A Traffic Signal Model Predictive Control Method Based on Uncertain Parameter Prediction

Abstract

This invention discloses a traffic signal model predictive control method based on uncertain parameter prediction. First, an improved Louvain algorithm is used to divide traffic sub-regions. Then, an uncertain parameter prediction model is constructed. Finally, for each sub-region, a corresponding local optimization model is established, and a sequential quadratic programming algorithm is used to solve the local optimization model to obtain the optimal solution for the green light time of each phase at each intersection. This invention is based on a distributed control structure and uses the improved Louvain algorithm to divide traffic sub-regions, thereby decomposing the traffic signal control problem of the entire regional road network into several sub-problems. This reduces computational complexity, ensures the real-time performance of model predictive control, meets the online traffic signal optimization needs of multiple control intersections, and improves the accuracy of traffic signal control by establishing an uncertain parameter prediction model and combining historical and online uncertain prediction parameters.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic control, specifically a traffic signal model predictive control method based on the prediction of uncertain parameters. Background Technology

[0002] In recent years, China's car ownership has been increasing, and traffic congestion has become increasingly prominent, greatly impacting traffic efficiency and people's daily travel. To alleviate traffic pressure, traffic signal control has become a hot research topic.

[0003] Compared to traditional traffic signal control and reinforcement learning, model predictive control (MMC) methods predict future traffic conditions, perform rolling optimization, and provide feedback corrections. For dynamically changing traffic networks, MMC methods can produce a stable, reliable, and practically deployable time-series traffic signaling scheme. However, current MMC methods often handle uncertain parameters in a simplistic way. Simple modeling based solely on data from the previous time step cannot accurately estimate these uncertain traffic disturbance parameters, such as input and output flow errors and the number of vehicles entering and exiting road segments. This results in extremely coarse uncertainties that are difficult to use to obtain high-quality traffic signal control schemes. Especially during morning rush hour, large fluctuations in traffic disturbances lead to inaccurate and significant uncertainties, which greatly affects the accuracy of traffic signal control.

[0004] Meanwhile, compared with centralized control structures, distributed control structures can significantly reduce computational complexity, ensure the real-time performance of model predictive control, and meet the online traffic signal optimization needs of multiple control intersections. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a traffic signal model predictive control method based on the prediction of uncertain parameters, which aims to accurately estimate the uncertain parameters of the system, thereby improving the accuracy of traffic signal control.

[0006] The technical solution of this invention is as follows:

[0007] A traffic signal model predictive control method based on uncertain parameter prediction specifically includes the following steps:

[0008] (1) The improved Louvain algorithm is used to divide traffic sub-regions: Specifically, nodes are initialized, each intersection in the regional road network is regarded as a node, the initial number of sub-regions is the same as the number of nodes, and then the modularity increment between two adjacent nodes is calculated. ,when The maximum value is greater than When this happens, two adjacent nodes are assigned to the same sub-region, where... The calculation formula is shown in the following formula (1):

[0009] (1);

[0010] In equation (1), For nodes The maximum modularity with neighboring nodes, For nodes The minimum modularity with neighboring nodes. For nodes Average modularity, This represents the total number of nodes in the regional road network.

[0011] Then repeat the above process until the module degree increment of all sub-regions can no longer increase, so that all nodes in the same sub-region are compressed into a new node, and the edge weights between sub-regions are updated; repeat the above process until the module degree of all sub-regions in the entire regional road network is... The division of the traffic sub-region ends when no further changes occur.

[0012] (2) Construct an uncertain parameter prediction model, as shown in equation (2):

[0013] (2);

[0014] In equation (2), For uncertain parameters, and These are empirical parameters. ; Indicates parameters for historical uncertainty prediction; Indicates online uncertain prediction parameters;

[0015] Historical Uncertainty Prediction Parameters The calculation formula is shown in the following formula (3):

[0016] (3);

[0017] In equation (3), The current system time. For historical days, For the first The historical uncertain parameters of the day;

[0018] Online Uncertainty Prediction Parameters It consists of trend prediction component, time-invariant prediction component and time-varying prediction component, and the calculation formula is shown in the following formula (4):

[0019] (4);

[0020] In equation (4), Indicates the trend prediction component. Represents the time-invariant prediction component. Indicates the time-varying prediction component;

[0021] (3) First, for each divided sub-region, establish the corresponding local optimization model, as shown in equations (5)-(6) below:

[0022] (5);

[0023] st

[0024] (6);

