A method for predicting lane-level conflicts in expressway weaving zones based on real-time traffic flow data

Abstract

This invention discloses a lane-level conflict prediction method for expressway weaving areas based on real-time traffic flow data, comprising the following steps: S1, selecting a suitable expressway weaving area as the survey location and conducting on-site investigation; S2, collecting traffic flow data and traffic conflict data for each lane, and defining thresholds for different conflict types; S3, building a traffic simulation model using the traffic parameters obtained from the survey, defining different conflict types according to the conflict thresholds obtained in step S2, and obtaining traffic conflict data for different types; S4, extracting feature variables from the data obtained in step S3, and constructing datasets for prediction models of different conflict types; S5, constructing a conflict analysis model; S6, constructing a conflict prediction model. This invention can achieve real-time lane-level traffic conflict prediction in expressway weaving areas, thereby providing more accurate lane-level conflict risk information for vehicles in each lane, helping drivers choose safer lanes and routes.

Description

Technical Field

[0001] This invention relates to a method for predicting lane-level traffic conflicts in expressway weaving areas, specifically a method for predicting lane-level conflicts in expressway weaving areas based on real-time traffic flow data. Background Technology

[0002] The prerequisite for solving road traffic safety problems is to conduct traffic safety analysis. Through traffic safety analysis, we can fully understand the influencing factors and evolution mechanisms of accidents, and thus provide a basis for improving traffic safety.

[0003] Traditional urban expressway traffic safety analysis largely relies on historical traffic accident data. Since accident data is mostly obtained through on-site investigations and manual recording, the data itself has a degree of subjectivity and the possibility of missing minor accidents, thus limiting the effectiveness of this method. Traffic Conflict Technique (TCT), as an alternative method to traffic accident analysis, analyzes incidents before they occur, offering a proactive approach. Existing research has demonstrated a correlation between traffic conflicts and traffic accidents, showing that both share similar evolutionary mechanisms.

[0004] Based on this, traffic conflict analysis has been widely applied in traffic safety analysis in recent years, yielding numerous research results. Traffic conflict prediction based on real-time traffic flow data is one of the important bases for achieving proactive road traffic safety management. Real-time traffic conflict prediction at the lane level can further provide drivers in each lane with more accurate lane-level conflict risk information, helping them choose safer lanes and routes. Summary of the Invention

[0005] The problem to be solved by this invention is to provide a lane-level conflict prediction method for expressway weaving areas based on real-time traffic flow data. By establishing a lane-level traffic conflict prediction model for expressway weaving areas, the method predicts the number of different types of traffic conflicts occurring in each lane of the weaving area within a short period of time, thus providing a traffic safety basis for realizing lane-level navigation for drivers.

[0006] This invention adopts the following technical solution: a method for predicting lane-level conflicts in expressway weaving areas based on real-time traffic flow data, comprising the following steps:

[0007] S1. Conduct on-site investigations of the weaving areas of the expressways to be tested and obtain real-time video of traffic conditions.

[0008] S2. Number each lane in the expressway weaving area, collect traffic flow data and traffic conflict data for each lane, and set collision time (TTC) thresholds based on different conflict types.

[0009] S3. Build a traffic simulation model and perform data calibration. Obtain traffic flow data and vehicle trajectory data. Based on the conflict angle and the conflict threshold set in step S2, define different conflict types and use conflict analysis software to obtain different types of traffic conflict data.

[0010] S4. Based on the traffic flow data obtained in step S3, extract feature variables and construct datasets for prediction models of different conflict types.

[0011] S5. Based on the negative binomial regression method, construct traffic conflict analysis models for different conflict types and analyze the impact of different influencing factors on the number of traffic conflicts.

[0012] S6. Based on machine learning methods, construct traffic conflict prediction models for different conflict types, predict the number of various conflicts that will occur in each lane in the future, and provide lane-level conflict risk alerts for drivers in each lane.

[0013] Furthermore, in step S1, a drone is used to film and record the weaving area of ​​the expressway under test to obtain video of traffic operation.

[0014] Furthermore, in step S2, traffic conflict data for each lane is collected using video tracking software, and collision time (TTC) thresholds for different conflict types are determined based on the Pareto curve. The conflict types include: severe side impact conflict, general side impact conflict, severe rear-end collision conflict, and general rear-end collision conflict.

