Characteristic gas detection method for cold-chain fruit and vegetable logistics transport vehicle

Abstract

The invention discloses a characteristic gas detection method for a cold-chain fruit and vegetable logistics transport vehicle, and the method comprises the steps: intermittently introducing oxygen and ammonia gas into a cold-chain logistics transport vehicle model, collecting the real-time concentration of oxygen and ammonia gas through an oxygen sensor and an ammonia gas sensor, obtaining the output voltage signals of the oxygen sensor and the ammonia gas sensor in a set number group, and carrying out the detection of the characteristic gas in the cold-chain logistics transport vehicle model; parts of voltage signals are randomly selected to serve as a training set, the rest voltage signals serve as a prediction set, then classification prediction is carried out based on an extreme learning machine optimized by a white whale algorithm, and data are calculated and classified. The stability, accuracy and intelligence of characteristic gas detection of the cold-chain fruit and vegetable logistics transport vehicle are improved through classification prediction of the extreme learning machine optimized based on the white whale algorithm, and real-time detection of the health state of fruits and vegetables is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of characteristic gas detection for cold chain fruit and vegetable logistics transport vehicles, and in particular to a characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles. Background Art

[0002] my country is a major producer, consumer and exporter of fruits and vegetables. In order to reduce the loss of fruits and vegetables during transportation and realize real-time detection of the health status of fruits and vegetables, it is becoming increasingly important to optimize the existing cold chain logistics and transportation management system. The existing characteristic gas detection algorithms have their own shortcomings. At present, the algorithms based on sensor arrays and pattern recognition are easily affected by sensor drift and environmental interference, and the data processing complexity is high. Spectral analysis algorithms rely on expensive and complex equipment and have strict requirements on environmental conditions. Electrochemical sensors and chemical sensor algorithms have long response times, poor selectivity, and limited sensor life, which requires regular replacement. Machine learning and deep learning algorithms require a large amount of high-quality labeled data, complex models, and training and deployment require high-performance computing resources, and there is a risk of overfitting. In summary, the existing algorithms have varying degrees of deficiencies in detection accuracy, response time, equipment cost, and environmental adaptability.

[0003] The Beluga Optimizer (BWO) is a novel optimization technique based on the beluga’s predation behavior, simulating the natural behavior of beluga whales in the ocean through sonic positioning and cooperative predation. It is known for its fast convergence and strong global search capabilities, and can effectively find the optimal solution to a problem. The Extreme Learning Machine (ELM) is a single hidden layer feedforward neural network that has attracted widespread attention due to its fast training speed and strong generalization ability. ELM avoids the complex iterative training process in traditional neural networks by randomly generating weights from the input layer to the hidden layer and completing the training by minimizing the weights of the output layer.

[0004] The BWO-ELM algorithm, namely the Beluga Whale Optimization Extreme Learning Machine, is a hybrid intelligent algorithm that combines the advantages of the Beluga Whale Optimization Algorithm and the Extreme Learning Machine to improve the performance and accuracy of the prediction model. Summary of the invention

[0005] In view of this, the present invention aims to improve the stability, accuracy and intelligence of characteristic gas detection of cold chain fruit and vegetable logistics transport vehicles through extreme learning machine classification prediction based on Beluga algorithm optimization, realize real-time detection of the health status of fruits and vegetables, and provide reliable detection results for optimizing the existing cold chain logistics and transportation management system.

[0006] The characteristic gas detection method for a cold chain fruit and vegetable logistics transport vehicle of the present invention comprises the following steps: S1. placing the compartment of the cold chain logistics transport vehicle model in the air, and collecting the voltage values ​​output by the oxygen and ammonia sensor driving circuits for a set time; S2. closing the compartment and keeping it sealed, introducing oxygen into the compartment for a set time, placing a container containing ammonia water in the compartment, and continuing to collect the voltage values ​​output by the oxygen and ammonia sensor driving circuits for a set time; S3. ending the oxygen supply, taking out the container containing ammonia water, opening the compartment door, and continuing to collect the voltage values ​​output by the oxygen and ammonia sensor driving circuits for a set time; S4. repeating the above steps to obtain several groups of oxygen and ammonia sensor output voltage signals; S5. randomly selecting some groups of voltage signals from the voltage signals obtained in step S4 as training sets, and the remaining groups of voltage signals as prediction sets, and then calculating and classifying the data based on an extreme learning machine optimized by the Beluga algorithm.