[0025] In equation (5), Indicates based on intersection No. Traffic volume predictions for each road segment at each stage Traffic volume on various road sections at different stages; express transpose; Indicates based on intersection No. The predicted green light time for each phase of the stage Green light duration for each phase of the process; express transpose; Indicates the number of intersections within the sub-region; Indicates the step size for traffic volume forecasting; This indicates the step size for predicting the green light time. This represents the output error weight matrix related to traffic volume; This represents the control weight matrix related to the green light time;

[0026] In equation (6), the state matrix The identity matrix; input matrix It reflects the characteristics of topology, phase, period, saturation flow rate, and turning rate; The intersection predicted by the uncertain parameter prediction model based on time series decomposition No. Uncertain parameters of the stage; Indicates the green light time and lost time for all phases. The sum is less than the signal period Lost time It consists of startup loss time, yellow light time, and green light interval; This indicates that the green light time for each phase at a single intersection falls between the shortest and longest green light times; 0 Used to limit the maximum traffic volume that a road segment can accommodate;

[0027] Then, the sequential quadratic programming algorithm is used to solve the local optimization model to obtain the optimal solution for the green light time of each phase at each intersection.

[0028] The aforementioned modularity The calculation formula is shown in equation (7):

[0029] (7);

[0030] in, ;

[0031] In equation (7), For all connected nodes The sum of the weights of the edges between two nodes constitutes an edge. ; For nodes The subregion to which it belongs; For nodes Subregion to which it belongs; Used for judgment and Whether they belong to the same sub-region. If they belong to the same sub-region. The value is 1 if it is set to 1, otherwise the value is 0. Represents all connected nodes The weight matrix of the edges is calculated using the following formula (8):

[0032] (8);

[0033] In equation (8), Indicates from the intersection To the intersection The correlation of a one-way road segment is calculated using the formula (9). Indicates an intersection and intersection The correlation between the two-way road sections is calculated using the formula (10). This is the sum of the weights of all edges in the regional road network;

[0034] (9);

[0035] (10);

[0036] In equations (9) and (10), and Represents the balance factor. ; Indicates from the intersection To the intersection Traffic volume on one-way road sections; Indicates from the intersection To the intersection Traffic volume on a one-way road segment; Indicates from the intersection To the intersection The degree of connectivity of one-way road segments; Indicates from the intersection To the intersection The number of lanes on a one-way road section; Indicates an intersection and intersection The distance between them; Indicates an intersection and intersection The correlation between sub-regions can be set to 0, 0.1, or 0.2. A value of 0 indicates that the sub-regions are unrelated, a value of 0.1 indicates that the sub-regions are similar, and a value of 0.2 indicates that the sub-regions are consistent.

[0037] The aforementioned module degree increment The calculation formula is shown in equation (11):

[0038] (11);

[0039] In equation (11), Indicates connecting nodes within the same sub-region The sum of the weights of the edges with other nodes. ; Indicates the connection node With nodes The weights of edges from other nodes outside the sub-region. .

[0040] The calculation process for the trend prediction component, time-invariant prediction component, and time-varying prediction component is as follows:

[0041] a. Solving for the median of the upper and lower envelopes of a time series: Find the median of the time series... All maximum and minimum points, where T represents a historical time period before the current system time t, and the value range is T. The upper and lower envelopes are formed by fitting a cubic spline interpolation function, and the median of the upper and lower envelopes is calculated. ;

[0042] b. Solving for new signals: Calculating time series With median The difference between them yields a new signal. ,Right now ;

[0043] c. Intrinsic modulus function condition judgment: Judging new signals Whether the intrinsic modulus function (IMF) satisfies the two conditions is as follows: the number of local extrema and zero crossings must be equal or differ by at most one over the entire time range; and at any given time point, the average of the envelopes of the local maxima and local minima must be zero.

[0044] d. When a new signal When the two conditions for the intrinsic modulus function (IMF) are not met, use replace As a new time series, step ac is repeated for analysis; when the new signal... When the two conditions of the intrinsic mode function (IMF) are satisfied, This is denoted as the first IMF component. The acquisition of IMF components involved several iterations.