[0015] Furthermore, in step S3, a simulation model is built using VISSIM traffic simulation software, and the collected traffic flow data is used for calibration. The parameters used for calibration include: geometric design parameters of the weaving area, vehicle composition, traffic volume, vehicle speed distribution, vehicle car-following model parameters, and vehicle lane-changing model parameters.

[0016] Furthermore, in step S4, three types of feature variables are extracted: weaving zone feature variables, conflict lane traffic flow feature variables, and conflict lane and adjacent lane difference variables.

[0017] Furthermore, in step S5, the variance inflation factor (VIF) of each feature variable is calculated to screen variables and build traffic conflict analysis models for different conflict types. The formula for calculating the variance inflation factor (VIF) is as follows:

[0018]

[0019] in, The independent variable is x i Other independent variables x j The goodness of fit obtained after performing linear regression.

[0020] Specifically, a VIF value greater than 10 is considered to indicate multicollinearity. Feature variables with VIF values ​​greater than 10 are successively removed until the VIF values ​​of all remaining feature variables are less than 10. This indicates that there is mild multicollinearity among the remaining variables, which will not interfere with the interpretability of the analytical model.

[0021] The feature variables that have undergone multicollinearity screening will be used in subsequent traffic conflict analysis models.

[0022] The negative binomial regression method is used to construct traffic conflict analysis models for different conflict types, as shown in the following formula:

[0023] N i =exp(β0+β1x1+β2x2+...+β) n x n )

[0024] Where, N i Let β0, β1, ..., β be the number of conflicts of different conflict types. n Let x1, x2, ..., x be the regression coefficients. n These are the feature variables extracted in step S4.

[0025] Furthermore, in step S6, different machine learning methods are used to construct prediction models for different conflict types and evaluate the prediction results. The model with the best prediction performance is selected by calculating the root mean square error (RSME) and the mean absolute error (MAE).

[0026] Root mean square error (RSME) and mean absolute error (MAE) are commonly used evaluation metrics for regression models, reflecting the error level between measured values ​​and model predictions. The smaller the RSME and MAE, the higher the accuracy of the model prediction.

[0027] The formula for calculating the root mean square error is as follows:

[0028]

[0029] The formula for calculating the mean absolute error is as follows:

[0030]

[0031] Where n is the number of actual values, Y i It is the true value of the i-th actual value. It is the predicted value of the i-th actual value.

[0032] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0033] 1. This invention uses real-time traffic flow data to predict traffic conflicts. Real-time traffic flow data provides real-time traffic flow operation status and road condition information, making the prediction results of traffic conflicts more timely and accurate.

[0034] 2. Before conducting conflict prediction, this invention constructs a negative binomial regression model for traffic conflict analysis. The negative binomial regression model is suitable for count data analysis and can more intuitively explain the impact of changes in characteristic variables on the number of traffic conflicts.

[0035] 3. This invention considers the differences in traffic conflicts of different categories and severity when making conflict predictions, and builds prediction models for four different types of traffic conflicts, making the prediction results more accurate.

[0036] 4. This invention uses lane-level feature variables to predict lane-level traffic conflicts, which can provide more accurate lane-level conflict risk information for vehicles in each lane, helping drivers choose safer lanes and routes. Attached Figure Description

[0037] Figure 1 This is a flowchart of the expressway lane-level conflict prediction method based on real-time traffic flow data according to the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] This invention relates to a lane-level conflict prediction method for expressway weaving areas based on real-time traffic flow data, such as... Figure 1 As shown, it includes the following steps:

[0040] S1. Conduct on-site investigations of the weaving areas of the expressways to be tested and obtain real-time video of traffic conditions.

[0041] S2. Number each lane in the expressway weaving area, collect traffic flow data and traffic conflict data for each lane, and set collision time (TTC) thresholds based on different conflict types.

[0042] S3. Build a traffic simulation model and calibrate it using the collected traffic flow data. Based on the conflict angle and the conflict threshold set in step S2, define different conflict types and obtain different types of traffic conflict data.

