Ramp control and ramp control model training method and device

Abstract

The invention provides a ramp control method and device and a ramp control model training method and device, and relates to the technical field of artificial intelligence, in particular to the technical field of intelligent cloud, intelligent traffic and the like. The ramp control method comprises the following steps: obtaining a target representation; the target characterization comprises historical traffic characterization of a target road connected with a target ramp before a target moment; determining a target control scheme of the target ramp at the target moment based on the target representation; and at the target moment, controlling the target ramp based on the target control scheme.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to the technical fields of smart cloud, smart transportation, etc., and in particular to a ramp control and ramp control model training method and device. Background Art

[0002] A ramp usually refers to a short connecting road section provided for vehicles to enter and exit a main road (such as a highway, an elevated road, a bridge, and a driving tunnel, etc.). The connecting road section connects a main road and a secondary road adjacent to the main road.

[0003] By opening or closing the ramp at the entrance of the main road, the number of vehicles traveling on the auxiliary road entering the main road through the ramp can be controlled, thereby improving the traffic efficiency of the main road. Summary of the invention

[0004] The present disclosure provides a ramp control and a ramp control model training method and device.

[0005] According to one aspect of the present disclosure, a ramp control method is provided, comprising: obtaining a target representation; the target representation comprising: a historical traffic representation of a target road connected to a target ramp before a target time; determining a target control scheme for the target ramp at the target time based on the target representation; and controlling the target ramp at the target time based on the target control scheme.

[0006] According to another aspect of the present disclosure, a ramp control model training method is provided, wherein the ramp control model includes: a representation extraction network and a solution generation network, and the method includes: using the representation extraction network to obtain a target representation; the target representation includes: a historical traffic representation of a target road connected to a target ramp before a target time; using the solution generation network to determine a prediction control solution for the target ramp at the target time based on the target representation; constructing a target loss function based on the prediction control solution and the label control solution; and adjusting parameters of the solution generation network based on the target loss function, and fixing parameters of the representation extraction network.

[0007] According to another aspect of the present disclosure, a ramp control device is provided, including: an acquisition module for acquiring a target representation; the target representation includes: a historical traffic representation of a target road connected to a target ramp before a target time; a determination module for determining a target control scheme for the target ramp at the target time according to the target representation; and a control module for controlling the target ramp at the target time based on the target control scheme.

[0008] According to another aspect of the present disclosure, a ramp control model training device is provided, wherein the ramp control model includes: a representation extraction network and a solution generation network, and the device includes: an acquisition module, used to adopt the representation extraction network to acquire a target representation; the target representation includes: a historical traffic representation of a target road connected to a target ramp before a target time; a determination module, used to adopt the solution generation network, and determine a prediction control solution for the target ramp at the target time based on the target representation; a construction module, used to construct a target loss function according to the prediction control solution and the label control solution; and an adjustment module, used to adjust the parameters of the solution generation network according to the target loss function, and fix the parameters of the representation extraction network.

[0009] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any method as described in any of the above aspects.

[0010] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any one of the methods according to any one of the above aspects.

[0011] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the methods described in any one of the above aspects.

[0012] The present disclosure can improve ramp control effect.

[0013] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0015] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0016] Figure 2 is a schematic diagram of an application scenario for implementing an embodiment of the present disclosure;

[0017] Figure 3 is a schematic diagram of determining a target control scheme based on a ramp control model according to an embodiment of the present disclosure;

[0018] Figure 4 is a schematic diagram of a representation extraction network provided according to an embodiment of the present disclosure;

[0019] Figure 5 is a schematic diagram of a network generated according to the solution provided by an embodiment of the present disclosure;

[0020] Figure 6 is a schematic diagram according to a second embodiment of the present disclosure;

[0021] Figure 7 is a schematic diagram according to a third embodiment of the present disclosure;

[0022] Figure 8 is a schematic diagram of a network involved in a training process provided by an embodiment of the present disclosure;

[0023] Fig. 9 is a schematic diagram of a solution evaluation network provided by an embodiment of the present disclosure;

[0024] Fig.10 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0025] Fig.11 is a schematic diagram according to a fifth embodiment of the present disclosure;

[0026] Fig.12 is a schematic diagram according to a sixth embodiment of the present disclosure;

[0027] Fig.13 It is a schematic diagram of an electronic device used to implement the ramp control method or ramp control model training method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0029] In the related art, a timed closing method or a threshold closing method is usually used to control the ramp. The timed closing method refers to closing the ramp after a preset time, and the threshold closing method refers to closing the ramp after the traffic flow on the main road reaches a preset threshold.