[0045] e. Obtain the first IMF component Afterwards, use minus Obtain the remaining components Then, through the analysis in step ad, the second IMF component is obtained. Then use minus Obtain the remaining components ,Right now ;go through Decompose the components repeatedly until the remaining components are obtained. The remaining components are those that cannot be further decomposed. As a trend component ,Right now ;

[0046] f. Trend prediction component Using the SVR algorithm in conjunction with the Sigmoid kernel function To perform fitting prediction, the formula for the SVR algorithm is shown in equation (12):

[0047] (12);

[0048] in, ;

[0049] In equation (12), for The number of parameters at time t; Represents the hyperplane weight vector; This represents the penalty coefficient, which is not less than 0; Indicates the insensitive loss function; ; Indicates the loss threshold; This represents the trend prediction component to be fitted;

[0050] Sigmoid kernel function For nonlinearly inseparable datasets, the low-dimensional to high-dimensional mapping is used to achieve linear separability of the dataset. The calculation formula is shown in the following formula (13):

[0051] (13);

[0052] In equation (13), express Activation function and All represent trend component data. express transpose, and All represent hyperparameters;

[0053] g. Composed of non-stationary components Obtained through a Fourier filter and The calculation formulas are shown in equations (14) and (15) below:

[0054] (14);

[0055] (15);

[0056] In equations (14) and (15), Indicates a Fourier filter; Represents a time series; Indicates trend components;

[0057] h, Input Through the encoder, learnable matrix, and decoder, we obtain The calculation formula is shown in the following formula (16):

[0058] (16);

[0059] In equation (16), Indicates encoder; Represents a learnable matrix; The decoder is a commonly used MLP structure; time-invariant predictive components. for The number of parameters at time t;

[0060] i. Time-varying prediction component Calculation: First, based on the observation sequence T segments prior to the current system time t, the time-varying components corresponding to the previous T segments are divided into several equal-length segments, each segment having a length of... Then it is divided into Section, of which, the first Time-varying components The representation of is shown in equation (17):

[0061] (17);

[0062] For each small segment of the time-varying component, the encoder maps it to a nonlinear dynamic space, resulting in... The calculation formula is shown in the following formula (18):

[0063] (18);

[0064] In equation (18), This represents the encoder, which is a commonly used MLP structure.

[0065] Time-varying operators The optimal fitting matrix is ​​obtained through the eDMD algorithm, and the calculation formulas are shown in equations (19)-(21) below:

[0066] (19);

[0067] (20);

[0068] (twenty one);

[0069] In equations (19)-(21), and All of these are characteristics of past time windows. for The inverse matrix;

[0070] Through time-varying operators The past window time period features are fitted, as shown in the following formula (22): (twenty two);

[0071] To obtain the future time length The prediction is used to fit the characteristics of the future window time period, as shown in the following formula (23):

[0072] (twenty three);

[0073] In equation (23), This indicates the characteristics of a future window of time in the h-th time period. Indicates the past number Characteristics of past time windows in a time period;

[0074] Finally, it is obtained through the encoder. Time-varying prediction components for The number of parameters at time t. The calculation formula is shown in the following formula (24):

[0075] (twenty four);

[0076] In equation (24), This represents the decoder, which is a commonly used MLP structure.

[0077] In the calculation of the time-varying and time-invariant prediction components, the MSE loss function is used as the objective function, the non-stationary component is used as the input value, and the encoder and learnable matrix are used. Both the decoder and the time-varying predictor are learnable parameters. Through parameter optimization, more accurate time-varying and time-invariant predictor components can be obtained.

[0078] For each divided sub-region, before establishing the local optimization model, an intersection and traffic flow model is constructed. The specific modeling process for the intersection and traffic flow model is as follows:

[0079] Adjacent intersections Intersection The middle section is , For discrete time steps, At the starting time, For the first Stages and Intersections Intersecting entrance lanes The flow rate of the inflow segment z is calculated using the following formula (25):

[0080] (25);

[0081] In equation (25), This represents the road segment opposite to road segment z. Indicates an intersection At each entrance The turning rate relative to road segment z For the first Stage intersection At each entrance The flow rate relative to road segment z;

[0082] For the first Stage Z section enters the intersection The flow rate is calculated as shown in the following formula (26):

[0083] (26);

[0084] In equation (26), This represents the saturation flow rate of segment z. Indicates the signal period. Indicates the first Stage intersection Green light times for each phase;

[0085] In considering the first Uncertain parameters of segment z in stage In the case of the first Traffic volume at time and on road segment z The calculation is shown in the following formula (27):

[0086] (27);

[0087] In equation (27), Indicates the first Traffic volume at time and on road segment z; and The flow rates entering and leaving road segment z are respectively represented by the flow rates measured by roadside detection equipment; the aforementioned It is obtained by transforming equations (26) and (27).

[0088] Advantages of this invention:

[0089] (1) Based on the distributed control structure, the present invention first uses the improved Louvain algorithm to divide the traffic sub-regions, thereby decomposing the traffic signal control problem of the entire regional road network into several sub-problems, thereby reducing the computational complexity, ensuring the real-time performance of model predictive control, and meeting the online traffic signal optimization needs of multiple control intersections.