[0043] S4. Based on the data obtained in step S3, extract feature variables and construct datasets for prediction models of different conflict types;

[0044] S5. Construct traffic conflict analysis models for different conflict types and analyze the impact of different influencing factors on the number of traffic conflicts.

[0045] S6. Based on machine learning methods, construct traffic conflict prediction models for different conflict types, predict the number of various conflicts that will occur in each lane in the future, and provide lane-level conflict risk alerts for drivers in each lane.

[0046] In one embodiment of the present invention, step S1 involves selecting a certain expressway weaving area in Nanjing City, Jiangsu Province as the investigation location, and using a drone equipped with a camera to record video of the traffic operation status of the selected expressway weaving area.

[0047] Step S2: Number each lane in the expressway weaving area, collect traffic flow data and traffic conflict data for each lane, and determine the collision time (TTC) threshold for different conflict types based on the Pareto curve. The TTC thresholds for different conflict types are shown in Table 1.

[0048] Table 1 Collision Time-To-Cost (TTC) Thresholds for Different Collision Types

[0049]

[0050] Step S3: Build a traffic simulation model using the traffic parameters obtained from the survey, define different conflict types based on the conflict thresholds obtained in step S2, and obtain traffic conflict data of different types.

[0051] In this embodiment, a simulation model is built using VISSIM traffic simulation software and calibrated using traffic flow parameters obtained from field surveys. The parameters used for calibration include geometric design parameters of the weaving zone, vehicle composition, traffic volume, vehicle speed distribution, vehicle following model parameters, and vehicle lane-changing model parameters. Based on the conflict angle and the collision time time (TTC) threshold defined in step 2, four conflict types are classified, and traffic conflict data of different types are obtained.

[0052] Step S4: Based on the data obtained in Step 3, extract feature variables and construct datasets for prediction models of different conflict types.

[0053] Three types of characteristic variables were extracted: weaving zone characteristic variables, conflict lane traffic flow characteristic variables, and conflict lane and adjacent lane difference variables.

[0054] Among them, the weaving zone characteristic variables include weaving zone length, speed limit, weaving ratio, and weaving flow ratio; the conflict lane traffic flow characteristic variables include traffic flow data upstream and downstream of the conflict lane; the conflict lane and adjacent lane difference variables include the differences in various traffic flow parameters upstream and downstream between the conflict lane and the adjacent lane.

[0055] Step S5, Construct a conflict analysis model: Build traffic conflict analysis models for different conflict types and analyze the impact of different influencing factors on the number of traffic conflicts.

[0056] In this embodiment, the formula for the severe side-impact traffic conflict analysis model is as follows:

[0057] N cz =exp(-1.52+0.021V) U +4.251TR U +0.008O_U+1.443TR D -0.012AvgS_D

[0058] +0.180AvgA_D+2.50VR±2.549DiffTR_U+0.036DiffAvgS_U

[0059] -0.313DiffAvgA_U)

[0060] Where, N cz This represents the number of serious side-impact traffic conflicts. V_U is the traffic volume upstream of the lane where the serious side-impact conflict occurred (veh / h), TR_U is the vehicle type ratio upstream of the lane where the serious side-impact conflict occurred, O_U is the occupancy rate upstream of the lane where the serious side-impact conflict occurred, TR_D is the vehicle type ratio downstream of the lane where the serious side-impact conflict occurred, AvgS_D is the average speed downstream of the lane where the serious side-impact conflict occurred (km / h), and AvgA_D is the average acceleration downstream of the lane where the serious side-impact conflict occurred (m / s²). 2 VR is the weaving flow ratio in the weaving zone; DiffTR_U is the difference in vehicle type ratio between the lane where a severe side collision occurs and the upstream of the adjacent lane; DiffAvgS_U is the difference in average speed (km / h) between the lane where a severe side collision occurs and the upstream of the adjacent lane; DiffAvgA_U is the difference in average acceleration (m / s²) between the lane where a severe side collision occurs and the upstream of the adjacent lane. 2 ).