[0007] Further, in step S5, the specific steps of the extreme learning machine algorithm optimized by the Beluga algorithm include: S51. Randomly initialize the Beluga population, each individual represents a possible ELM parameter combination; S52. Perform fitness calculation; for each individual, use the corresponding ELM parameter to train the model and calculate its prediction error as the fitness value of the individual; S53. The algorithm performs selection and crossover, selects individuals with high fitness according to the fitness value for crossover operation, and generates new candidate solutions; S54. Updates the population, compares the newly generated candidate solution with the current population, and updates the population with individuals with higher fitness; S55. Repeats steps S51-S54 in a loop until a predetermined number of iterations T is reached or the fitness value converges; S56. Uses the optimized parameters to train the final extreme learning machine model for classification or regression tasks; the optimized parameters include the number of hidden layer neurons n h , input weight matrix W in And the output weight matrix W out ; S57. Use the trained ELM model to make predictions and output the final results.

[0008] Further, in step s52, the fitness function may be a mean square error, and the specific formula is: Where n represents the number of samples, that is, the total number of data points in the data set. i Represents the i-th actual value, that is, the true target value. Represents the i-th predicted value, that is, the predicted result output by the model.

[0009] Furthermore, the set duration is 4 minutes.

[0010] Furthermore, in step s2, before oxygen is injected into the vehicle compartment through the air pipe, in order to ensure sealing, the through hole through which the air pipe passes through the side wall of the vehicle compartment is sealed with sealing putty.

[0011] Furthermore, in step s2, a low concentration of ammonia water with a concentration of 0.1% is selected.

[0012] Furthermore, the oxygen sensor and the ammonia sensor collect voltage signals every 2 seconds, and 100 sampling points are set for each set of data.

[0013] Furthermore, the oxygen sensor AOX3000 and the ammonia sensor JEC-NH3 were used to collect the real-time concentrations of oxygen and ammonia.

[0014] Beneficial effects of the invention: The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles proposed in the invention has the advantages of high precision, fast response time, and low cost of data acquisition equipment. The method is conducive to optimizing the cold chain logistics transportation management system, reducing the loss and waste of fruit and vegetable cold chain transportation, and improving the level of intelligent management, providing technical support for the modernization and intelligent development of cold chain logistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0016] Figure 1 It is a flowchart of the BWO-ELM algorithm in the present invention.

[0017] Figure 2 Schematic diagram of the output voltage of the ammonia sensor in the present invention.

[0018] Figure 3 Schematic diagram of the output voltage of the oxygen sensor in the present invention.

[0019] Figure 4 This is a comparison chart of the prediction results of the BWO-ELM used in the present invention and the traditional SVM training set.

[0020] Figure 5 This is a comparison chart of the prediction results of the BWO-ELM and traditional SVM test sets used in the present invention.

[0021] Figure 6 This is the confusion matrix of the BWO-ELM and traditional SVM training sets used in the present invention.

[0022] Figure 7 This is the confusion matrix of the BWO-ELM and traditional SVM test sets used in the present invention.

[0023] Figure 8 It is the fitness curve of the BWO-ELM algorithm in the present invention. DETAILED DESCRIPTION

[0024] like Figure 1 As shown, a characteristic gas detection method for a cold chain fruit and vegetable logistics transport vehicle in this embodiment includes the following steps:

[0025] S1. Place the compartment of the cold chain logistics transport vehicle model in the air, use the oxygen sensor AOX3000 and the ammonia sensor JEC-NH3 to collect the real-time concentration of oxygen and ammonia, and collect the voltage value output by the sensor drive circuit for four minutes.

[0026] S2. Close the compartment and keep it sealed, introduce oxygen into the compartment for four minutes, place a container containing ammonia water in the compartment, and continue to collect the voltage values ​​output by the oxygen and ammonia sensor drive circuit for four minutes; in this step, use a portable oxygen cylinder with an oxygen purity of ≥99.5% to perform repeatability testing, and discharge the oxygen in the oxygen cylinder into the cold chain logistics transport vehicle model compartment through a gas pipe. In order to ensure the sealing, the gas pipe is sealed with sealing glue around it. Put ammonia water in a beaker and keep it still in the compartment of the cold chain logistics transport vehicle model. The volatilization of ammonia water simulates the situation that the fruit slowly releases ammonia when it is corrupted. Collect sensor data to verify whether it can identify the change of ammonia concentration in the compartment. At the same time, to ensure the safety of the experiment, a low concentration of ammonia water with a concentration of 0.1% is selected.