[0090] (2) The present invention establishes an uncertain parameter prediction model, which combines historical uncertain prediction parameters and online uncertain prediction parameters to improve the accuracy of traffic signal control and enhance the stability and robustness of traffic control.

[0091] (3) The present invention establishes a local optimization model for distributed control structure, considering the interaction between various intersections in the regional road network, and uses model predictive control to optimize traffic signals, thereby achieving precise control of traffic signals. Attached Figure Description

[0092] Figure 1 This is a diagram illustrating the road segment principle of constructing an intersection and traffic flow model according to the present invention. Detailed Implementation

[0093] 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.

[0094] A traffic signal model predictive control method based on uncertain parameter prediction specifically includes the following steps:

[0095] (1) The improved Louvain algorithm is used to divide the traffic sub-regions:

[0096] S11. The basis for sub-region division mainly depends on the degree of correlation between adjacent intersections. The stronger the correlation, the easier it is to be divided into the same sub-region. Using graph theory, each intersection in the regional road network is abstracted as a node, each road segment is abstracted as an edge, and the entire regional road network is abstracted as a network topology graph, i.e., intersections. Corresponding nodes in the diagram intersection Intersection The road segments connecting the intersections are considered as edges. The correlation between the intersections is calculated using equations (9) and (10) below:

[0097] (9);

[0098] (10);

[0099] In equations (9) and (10), Indicates from the intersection To the intersection The degree of connectivity of one-way road segments; Indicates an intersection and intersection The degree of correlation between the two-way road sections; and Represents the balance factor. ; Indicates from the intersection To the intersection Traffic volume on one-way road sections; Indicates from the intersection To the intersection Traffic volume on a one-way road segment; Indicates from the intersection To the intersection The degree of connectivity of one-way road segments; Indicates from the intersection To the intersection The number of lanes on a one-way road section; Indicates an intersection and intersection The distance between them; Indicates an intersection and intersection The correlation between sub-regions can be set to 0, 0.1, or 0.2. A value of 0 indicates that the sub-regions are unrelated, a value of 0.1 indicates that the sub-regions are similar, and a value of 0.2 indicates that the sub-regions are identical.

[0100] In equation (8), the calculation formula is shown in equation (9); and the calculation formula is shown in equation (10). This is the sum of the weights of all edges in the regional road network;

[0101] S12. Use the correlation degree as the weight of the edge to create a weight matrix. ,in Represents all connected nodes The weight matrix of the edges is calculated using the following formula (8):

[0102] (8);

[0103] In equation (8), This is the sum of the weights of all edges in the regional road network;

[0104] S13, Modularity It is a measure for evaluating the quality of regional division. The larger the value, the better the region partitioning result and the higher the modularity. The calculation formula is shown in equation (7):

[0105] (7);

[0106] in, ;

[0107] In equation (7), For all connected nodes The sum of the weights of the edges between two nodes constitutes an edge. ; For nodes The subregion to which it belongs; For nodes The subregion to which it belongs; Used for judgment and Whether they belong to the same sub-region. If they belong to the same sub-region. The value is 1 if it is set to 1, otherwise the value is 0.

[0108] S14, Node Modularity increment of the sub-region The calculation formula is shown in equation (11):

[0109] (11);

[0110] In equation (11), Indicates connecting nodes within the same sub-region The sum of the weights of the edges with other nodes. ; Indicates the connection node With nodes The weights of edges from other nodes outside the sub-region. ;

[0111] S15. Perform sub-region division: Specifically, initialize nodes, treating each intersection in the regional road network as a node. The initial number of sub-regions is the same as the number of nodes. Then, calculate the modularity increment between two adjacent nodes. ,when The maximum value is greater than When this happens, two adjacent nodes are assigned to the same sub-region, where... The calculation formula is shown in the following formula (1):

[0112] (1);

[0113] In equation (1), For nodes The maximum modularity with neighboring nodes, For nodes The minimum modularity with neighboring nodes. For nodes Average modularity, This represents the total number of nodes in the regional road network.

[0114] Then repeat the above process until the module degree increment of all sub-regions can no longer increase, so that all nodes in the same sub-region are compressed into a new node, and the edge weights between sub-regions are updated; repeat the above process until the module degree of all sub-regions in the entire regional road network is... The division of the traffic sub-region ends when no further changes occur.