[0061] The general model for analyzing traffic conflicts is as follows:

[0062] N cyb =exp(1.579+0.017V_U+0.568AvgA_U+1.461TR_D+0.489AvgA_D-

[0063] 0.001L+2.072R+1.661VR+0.008DiffV_U-1.554DiffTR_U+0.007DiffO_U+

[0064] 0.018DiffAvgS_U+0.168DiffAvgA_U)

[0065] Where, N cyb V_U represents the number of typical side-impact collisions, V_U represents the upstream traffic volume (veh / h) of the lane where the typical side-impact collision occurred, and AvgA_U represents the average upstream acceleration (m / s²) of the lane where the typical side-impact collision occurred. 2 TR_D is the ratio of vehicles downstream of the lane where a typical side impact collision occurs, and AvgA_D is the mean acceleration downstream of the lane where a typical side impact collision occurs (m / s²). 2 L is the length of the expressway weaving zone (m), R is the weaving ratio of the expressway weaving zone, VR is the weaving flow ratio of the expressway weaving zone, DiffV_U is the difference in traffic volume between the lane where the side collision occurs and the upstream of the adjacent lane (veh / h), DiffTR_U is the difference in vehicle type ratio between the lane where the side collision occurs and the upstream of the adjacent lane, DiffO_U is the difference in occupancy between the lane where the side collision occurs and the upstream of the adjacent lane, DiffAvgS_U is the difference in average speed between the lane where the side collision occurs and the upstream of the adjacent lane (veh / h), and DiffAvgA_U is the difference in average acceleration between the lane where the side collision occurs and the upstream of the adjacent lane (m / s²). 2 ).

[0066] The formula for analyzing severe rear-end collision traffic conflicts is as follows:

[0067] N zyz =exp(0.135+6.931TR) U +0.006O U +0.351AvgA U -0.020AvgS D +0.230AvgA D

[0068] -0.994VR -4.494DiffTR U +0.231DiffAvgA D )

[0069] Where, N zyz This represents the number of serious rear-end collisions, TR_U is the ratio of vehicle types upstream of the lane where the serious rear-end collision occurred, O_U is the upstream occupancy rate of the lane where the serious rear-end collision occurred, and AvgA_U is the average acceleration (m / s²) upstream of the lane where the serious rear-end collision occurred. 2AvgS_D is the mean speed (km / h) downstream of the lane where the serious rear-end collision occurred, and AvgA_D is the mean acceleration (m / s²) downstream of the lane where the serious rear-end collision occurred. 2 VR is the weaving flow ratio in the weaving zone, DiffTR_U is the difference in vehicle type ratio between the lane where the serious rear-end collision occurred and the upstream lane of the adjacent lane, and DiffAvgA_U is the difference in mean acceleration (m / s²) between the lane where the serious rear-end collision occurred and the upstream lane of the adjacent lane. 2 ).

[0070] The general rear-end collision traffic conflict analysis model has the following formula:

[0071] N zyb =exp(1.715+0.007V) U +1.974TR U +0.006O U +0.414AvgA U +0.673TR D

[0072] -0.016AvgS D +0.316AvgA D +1.085R +0.572VR +0.011DiffV U

[0073] -2.607DiffTR U -0.014DiffAvgS U +0.010DiffV D -0.032DiffO D

[0074] -0.353DiffAvgA D )

[0075] Where, N zyz V_U represents the number of typical rear-end collisions, V_U represents the traffic volume upstream of the lane where the typical rear-end collision occurred (veh / h), TR_U represents the vehicle type ratio upstream of the lane where the typical rear-end collision occurred, O_U represents the occupancy rate upstream of the lane where the typical rear-end collision occurred, and AvgA_U represents the average acceleration upstream of the lane where the typical rear-end collision occurred (m / s²). 2 TR_D is the vehicle ratio downstream of the lane where a typical rear-end collision occurs, AvgS_D is the average speed (km / h) downstream of the lane where a typical rear-end collision occurs, and AvgA_D is the average acceleration (m / s²) downstream of the lane where a typical rear-end collision occurs. 2 R is the interleaving ratio in the interleaving zone, and VR is the interleaving flow rate in the interleaving zone.

