, temperature and humidity state model includes a function for responding to temperature and humidity step changes caused by opening and closing of the container door.

[0074] The four models run simultaneously and make different predictions for the future state. For example, when a refrigerated vehicle is driving on a highway at a speed of 90 kilometers per hour, the normal transport model will predict that the vehicle will continue to move at a constant speed, while the temperature will stabilize at the set value of 4 degrees Celsius. At the same time, the abnormal stop model will predict that the vehicle speed is zero and the position is unchanged, which is not consistent with the actual GPS data. The refrigeration failure temperature rise model will predict that while the vehicle speed remains 90 kilometers per hour, the temperature will rise at a rate of 0.2 degrees Celsius per minute. The loading and unloading operation model predicts that the vehicle is stationary and the temperature and humidity will change dramatically due to the opening of the door. By comparing the prediction results of the four models with the actual measurement data returned by the sensors at the next time, it can be determined which model best describes the current reality. If the GPS shows that the vehicle speed is still 90 kilometers per hour and the thermometer reading is stable at 4 degrees Celsius, then the normal transport model has the highest matching degree, and its weight will increase significantly. Conversely, if the vehicle speed is still normal but the temperature starts to rise continuously, then the weight of the refrigeration failure temperature rise model will increase, thereby automatically identifying potential equipment failure risks.

[0075] In a preferred embodiment, the process noise covariance matrix Q is dynamically configured according to information including the type of road corresponding to the current position of the vehicle and real-time traffic flow, comprising:

[0076] When it is determined that the vehicle is in a stable driving state, the process noise parameters corresponding to the position and velocity components in Q are set to a set of pre-set low values;

[0077] When it is determined that the vehicle is in a non-stable driving state, the process noise parameters corresponding to the position and velocity components in Q are set to a set of pre-set high values.

[0078] The process noise covariance matrix Q quantifies the uncertainty of the prediction model. When the refrigerated truck is driving on a highway with good road conditions, its motion state is very stable and can be well described by a constant velocity motion model. At this time, the term related to acceleration in the Q matrix is set to a very small value, such as 0.05, indicating that the system has a strong belief in the prediction ability of its motion model and believes that the vehicle will not have sudden and unpredictable acceleration and deceleration. The state estimator will rely more on model prediction, resulting in a smooth trajectory. When the vehicle enters the city and faces traffic lights and traffic congestion, its motion becomes frequent start and stop, which is difficult to predict. In this unstable state, the prediction of the constant velocity motion model will produce a larger error. In order to adapt to this situation, the term related to acceleration in the Q matrix is adjusted to a larger value, for example, 2.5. This is equivalent to telling the state estimator that the reliability of the prediction is reduced, and at this time the real-time GPS measurement data should be more believed and relied on. Through dynamic adjustment, the accurate tracking of the vehicle state can be maintained in different driving scenarios, Figure 3 The recognition graph of different motion modes is shown.

[0079] In order to filter out abnormal sensor data, in a preferred embodiment, the data validity test by the Mahalanobis distance-based validation gate comprises:

[0080] Calculate the residual v between the new measurement value and the predicted measurement value generated based on the state prediction k and the covariance matrix S of the residual k ;

[0081] Calculate the square of the Mahalanobis distance

[0082] Compare d 2 with the threshold value γ determined from the chi-square distribution according to the dimension of the measurement vector and the preset confidence level, when d 2 < γ, it is determined that the measurement data is valid.

[0083] Exemplarily, it is assumed that the vehicle's position at the current time should be point A according to the state prediction at the last time. However, the new GPS measurement data shows that the vehicle is at point B. The difference between points A and B, i.e. the residual, is calculated. Then, the statistical quantity d 2 is calculated by using the Mahalanobis distance formula, which comprehensively considers the size of the residual and the prediction uncertainty. The statistical quantity d 2 considers the error range that may exist in each direction. The preset confidence level, such as 0.95, and a threshold value γ, for example, 5.99, are obtained from the chi-square distribution table according to the preset confidence level and the dimension of the measurement data. If the calculated Mahalanobis distance square d 2is less than 5.99, it indicates that although there is a bias between the measured value and the predicted value, the bias is within an acceptable statistical range, thus the GPS data is determined to be valid. If the GPS suddenly reports a C-point far away from the vehicle by several kilometers due to signal interference, the calculated d 2 will be a very large value, far exceeding 5.99, and the data is determined to be invalid outliers and discarded, thus avoiding the pollution of the overall state estimation by the false data.