[0115] (2) Construct an uncertain parameter prediction model, as shown in equation (2):

[0116] (2);

[0117] In equation (2), For uncertain parameters, and These are empirical parameters. ; Indicates parameters for historical uncertainty prediction; Indicates online uncertain prediction parameters;

[0118] Historical Uncertainty Prediction Parameters The calculation formula is shown in the following formula (3):

[0119] (3);

[0120] In equation (3), The current system time. For historical days, For the first The historical uncertain parameters of the day;

[0121] Online Uncertainty Prediction Parameters It consists of trend prediction component, time-invariant prediction component and time-varying prediction component, and the calculation formula is shown in the following formula (4):

[0122] (4);

[0123] In equation (4), Indicates the trend prediction component. Represents the time-invariant prediction component. Indicates the time-varying prediction component;

[0124] The calculation process for the trend prediction component, time-invariant prediction component, and time-varying prediction component is as follows:

[0125] a. Solving for the median of the upper and lower envelopes of a time series: Find the median of the time series... All maximum and minimum points, where T represents a historical time period before the current system time t, and the value range is T. The upper and lower envelopes are formed by fitting a cubic spline interpolation function, and the median of the upper and lower envelopes is calculated. ;

[0126] b. Solving for new signals: Calculating time series With median The difference between them yields a new signal. ,Right now ;

[0127] c. Intrinsic modulus function condition judgment: Judging new signals Whether the intrinsic modulus function (IMF) satisfies the two conditions is as follows: the number of local extrema and zero crossings must be equal or differ by at most one over the entire time range; and at any given time point, the average of the envelopes of the local maxima and local minima must be zero.

[0128] d. When a new signal When the two conditions for the intrinsic modulus function (IMF) are not met, use replace As a new time series, step ac is repeated for analysis; when the new signal... When the two conditions of the intrinsic mode function (IMF) are satisfied, This is denoted as the first IMF component. The acquisition of IMF components involved several iterations.

[0129] e. Obtain the first IMF component Afterwards, use minus Obtain the remaining components Then, through the analysis in step ad, the second IMF component is obtained. Then use minus Obtain the remaining components ,Right now ;go through Decompose the components repeatedly until the remaining components are obtained. The remaining components are those that cannot be further decomposed. As a trend component ,Right now ;

[0130] f. Trend prediction component Using the SVR algorithm in conjunction with the Sigmoid kernel function To perform fitting prediction, the formula for the SVR algorithm is shown in equation (12):

[0131] (12);

[0132] in, ;

[0133] In equation (12), for The number of parameters at time t; Represents the hyperplane weight vector; This represents the penalty coefficient, which is not less than 0; Indicates the insensitive loss function; ; Indicates the loss threshold; This represents the trend prediction component to be fitted;

[0134] Sigmoid kernel function For nonlinearly inseparable datasets, the low-dimensional to high-dimensional mapping is used to achieve linear separability of the dataset. The calculation formula is shown in the following formula (13):

[0135] (13);

[0136] In equation (13), express Activation function and All represent trend component data. express transpose, and All represent hyperparameters;

[0137] g. Composed of non-stationary components Obtained through a Fourier filter and The calculation formulas are shown in equations (14) and (15) below:

[0138] (14);

[0139] (15);

[0140] In equations (14) and (15), Indicates a Fourier filter; Represents a time series; Indicates trend components;

[0141] h, Input Through the encoder, learnable matrix, and decoder, we obtain The calculation formula is shown in the following formula (16):

[0142] (16);

[0143] In equation (16), Indicates encoder; Represents a learnable matrix; The decoder is a commonly used MLP structure; time-invariant predictive components. for The number of parameters at time t;

[0144] i. Time-varying prediction component Calculation: First, based on the observation sequence T segments prior to the current system time t, the time-varying components corresponding to the previous T segments are divided into several equal-length segments, each segment having a length of... Then it is divided into Section, of which, the first Time-varying components The representation of is shown in equation (17):

[0145] (17);

[0146] For each small segment of the time-varying component, the encoder maps it to a nonlinear dynamic space, resulting in... The calculation formula is shown in the following formula (18):

[0147] (18);

[0148] In equation (18), This represents the encoder, which is a commonly used MLP structure.

[0149] Time-varying operators The optimal fitting matrix is ​​obtained through the eDMD algorithm, and the calculation formulas are shown in equations (19)-(21) below:

[0150] (19);

[0151] (20);

[0152] (twenty one);

[0153] In equations (19)-(21), and All of these are characteristics of past time windows. for The inverse matrix;

[0154] Through time-varying operators The past window time period features are fitted, as shown in the following formula (22): (twenty two);

[0155] To obtain the future time length The prediction is used to fit the characteristics of the future window time period, as shown in the following formula (23):

[0156] (twenty three);

[0157] In equation (23), This indicates the characteristics of a future window of time in the h-th time period. Past Characteristics of past time windows in a time period;

[0158] Finally, it is obtained through the encoder. Time-varying prediction components for The number of parameters at time t. The calculation formula is shown in the following formula (24):

[0159] (twenty four);

[0160] In equation (24), The decoder is represented by a commonly used MLP structure.