[0076] DiffV_U is the difference in traffic volume (veh / h) between the lane where the rear-end collision occurred and the upstream lane of the adjacent lane. DiffTR_U is the difference in vehicle type ratio between the lane where the rear-end collision occurred and the upstream lane of the adjacent lane. DiffAvgS_U is the difference in average speed (km / h) between the lane where the rear-end collision occurred and the upstream lane of the adjacent lane. DiffV_D is the difference in traffic volume (veh / h) between the lane where the rear-end collision occurred and the downstream lane of the adjacent lane. DiffO_D is the difference in occupancy between the lane where the rear-end collision occurred and the downstream lane of the adjacent lane. DiffAvgA_D is the difference in average acceleration (m / s²) between the lane where the rear-end collision occurred and the downstream lane of the adjacent lane. 2 ).

[0077] Step S6: Construct a conflict prediction model.

[0078] Predictive models for severe side-impact collisions, general side-impact collisions, severe rear-end collisions, and general rear-end collisions were constructed using different machine learning methods, including random forests and neural networks.

[0079] 1. Construction of a conflict prediction model based on random forest

[0080] In this embodiment, the sample set is randomly allocated to the training set and the test set in an 8:2 ratio, and the random forest regression prediction model is implemented using the sklearn library and the RandomForestRegressor library based on Python 3.9.

[0081] Use grid search to find the optimal model parameters, including the maximum depth of decision trees in a random forest, the minimum number of samples in a leaf node, the minimum number of samples required to split an internal node, and the number of decision trees.

[0082] The optimal parameter values ​​for the four conflict prediction models are shown in Table 2.

[0083] Table 2 Optimal Parameter Values ​​for Conflict Prediction Model

[0084]

[0085] 2. Construction of a Conflict Prediction Model Based on Neural Networks

[0086] In this embodiment, the sample set is randomly allocated to the training set and the test set in an 8:2 ratio, and the neural network structure includes:

[0087] ① Input layer: It contains 256 neurons, selects ReLU as the activation function, and specifies that the input dimension is equal to the number of features in the training set;

[0088] ②Hidden layers: The model contains 4 hidden layers, with 128, 64, 32 and 16 neurons in each hidden layer, respectively.

[0089] ③ Output layer: There is only one neuron in the output layer.

[0090] L2 regularization is used to assist training when building neural networks.

[0091] In addition, the epoch, i.e. the number of iterations, needs to be adjusted to improve the model's prediction accuracy. The epoch values ​​for the four types of conflict prediction models are shown in Table 3.

[0092] Table 3 epoch values

[0093]

[0094] After the random forest and neural network conflict structures are built, the evaluation metrics are output as follows: root mean square error (RMSE) and mean absolute error (MAE). The statistical comparison results of the evaluation metrics are shown in Table 4.

[0095] Table 4. Statistical Comparison of Evaluation Indicators

[0096]

[0097] Finally, based on the comprehensive evaluation results, a neural network was selected to predict severe and general side impact collisions, while a random forest method was used to predict severe and general rear-end collisions.

[0098] In this embodiment, the prediction results are applied to actual traffic. By collecting and inputting the weaving zone characteristic variables, traffic flow characteristic variables of each lane, and the difference variables between each lane and adjacent lanes within the current 2-minute time interval, the number of various types of conflicts occurring in each lane within the next 2-minute time interval is predicted. This provides more accurate lane-level conflict risk information for vehicles in each lane, helping drivers choose safer lanes to drive in.