[0084] In a preferred embodiment, the weighted fusion of the posterior state estimates of all parallel models comprises:

[0085] The optimal fused state estimate is calculated by the formula where is the state estimate at time k, is the posterior state estimate of the jth model at time k, is the posterior probability of the jth model at time k, and the sum is taken over all models j.

[0086] At any time k, the four parallel models each give their estimate of the current state. For example, the normal transport model estimates the vehicle speed to be 90 km / h and the temperature to be 4.1°C; while the refrigeration failure model estimates the vehicle speed to be 90 km / h but the temperature to have risen to 4.8°C. By comparing the agreement of each model's prediction with the actual data, the probability or weight of each model is calculated. If the vehicle is currently driving smoothly and the temperature is normal, the probability of the normal transport model may be as high as 0.97, while the probability of the refrigeration failure model may be only 0.01, and the probabilities of the other two models add up to 0.02. The state estimate is obtained by weighted averaging, and the final temperature estimate will be 0.97 times 4.1°C, plus 0.01 times 4.8°C, plus the temperature estimates of the other two models with very low weights. The result will be very close to 4.1°C, as shown in Figure 4 because it is mainly determined by the most credible model. If the situation changes, for example, the temperature starts to rise abnormally, the probability of the refrigeration failure model will quickly increase, and its influence in the weighted average will also increase, so that the final fused state can quickly and accurately reflect this change.

[0087] Embodiment two proposes a fresh food transportation whole-process tracking system based on RFID, which comprises the following units:

[0088] The acquisition unit is used for acquiring the fresh goods identification, temperature and humidity collected by the RFID tag, and the GPS position and vehicle operation state data sent by the vehicle terminal; a coupling state vector is constructed, which includes the three-dimensional position, speed, internal temperature and humidity of the goods, and the remaining shelf life of the goods based on the Arrhenius equation;

[0089] The model construction unit is used for constructing a parallel system model covering possible states of the whole transportation process, and the model includes a normal transportation model, an abnormal stop model, a refrigeration failure temperature rise model and a loading and unloading operation model; the state estimation of each model is interactively mixed based on the posterior probability of each model at the previous moment, and the initial state for the prediction of each model is generated;

[0090] The prediction unit is used for performing Kalman filter state prediction for each model after the interaction and mixing, wherein the evolution of the remaining shelf life state follows a dynamic equation coupled with the temperature state, and the process noise covariance matrix Q is dynamically configured according to information including the road type corresponding to the current position of the vehicle and the real-time traffic flow;

[0091] The estimation unit is used for receiving new measurement data, performing data validity test through a verification gate based on Mahalanobis distance, updating the state estimation and likelihood of each model using the valid data, and calculating the posterior probability of each model; the posterior state estimation of all parallel models is weighted and fused according to the updated model posterior probability, and the optimal state estimation at the current moment is output.

[0092] Optionally, the coupling state vector is defined as X k =[x k ,y k ,z k ,vx k ,vy k ,vz k ,T k ,H k ,SL k ] T , wherein [x k ,y k ,z k ] is the three-dimensional position coordinate of the vehicle at time k, [vx k ,vy k ,vz k ] is the three-dimensional speed of the vehicle at time k, T k and H k are the temperature and humidity inside the goods at time k, and SL k is the remaining shelf life of the goods at time k.