[0027] S3. End the oxygen supply, take out the container containing ammonia water, open the car door, and continue to collect the voltage values ​​output by the oxygen and ammonia sensor drive circuits for 4 minutes.

[0028] S4. Repeat the above steps to obtain 200 sets of oxygen and ammonia sensor output voltage signals; the sensor collects voltage signals every 2 seconds, and 100 sampling points are set for each set of data, where the standard deviation mean of the oxygen group is 0.3629, and the standard deviation mean of the ammonia group is 0.0354. The 30 sets of data within 0-200S are as follows Figure 1 and Figure 2 shown.

[0029] S5. Randomly select 140 groups of voltage signals as training sets and 60 groups of voltage signals as prediction sets from the voltage signals obtained in step S4, and then calculate and classify the data based on the extreme learning machine optimized by the Beluga algorithm.

[0030] The specific steps of the extreme learning machine algorithm optimized by the white whale algorithm in this embodiment include: S51. Randomly initialize the white whale population, each individual represents a possible ELM parameter combination; for example, the number of hidden layer neurons n h , input weight matrix W inetc.; S52. Perform fitness calculation; for each individual, use the corresponding ELM parameter to train the model and calculate its prediction error as the individual's fitness value; the fitness function is the mean square error, and the specific formula is: Where n represents the number of samples, that is, the total number of data points in the data set. i Represents the i-th actual value, that is, the true target value. Represents the i-th predicted value, that is, the predicted result output by the model.

[0031] S53. After the fitness calculation, the algorithm performs selection and crossover, and selects individuals with high fitness according to the fitness value for crossover operation to generate new candidate solutions; this step simulates the cooperative predation behavior of beluga whales. Then mutation is performed to mutate some candidate solutions to increase the diversity of the population and prevent falling into the local optimal solution. Mutation probability p m Determines the frequency of mutation operations.

[0032] S54. Update the population, compare the newly generated candidate solution with the current population, and update the population with individuals with higher fitness.

[0033] S55. Repeat steps S51-S54 in a loop, through iterations of fitness calculation, selection and crossover, mutation and population update, until a predetermined number of iterations T is reached or the fitness value converges.

[0034] S56. After reaching the stopping condition, the final extreme learning machine model is trained using the optimized parameters for classification or regression tasks; the optimized parameters include the number of hidden layer neurons n h , input weight matrix W in And the output weight matrix W out .

[0035] S57. Finally, use the trained ELM model to make predictions and output the final results.

[0036] The BWO-ELM and traditional SVM gas classification prediction models were compared and tested, where 1 means that the data set is oxygen, 2 means that the data set is ammonia, the predicted value refers to the data type result predicted by the model, and the true value refers to the actual type of the data. If the model misjudges the type of a certain data set, the blue circle at the corresponding position of the data jumps to the opposite side of the true value. The simulation results are shown in Figure 1. Figure 4-7 shown.

[0037] like Figure 8is the fitness curve of the BWO-ELM algorithm. The initial fitness value is high, close to 1.1, indicating that the initial solution performs well, but there is still room for optimization. In the first few iterations, the fitness value drops rapidly to around 0.7, showing that the Beluga optimization algorithm can quickly find a better combination of parameters in the early stage and greatly improve the model performance. Subsequently, the fitness value tends to stabilize and remains at around 0.7, indicating that the algorithm has found the global optimal solution or a region close to the optimal solution. The stability of the fitness value indicates the convergence and effectiveness of the algorithm. The final fitness value is 0.7, which is a significant improvement compared to the initial value, indicating the effectiveness of the BWO-ELM algorithm in improving the performance of the ELM model.