[0099] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting lane-level conflicts in expressway weaving zones based on real-time traffic flow data, characterized in that, Includes the following steps: S1. Conduct on-site investigations of the weaving areas of the expressways to be tested and obtain real-time video of traffic conditions. S2. Number each lane in the expressway weaving area, collect traffic flow data and traffic conflict data for each lane, and set collision time (TTC) thresholds based on different conflict types. S3. Build a traffic simulation model and perform data calibration. Obtain traffic flow data and vehicle trajectory data. Based on the conflict angle and the conflict threshold set in step S2, define different conflict types and use conflict analysis software to obtain different types of traffic conflict data. S4. Based on the traffic flow data obtained in step S3, extract feature variables and construct datasets for prediction models of different conflict types. Extract feature variables, including: weaving zone feature variables, conflict lane traffic flow feature variables, and difference variables between conflict lanes and adjacent lanes; weaving zone feature variables, including: weaving zone length, speed limit, weaving ratio, and weaving flow ratio; conflict lane traffic flow feature variables, including traffic flow data upstream and downstream of the conflict lane; and difference variables between conflict lanes and adjacent lanes, including the differences in various traffic flow parameters upstream and downstream between the conflict lane and adjacent lanes. S5. Based on the negative binomial regression method, construct traffic conflict analysis models for different conflict types and analyze the impact of different influencing factors on the number of traffic conflicts. Calculate the variance inflation factor (VIF) for each characteristic variable, perform variable selection, and build traffic conflict analysis models for different conflict types. The formula for calculating the variance inflation factor (VIF) is as follows: ； in, It is the independent variable Other independent variables The goodness of fit obtained after performing linear regression; Variables exhibiting multicollinearity are gradually eliminated until all remaining feature variables exhibit mild collinearity, which does not interfere with the interpretability of the analysis model. The feature variables that have undergone multicollinearity screening are then used in the traffic conflict analysis model. The negative binomial regression method is used to construct traffic conflict analysis models for different conflict types, as shown in the following formula: ； in, The number of conflicts for different conflict types, , ,… For regression coefficients, , , ..., These are the feature variables extracted in step S4; S6. Based on machine learning methods, construct traffic conflict prediction models for different conflict types, predict the number of various conflicts that will occur in each lane in the future, and provide lane-level conflict risk alerts for drivers in each lane.

2. The method for predicting lane-level conflicts in expressway weaving zones according to claim 1, characterized in that, In step S1, a drone is used to film and record the weaving area of ​​the expressway under test to obtain video of traffic conditions.

3. The method for predicting lane-level conflicts in expressway weaving zones according to claim 2, characterized in that, In step S2, video tracking software is used to collect traffic conflict data for each lane, and collision time (TTC) thresholds for different conflict types are defined based on the Pareto curve. The conflict types include: severe side impact conflict, general side impact conflict, severe rear-end collision conflict, and general rear-end collision conflict.

4. The method for predicting lane-level conflicts in expressway weaving zones according to claim 3, characterized in that, In step S3, a simulation model is built using VISSIM traffic simulation software, and the collected traffic flow data is used for calibration. The parameters used for calibration include: geometric design parameters of the weaving area, vehicle composition, traffic volume, vehicle speed distribution, vehicle car-following model parameters, and vehicle lane-changing model parameters.

5. The method for predicting lane-level conflicts in expressway weaving zones according to claim 1, characterized in that, In step S6, the sample set is proportionally divided into a training set and a test set, and machine learning methods are used to construct prediction models for severe side-impact collisions, general side-impact collisions, severe rear-end collisions, and general rear-end collisions, respectively. The machine learning methods include random forests and neural networks.

6. The method for predicting lane-level conflicts in expressway weaving zones according to claim 5, characterized in that, The machine learning approach based on random forests uses the sklearn library and the RandomForestRegressor library based on Python to build a conflict prediction model. Grid search is used to find the optimal model parameters, including: the maximum depth of the decision trees in the random forest, the minimum number of samples in the leaf nodes, the minimum number of samples required to split internal nodes, and the number of decision trees.

7. The method for predicting lane-level conflicts in expressway weaving zones according to claim 6, characterized in that, A conflict prediction model is constructed using a machine learning method based on neural networks. The neural network structure includes: Input layer: Contains 256 neurons, selects ReLU as the activation function, and specifies that the input dimension is equal to the number of features in the training set; Hidden layers: The model contains 4 hidden layers, with 128, 64, 32 and 16 neurons in each hidden layer, respectively; Output layer: The output layer contains only one neuron; The neural network uses L2 regularization to assist training and adjusts the number of iterations to improve the model's prediction accuracy.

8. The method for predicting lane-level conflicts in expressway weaving zones according to claim 1, characterized in that, In step S6, different machine learning methods are used to construct prediction models for different types of conflicts and evaluate the prediction results. The root mean square error and mean absolute error are calculated, and the prediction model with the smallest root mean square error and mean absolute error is selected to predict the number of various types of conflicts that will occur in each lane in the future, providing lane-level conflict risk information for driving vehicles in each lane. Root mean square error The calculation formula is as follows: ； Mean Absolute Error The calculation formula is as follows: ； in, This refers to the actual quantity. It is the true value of the i-th actual value. It is the predicted value of the i-th actual value.

Images