[0093] Optionally, the remaining shelf life SL kThe state evolution of SL follows a kinetic model based on the Arrhenius equation: SL k = SL k-1 -Δt*k ref *exp[(Ea / R)*(1 / T ref -1 / T k )], where Δt is the sampling time interval, k ref is the quality deterioration rate constant at the reference temperature T ref , Ea is the activation energy of the deterioration reaction, R is the ideal gas constant, and T k is the internal temperature state of the cargo at time k.

[0094] Optionally, the parallel system model is specifically defined as follows:

[0095] Normal transportation model: the vehicle position and speed follow the constant speed kinematics model, and the temperature and humidity are stable around the set value;

[0096] Abnormal stop model: the vehicle speed is zero, the position remains unchanged, and the temperature and humidity remain stable;

[0097] Refrigeration failure temperature rise model: the vehicle motion state is consistent with the normal transportation model, and the temperature state follows a kinetic model representing continuous temperature rise;

[0098] Loading and unloading operation model: the vehicle speed is zero, and the temperature and humidity state model includes a function for responding to temperature and humidity step changes caused by the opening and closing of the box door.

[0099] Optionally, the process noise covariance matrix Q is dynamically configured according to information including the road type corresponding to the current position of the vehicle and the real-time traffic flow, including:

[0100] When it is determined that the vehicle is in a stable driving state, the process noise parameters corresponding to the position and speed components in Q are set to a set of preset low values;

[0101] When it is determined that the vehicle is in a non-stable driving state, the process noise parameters corresponding to the position and speed components in Q are set to a set of preset high values.

[0102] Optionally, the data validity test is performed through a verification gate based on the Mahalanobis distance, including:

[0103] Calculate the residual and its covariance matrix S k between the new measurement value and the predicted measurement value generated based on the state prediction;

[0104] Calculate the square of the Mahalanobis distance

[0105] d 2When d 2 <γ, the measurement data is determined to be valid.

[0106] Optionally, the weighted fusion of the posterior state estimates of all parallel models comprises:

[0107] The optimal state estimate after fusion is calculated by formula wherein is the state estimate at time k, is the posterior state estimate of the jth model at time k, is the posterior probability of the jth model at time k, and summation is performed for all models j.

[0108] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those ordinarily skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. In addition, the various different embodiments of the embodiments of the present application can also be combined arbitrarily, as long as it does not deviate from the idea of the embodiments of the present application, it should also be considered as the disclosed content of the embodiments of the present application.

Claims

1. A method for tracking the entire transportation process of fresh produce based on RFID, characterized in that, Includes the following steps: Acquire fresh produce identification, temperature, and humidity data collected by RFID tags, as well as GPS location and vehicle operating status data sent by the vehicle terminal; construct a coupled state vector that includes the vehicle's three-dimensional position, speed, internal temperature and humidity of the goods, and the remaining shelf life of the goods based on the Arrhenius equation; A parallel system model covering all possible states of the transportation process is constructed. The system model includes: a normal transportation model, an abnormal stagnation model, a refrigeration failure temperature rise model, and a loading and unloading operation model. Based on the posterior probability of each model at the previous time step, the state estimates of each model are interactively mixed to generate an initial state for each model for the current round of prediction. For each interactive hybrid model, Kalman filtering state prediction is performed, where the evolution of the remaining shelf life state follows a dynamic equation coupled with the temperature state, and the process noise covariance matrix Q is dynamically configured based on information including the road type corresponding to the vehicle's current location and real-time traffic flow. Upon receiving new measurement data, the system performs a data validity check using a validation gate based on Mahalanobis distance. Then, it updates the state estimates and likelihoods of each model using the valid data and calculates the posterior probability of each model. Based on the updated model posterior probabilities, the system performs a weighted fusion of the posterior state estimates of all parallel models and outputs the optimal state estimate at the current time step. The coupling state vector is defined as ,in Let K be the three-dimensional position coordinates of the vehicle at time k. Let K be the three-dimensional velocity of the vehicle at time k. and Let K represent the temperature and humidity inside the cargo at time k. Let k be the remaining shelf life of the goods at time k. The remaining shelf life of the goods The state evolution follows a dynamic model based on the Arrhenius equation: ,in The sampling time interval, Reference temperature The rate constant of degradation is given by Ea, where Ea is the activation energy of the degradation reaction and R is the ideal gas constant. Let k be the internal temperature state of the cargo at time k.