[0030] However, the method based on preset time or preset threshold is difficult to adapt to the complex actual road conditions and affects the traffic efficiency of the main road. Therefore, the ramp control effect needs to be improved.

[0031] In order to improve the ramp control effect, the present disclosure provides the following embodiments.

[0032] Figure 1 is a schematic diagram according to the first embodiment of the present disclosure. This embodiment provides a ramp control method, the method comprising:

[0033] 101. Obtain a target representation; the target representation includes: a historical traffic representation of a target road connected to a target ramp before a target time.

[0034] 102. Determine a target control scheme for the target ramp at the target time based on the target representation.

[0035] 103. At the target time, control the target ramp based on the target control scheme.

[0036] Among them, the target ramp refers to the ramp to be controlled.

[0037] The target road refers to the road connected to the target ramp.

[0038] Generally speaking, one side connected by the ramp is a main road, and the other side is a secondary road. Therefore, the target road includes: a target main road and a target secondary road.

[0039] The target time refers to the time to be controlled. For example, the target time is represented by time t.

[0040] Historical moment refers to the time point before the target moment.

[0041] The target control scheme at the target moment (t moment) is determined based on the historical traffic representations of a preset number of historical moments. The preset number can be represented by M, where M is a positive integer.

[0042] For example, M=5, then the target control scheme at time t is determined based on the historical traffic representations at times (t-1), (t-2) ... (t-5).

[0043] Traffic representation is used to represent traffic indicators on the target road, such as traffic flow. In a traffic scenario, traffic flow refers to the number of traffic participants on the road, such as the number of vehicles traveling on the road.

[0044] The historical popular representation refers to the popular representation of a historical moment. For example, the historical popular representation includes: the popular representation of moments (t-1), (t-2)...(t-5).

[0045] The target control plan refers to the control plan of the target ramp at the target time.

[0046] The target control scheme may specifically include: on, or off, for example, 1 represents on, and 0 represents off.

[0047] Specifically, assuming that the target representation obtained at time t is h t (usually a vector), then the target can be used to represent h t Get the target control solution S at time t t (on or off), and then at time t, the target control scheme S at time t is adopted t Control the target ramp.

[0048] For example, the target control scheme S t If it is on, the signal light corresponding to the target ramp can be controlled to be green, indicating that it is passable. At this time (time t), vehicles can enter the target main road from the target auxiliary road through the target ramp. t If it is off, the signal light is controlled to be red, indicating that passage is prohibited. At this time (time t), vehicles cannot enter the target main road from the target auxiliary road through the target ramp.

[0049] In this embodiment, the target control scheme is determined based on the target representation. Compared with the method based on preset time or preset threshold, the accuracy of the target control scheme can be improved and the control effect of the target ramp can be improved; the target representation includes a historical traffic representation, which can reflect the historical traffic conditions and further improve the control effect.

[0050] In order to better understand the embodiments of the present disclosure, application scenarios to which the embodiments of the present disclosure can be applied are described.

[0051] Figure 2 It is a schematic diagram of an application scenario for implementing the embodiment of the present disclosure.

[0052] like Figure 2 As shown, in this embodiment, it is taken as an example that the target road connected to the target ramp includes a target main road and a target auxiliary road.

[0053] like Figure 2 As shown, this scenario includes: a target data provider 201, a controller 202, and a traffic light 203 corresponding to the target ramp.

[0054] The target data provider 201 is used to provide target data.

[0055] The target data may include: historical traffic volume of the target road, and historical control plans of the target ramp.

[0056] Historical traffic may include: main road historical traffic and auxiliary road historical traffic.

[0057] Taking the historical traffic volume of a main road as an example, the historical traffic volume of a main road refers to the number of vehicles traveling on the target main road at a historical moment.

[0058] The control scheme of the target ramp at a historical moment is called a historical control scheme. For a certain historical control scheme, it can be represented by 1 (indicating that the target ramp at the historical moment is open) or 0 (indicating that the target ramp at the historical moment is closed).

[0059] The target data may also include: historical external data, where the historical external data refers to external data of a preset type at a historical moment.

[0060] The preset type of external data may include one or more of time information, date information, lane information, etc. Time information, such as 8:10 am on Monday; date information, such as working day or weekend; lane information, such as the target main road is 3 lanes and the target auxiliary road is 1 lane.

[0061] The target data provider may specifically include a database, in which time data and road network data are recorded. The time information and date information may be obtained through the time data, and the lane information may be obtained through the road network data.

[0062] The controller 202 is used to obtain a target representation according to the target data, and determine a target control scheme according to the target representation.

[0063] The controller may specifically include a server, a processing unit, etc.