[0038] The simulation results of the BWO-ELM model and the traditional SVM model are compared in Table 1. The gas classification prediction model based on BWO-ELM performs well in the simulation. The prediction accuracy on the training set is 99.29%, and the prediction accuracy on the test set is 100%, showing extremely high classification accuracy and generalization ability. The confusion matrix of the test set shows that all samples are correctly classified; in the training set, 2 samples are misclassified with accuracy rates of 100% and 98.6%, respectively. In contrast, the training set and test set accuracy rates of the traditional SVM model are 98.21% and 97.5%, respectively, and the confusion matrix shows more misclassifications. It can be concluded that the BWO-ELM model has higher accuracy and stability in the detection of characteristic gases in cold chain fruit and vegetable logistics transport vehicles than the traditional SVM model.

[0039] Table 1 Comparison of simulation results between BWO-ELM and SVM models

[0040] Model BWO-ELM BWO-ELM SVM SVM Dataset Training set Test Set Training set Test Set Accuracy / % 99.2857 100 98.2143 97.5 Actual oxygen / group 140 60 140 60 Actual ammonia / group 140 60 140 61 Model Classification Oxygen / Group 139 60 135 59 Model Classification Ammonia / Group 141 60 145 61 Mistakenly identified as oxygen / group 0 1 0 2 Mistakenly identified as ammonia / group 2 0 5 1

[0041] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the protection scope of the present invention.

Claims

1. A characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles, characterized in that: The following steps are involved: S1. Place the compartment of the cold chain logistics transport vehicle model in the air and collect the voltage values ​​output by the oxygen and ammonia sensor drive circuits for a set period of time; S2. The compartment is closed and kept sealed, oxygen is introduced into the compartment for a set time, and a container containing ammonia is placed in the compartment, and the voltage value output by the oxygen and ammonia sensor drive circuit is continuously collected for a set time; S3. End the oxygen supply, take out the container containing ammonia water, open the door of the carriage, and continue to collect the voltage value output by the oxygen and ammonia sensor drive circuit for a set time; S4. Repeat the above steps to obtain several groups of oxygen and ammonia sensor output voltage signals; S5. Randomly select some groups of voltage signals from the voltage signals obtained in step S4 as training sets, and the remaining groups of voltage signals as prediction sets, and then calculate and classify the data based on the extreme learning machine optimized by the Beluga algorithm.

2. The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles according to claim 1 is characterized in that: In step S5, the specific steps of the extreme learning machine algorithm optimized by the White Whale algorithm include: S51. Randomly initialize the beluga population, each individual represents a possible ELM parameter combination; S52. Perform fitness calculation; for each individual, use the corresponding ELM parameter to train the model and calculate its prediction error as the individual's fitness value; S53. The algorithm performs selection and crossover, selects individuals with high fitness according to the fitness value, performs crossover operation, and generates new candidate solutions; S54. Update the population, compare the newly generated candidate solution with the current population, and update the population with individuals with higher fitness; S55. Repeat steps S51-S54 in a loop until a predetermined number of iterations T is reached or the fitness value converges; S56. Use the optimized parameters to train the final extreme learning machine model for classification or regression tasks; the optimized parameters include the number of hidden layer neurons n h , input weight matrix W in And the output weight matrix W out ; S57. Use the trained ELM model to make predictions and output the final results.

3. The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles according to claim 2 is characterized in that: In step s52, the fitness function is the mean square error, and the specific formula is: Among them, n represents the number of samples, that is, the total number of data points in the data set; y i represents the ith actual value, that is, the true target value; Represents the i-th predicted value, that is, the predicted result output by the model.

4. The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles according to claim 3 is characterized in that: The set duration is 4 minutes.

5. The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles according to claim 4 is characterized in that: In step s2, before oxygen is injected into the vehicle compartment through the air supply pipe, the through hole through which the air supply pipe passes through the side wall of the vehicle compartment is sealed with sealing putty to ensure sealing.

6. The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles according to claim 5 is characterized in that: In step s2, a low concentration of ammonia water with a concentration of 0.1% is selected.

7. The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles according to claim 6 is characterized in that: The oxygen and ammonia sensors collect voltage signals every 2 seconds, and 100 sampling points are set for each set of data.

8. The characteristic gas detection method for cold chain fruit and vegetable logistics transport vehicles according to claim 1 is characterized in that: The oxygen sensor AOX3000 and the ammonia sensor JEC-NH3 are used to collect the real-time concentrations of oxygen and ammonia.

Images