[0161] In the calculation of time-varying and time-invariant predictive components, the MSE loss function is used as the objective function, and the non-stationary component is used as the input value. The encoder and the learnable matrix are also used. Both the decoder and the decoder are learnable parameters. Through parameter optimization, more accurate time-varying prediction components and time-invariant prediction components can be obtained.

[0162] (3) Distributed traffic signal control method based on model predictive control:

[0163] S31. First, for each divided sub-region, construct the intersection and traffic flow model. The specific modeling process for the intersection and traffic flow model is as follows:

[0164] See Figure 1 Adjacent intersections Intersection The middle section is , For discrete time steps, At the starting time, For the first Stages and Intersections Intersecting entrance lanes The flow rate of the inflow segment z is calculated using the following formula (25):

[0165] (25);

[0166] In equation (25), This represents the road segment opposite to road segment z. Indicates an intersection At each entrance The turning rate relative to road segment z For the first Stage intersection At each entrance The flow rate relative to road segment z;

[0167] For the first Stage Z section enters the intersection The flow rate is calculated as shown in the following formula (26):

[0168] (26);

[0169] In equation (26), This represents the saturation flow rate of segment z. Indicates the signal period. Indicates the first Stage intersection Green light times for each phase;

[0170] In considering the first Uncertain parameters of segment z in stage In the case of the first Traffic volume at time and on road segment z The calculation is shown in the following formula (27):

[0171] (27);

[0172] In equation (27), Indicates the first Traffic volume at time and on road segment z; and These represent the flow rates entering and leaving road segment z, respectively, measured by roadside detection equipment;

[0173] S32. For each sub-region after division, establish a corresponding local optimization model, as shown in equations (5)-(6) below:

[0174] (5);

[0175] st

[0176] (6);

[0177] In equation (5), Indicates based on intersection No. Traffic volume predictions for each road segment at each stage Traffic volume on various road sections at different stages; express transpose; Indicates based on intersection No. The predicted green light time for each phase of the stage Green light duration for each phase of the process; express transpose; Indicates the number of intersections within the sub-region; Indicates the step size for traffic volume forecasting; This indicates the step size for predicting the green light time. This represents the output error weight matrix related to traffic volume; This represents the control weight matrix related to the green light time;

[0178] In equation (6), The state matrix is ​​obtained by transforming equations (26) and (27). The identity matrix; input matrix It reflects the characteristics of topology, phase, period, saturation flow rate, and turning rate; The intersection predicted by the uncertain parameter prediction model based on time series decomposition No. Uncertain parameters of the stage; Indicates the green light time and lost time for all phases. The sum is less than the signal period Lost time It consists of startup loss time, yellow light time, and green light interval; This indicates that the green light time for each phase at a single intersection falls between the shortest and longest green light times; 0 Used to limit the maximum traffic volume that a road segment can accommodate;

[0179] Then, the sequential quadratic programming algorithm is used to solve the local optimization model to obtain the optimal solution for the green light time of each phase at each intersection.