2. The method according to claim 1, characterized in that, The parallel system model is specifically defined as follows: Normal transportation model: Vehicle position and speed follow a constant velocity kinematic model, and temperature and humidity remain stable around the set values; Abnormal stationary model: The vehicle speed is zero, the position remains unchanged, and the temperature and humidity remain stable; Cooling failure temperature rise model: The vehicle motion state is consistent with the normal transportation model, and the temperature state follows the dynamic model that characterizes the continuous temperature rise; Loading and unloading operation model: The vehicle speed is zero, and the temperature and humidity state model includes functions to respond to the step changes in temperature and humidity caused by the opening and closing of the box door.

3. The method according to claim 1, characterized in that, The dynamic configuration of the process noise covariance matrix Q based on information including the road type corresponding to the vehicle's current location and real-time traffic flow includes: When the vehicle is determined to be in a stable driving state, the process noise parameters corresponding to the position and velocity components in Q are set to a set of preset low values. When the vehicle is determined to be in an unstable driving state, the process noise parameters corresponding to the position and velocity components in Q are set to a set of preset high values.

4. The method according to claim 1, characterized in that, The data validity verification using a validation gate based on Mahalanobis distance includes: Calculate the residual between the new measurement and the predicted measurement generated based on state prediction. and its covariance matrix ; Calculate the square of the Mahalanobis distance ; Will The threshold determined from the chi-square distribution based on the measurement vector dimension and pre-set confidence level. When comparing, When the measurement data is valid, it is determined that the measurement data is valid.

5. The method according to claim 1, characterized in that, The weighted fusion of the posterior state estimates of all parallel models includes: Through formula Calculate the optimal state estimate after fusion, where For the state estimation at time k, Let be the posterior state estimate of the j-th model at time k. Let be the posterior probability of the j-th model at time k, summed over all models j.

6. A fresh produce transportation end-to-end tracking system based on RFID, characterized in that, Includes the following units: The data acquisition unit is used to acquire the identification, temperature, and humidity of fresh produce collected by RFID tags, as well as GPS location and vehicle operating status data sent by the vehicle terminal; and to construct a coupled state vector that includes the vehicle's three-dimensional position, speed, internal temperature and humidity of the goods, and the remaining shelf life of the goods based on the Arrhenius equation. The model building unit is used to build a parallel system model covering all possible states of the transportation process. The model includes: a normal transportation model, an abnormal stagnation model, a refrigeration failure temperature rise model, and a loading and unloading operation model. Based on the posterior probability of each model at the previous time step, the state estimates of each model are interactively mixed to generate an initial state for each model for this round of prediction. The prediction unit is used to perform Kalman filter state prediction for each interactively mixed model, where the evolution of the remaining shelf life state follows a dynamic equation coupled with the temperature state, and the process noise covariance matrix Q is dynamically configured based on information including the road type corresponding to the vehicle's current location and real-time traffic flow. The estimation unit receives new measurement data, performs data validity checks through a validation gate based on Mahalanobis distance, updates the state estimates and likelihoods of each model using valid data, and calculates the posterior probability of each model. Based on the updated model posterior probabilities, it performs weighted fusion of the posterior state estimates of all parallel models and outputs the optimal state estimate at the current time. The coupling state vector is defined as ,in Let K be the three-dimensional position coordinates of the vehicle at time k. Let K be the three-dimensional velocity of the vehicle at time k. and Let K represent the temperature and humidity inside the cargo at time k. Let k be the remaining shelf life of the goods at time k. The remaining shelf life of the goods The state evolution follows a dynamic model based on the Arrhenius equation: ,in The sampling time interval, Reference temperature The rate constant of degradation is given by Ea, where Ea is the activation energy of the degradation reaction and R is the ideal gas constant. Let k be the internal temperature state of the cargo at time k.

Images