[0180] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A traffic signal model predictive control method based on uncertain parameter prediction, characterized in that: Specifically, it includes the following steps: (1) The improved Louvain algorithm is used to divide traffic sub-regions: Specifically, nodes are initialized, each intersection in the regional road network is regarded as a node, the initial number of sub-regions is the same as the number of nodes, and then the modularity increment between two adjacent nodes is calculated. ,when The maximum value is greater than When this happens, two adjacent nodes are assigned to the same sub-region, where... The calculation formula is shown in the following formula (1): （1）； In equation (1), For nodes The maximum modularity with neighboring nodes, For nodes The minimum modularity with neighboring nodes. For nodes Average modularity, This represents the total number of nodes in the regional road network. Then repeat the above process until the module degree increment of all sub-regions can no longer increase, so that all nodes in the same sub-region are compressed into a new node, and the edge weights between sub-regions are updated; repeat the above process until the module degree of all sub-regions in the entire regional road network is... The division of the traffic sub-region ends when no further changes occur. (2) Construct an uncertain parameter prediction model, as shown in equation (2): （2）； In equation (2), For uncertain parameters, and These are empirical parameters. ; Indicates parameters for historical uncertainty prediction; Indicates online uncertain prediction parameters; Historical Uncertainty Prediction Parameters The calculation formula is shown in the following formula (3): （3）； In equation (3), The current system time. For historical days, For the first The historical uncertain parameters of the day; Online Uncertainty Prediction Parameters It consists of trend prediction component, time-invariant prediction component and time-varying prediction component, and the calculation formula is shown in the following formula (4): （4）； In equation (4), Indicates the trend prediction component. Represents the time-invariant prediction component. Indicates the time-varying prediction component; (3) First, for each divided sub-region, establish the corresponding local optimization model, as shown in equations (5)-(6) below: （5）； st （6）； In equation (5), Indicates based on intersection No. Traffic volume predictions for each road segment at each stage Traffic volume on various road sections at different stages; express Transpose of; Indicates based on intersection No. The predicted green light time for each phase of the stage Green light duration for each phase of the process; express Transpose of; Indicates the number of intersections within the sub-region; Indicates the step size for traffic volume forecasting; Indicates the green light time prediction step size; This represents the output error weight matrix related to traffic volume; This represents the control weight matrix related to the green light time; In equation (6), the state matrix The identity matrix; input matrix It reflects the characteristics of topology, phase, period, saturation flow rate, and turning rate; The intersection predicted by the uncertain parameter prediction model based on time series decomposition No. Uncertain parameters of the stage; Indicates the green light time and lost time for all phases. The sum is less than the signal period Lost time It consists of startup loss time, yellow light time, and green light interval; This indicates that the green light time for each phase at a single intersection falls between the shortest and longest green light times; 0 Used to limit the maximum traffic volume that a road segment can accommodate; Then, the sequential quadratic programming algorithm is used to solve the local optimization model to obtain the optimal solution for the green light time of each phase at each intersection.

2. The traffic signal model predictive control method based on uncertain parameter prediction according to claim 1, characterized in that: The aforementioned modularity The calculation formula is shown in equation (7): （7）； in, ; In equation (7), For all connected nodes The sum of the weights of the edges between two nodes constitutes an edge. ; For nodes The subregion to which it belongs; For nodes Subregion to which it belongs; Used for judgment and Whether they belong to the same sub-region. If they belong to the same sub-region. The value is 1 if it is set to 1, otherwise the value is 0. Represents all connected nodes The weight matrix of the edges is calculated using the following formula (8): （8）； In equation (8), Indicates from the intersection To the intersection The correlation of a one-way road segment is calculated using the formula (9). Indicates an intersection and intersection The correlation between the two-way road sections is calculated using the formula (10). This is the sum of the weights of all edges in the regional road network; （9）； （10）； In equations (9) and (10), and Represents the balance factor. ; Indicates from the intersection To the intersection Traffic volume on one-way road sections; Indicates from the intersection To the intersection Traffic volume on a one-way road segment; Indicates from the intersection To the intersection The degree of connectivity of one-way road segments; Indicates from the intersection To the intersection The number of lanes on a one-way road section; Indicates an intersection and intersection The distance between them; Indicates an intersection and intersection The correlation between sub-regions can be set to 0, 0.1, or 0.

2. A value of 0 indicates that the sub-regions are unrelated, a value of 0.1 indicates that the sub-regions are similar, and a value of 0.2 indicates that the sub-regions are consistent.

3. The traffic signal model predictive control method based on uncertain parameter prediction according to claim 2, characterized in that: The aforementioned module degree increment The calculation formula is shown in equation (11): （11）； In equation (11), Indicates connecting nodes within the same sub-region The sum of the weights of the edges with other nodes. ; Indicates the connection node With nodes The weights of edges from other nodes outside the sub-region. .

4. The traffic signal model predictive control method based on uncertain parameter prediction according to claim 1, characterized in that: The calculation process for the trend prediction component, time-invariant prediction component, and time-varying prediction component is as follows: a. Solving for the median of the upper and lower envelopes of a time series: Find the median of the time series... All maximum and minimum points, where T represents a historical time period before the current system time t, and the value range is T. The upper and lower envelopes are formed by fitting a cubic spline interpolation function, and the median of the upper and lower envelopes is calculated. ; b. Solving for new signals: Calculating time series With median The difference between them yields a new signal. ,Right now ; c. Intrinsic modulus function condition judgment: Judging new signals Whether the intrinsic modulus function (IMF) satisfies the two conditions is as follows: the number of local extrema and zero crossings must be equal or differ by at most one over the entire time range; and at any given time point, the average of the envelopes of the local maxima and local minima must be zero. d. When a new signal When the two conditions for the intrinsic modulus function (IMF) are not met, use replace As a new time series, step ac is repeated for analysis; when the new signal... When the two conditions of the intrinsic mode function (IMF) are satisfied, This is denoted as the first IMF component. The acquisition of IMF components involves several iterations. e. Obtain the first IMF component Afterwards, use minus Obtain the remaining components Then, through the analysis in step ad, the second IMF component is obtained. Then use minus Obtain the remaining components ,Right now ;go through Decompose the components repeatedly until the remaining components are obtained. The remaining components are those that cannot be further decomposed. As a trend component ,Right now ; f. Trend prediction component Using the SVR algorithm in conjunction with the Sigmoid kernel function For fitting and prediction, the formula for the SVR algorithm is shown in equation (12): （12）； in, ; In equation (12), for The number of parameters at time t; Represents the hyperplane weight vector; This represents the penalty coefficient, which is not less than 0; Indicates the insensitive loss function; ; Indicates the loss threshold; This represents the trend prediction component to be fitted; Sigmoid kernel function For nonlinearly inseparable datasets, mapping from low dimension to high dimension achieves linear separability of the dataset. The calculation formula is shown in equation (13): （13）； In equation (13), express Activation function and All represent trend component data. express transpose, and All represent hyperparameters; g. Composed of non-stationary components Obtained through a Fourier filter and The calculation formulas are shown in equations (14) and (15) below: （14）； （15）； In equations (14) and (15), Indicates a Fourier filter; Represents a time series; Indicates trend components; h, Input Through the encoder, learnable matrix, and decoder, we obtain The calculation formula is shown in the following formula (16): （16）； In equation (16), Indicates encoder; Represents a learnable matrix; The decoder is a commonly used MLP structure; time-invariant predictive components. for The number of parameters at time t; i. Time-varying prediction component Calculation: First, based on the observation sequence T segments prior to the current system time t, the time-varying components corresponding to the previous T segments are divided into several equal-length segments, each segment having a length of... Then it is divided into Section, of which, the first Time-varying components The representation of is shown in equation (17): （17）； For each small segment of the time-varying component, the encoder maps it to a nonlinear dynamic space, resulting in... The calculation formula is shown in the following formula (18): （18）； In equation (18), This represents the encoder, which is a commonly used MLP structure. Time-varying operators The optimal fitting matrix is ​​obtained through the eDMD algorithm, and the calculation formulas are shown in equations (19)-(21) below: （19）； （20）； （21）； In equations (19)-(21), and All of these are characteristics of past time windows. for The inverse matrix; Through time-varying operators The past window time period features are fitted, as shown in the following formula (22): (twenty two); To obtain the future time length The prediction is used to fit the characteristics of the future window time period, as shown in the following formula (23): （23）； In equation (23), This indicates the characteristics of a future window of time in the h-th time period. Indicates the past number Characteristics of past time windows in a time period; Finally, it is obtained through the encoder. Time-varying prediction components for The number of parameters at time t. The calculation formula is shown in the following formula (24): （24）； In equation (24), This represents the decoder, which is a commonly used MLP structure.

5. The traffic signal model predictive control method based on uncertain parameter prediction according to claim 4, characterized in that: In the calculation of the time-varying and time-invariant prediction components, the MSE loss function is used as the objective function, the non-stationary component is used as the input value, and the encoder and learnable matrix are used. Both the decoder and the time-varying predictor are learnable parameters. Through parameter optimization, more accurate time-varying and time-invariant predictor components can be obtained.

6. The traffic signal model predictive control method based on uncertain parameter prediction according to claim 1, characterized in that: For each divided sub-region, before establishing the local optimization model, an intersection and traffic flow model is constructed. The specific modeling process for the intersection and traffic flow model is as follows: Adjacent intersections Intersection The middle section is , For discrete time steps, At the starting time, For the first Stages and Intersections Intersecting entrance lanes The flow rate of the inflow segment z is calculated using the following formula (25): （25）； In equation (25), This represents the road segment opposite to road segment z. Indicates an intersection At each entrance The turning rate relative to road segment z For the first Stage intersection At each entrance The flow rate relative to road segment z; For the first Stage Z section enters the intersection The flow rate is calculated as shown in the following formula (26): （26）； In equation (26), This represents the saturation flow rate of segment z. Indicates the signal period. Indicates the first Stage intersection Green light times for each phase; In considering the first Uncertain parameters of segment z in stage In the case of the first Traffic volume at time and on road segment z The calculation is shown in the following formula (27): （27）； In equation (27), Indicates the first Traffic volume at time and on road segment z; and The flow rates entering and leaving road segment z are respectively represented by the flow rates measured by roadside detection equipment; the aforementioned It is obtained by transforming equations (26) and (27).

Images