[0064] Traffic light 203 is used to generate a corresponding control signal according to the target control scheme. For example, if the target control scheme is on, the traffic light will display a green light, indicating that vehicles in the target auxiliary road can enter the target main road through the target ramp; or if the target control scheme is off, the traffic light will display a red light, indicating that vehicles in the target auxiliary road cannot enter the target main road through the target ramp.

[0065] The controller may specifically adopt a pre-trained ramp control model and determine a target control scheme according to target data.

[0066] Figure 3 It is a schematic diagram of determining a target control scheme based on a ramp control model according to an embodiment of the present disclosure.

[0067] like Figure 3 As shown, the ramp control model 300 includes a representation extraction network 301 and a solution generation network 302 .

[0068] The input of the representation extraction network 301 is the target data, and the output is the target representation.

[0069] The input of the solution generation network 302 is the target representation, and the output is the target control solution of the target ramp at the target time.

[0070] Figure 4 is a schematic diagram of a representation extraction network provided according to an embodiment of the present disclosure.

[0071] like Figure 4 As shown, the input of the characterization extraction network is the target data, which includes: the historical traffic of the target road and the historical control scheme of the target ramp.

[0072] In this embodiment, it is taken that the target road includes a target main road and a target secondary road as an example, and therefore, the historical traffic includes: the historical traffic of the main road and the historical traffic of the secondary road.

[0073] The target data may also include: external data at historical moments, which may be referred to as historical external data. The representation corresponding to the historical external data may be referred to as historical external representation.

[0074] External data may include: time information, date information, and lane information.

[0075] Time information, such as 8:10 am on Monday; date information, such as working day or weekend; lane information, such as the target main road is 3 lanes and the target auxiliary road is 1 lane.

[0076] The output of the representation extraction network is the target representation, which includes the historical traffic representation, specifically: the main road traffic representation and the auxiliary road traffic representation.

[0077] Among them, the main road traffic characterization can be obtained based on the historical traffic of the auxiliary road, the historical control plan, the historical external data and the historical traffic of the main road; the auxiliary road traffic characterization can be obtained based on the historical traffic of the main road and the historical external data.

[0078] Specifically, the representation extraction network includes a main path representation extraction network and an auxiliary path representation extraction network.

[0079] The auxiliary road representation extraction network is used to learn the relationship between the auxiliary road traffic representation and the main road traffic data, such as Figure 4 As shown, it specifically includes: main road traffic encoding layer, external data encoding layer and auxiliary road representation decoding layer. The input of the auxiliary road representation extraction network includes the main road historical traffic and historical external data; the output is the auxiliary road traffic representation.

[0080] The main road representation extraction network is used to learn the impact of auxiliary road traffic on the main road, such as Figure 4 As shown, it specifically includes: auxiliary road traffic encoding layer, gating layer, external data encoding layer, main road traffic encoding layer and main road representation decoding layer. The input of the auxiliary road representation extraction network includes auxiliary road historical traffic, historical scheme, historical external data and main road historical traffic; the output is the main road traffic representation.

[0081] In this embodiment, the traffic characterization of the auxiliary road is obtained based on the historical traffic flow of the main road and the historical external data, and the influence of the main road and external data on the auxiliary road can be considered to improve the accuracy of the traffic characterization of the auxiliary road; the traffic characterization of the main road is obtained based on the historical traffic flow of the auxiliary road, the historical external data, the historical control scheme and the historical traffic flow of the main road, and the influence of various factors on the main road can be considered to improve the accuracy of the traffic characterization of the main road, thereby improving the ramp control effect.

[0082] Among them, for the auxiliary road traffic characterization: the main road historical traffic can be encoded to obtain the main road traffic characterization; the historical external data can be encoded to obtain the historical external characterization; the main road traffic characterization and the historical external characterization can be decoded to obtain the auxiliary road traffic characterization.

[0083] For the main road traffic characterization: encode the auxiliary road historical traffic to obtain the auxiliary road traffic characterization; obtain the weighted value based on the historical control scheme; obtain the weighted auxiliary road traffic characterization based on the weighted value and the auxiliary road traffic characterization; encode the historical main road traffic data to obtain the main road traffic characterization; encode the historical external data to obtain the historical external characterization; decode the weighted auxiliary road traffic characterization, the main road traffic characterization and the historical external characterization to obtain the main road traffic characterization.

[0084] Furthermore, for the main road traffic characterization: a main road traffic characterization layer is used to encode the main road historical traffic to obtain the main road traffic characterization.

[0085] Specifically, Figure 4 As shown, the main road traffic encoding layer includes a linear layer and an encoding layer, and the encoding layer can specifically adopt an encoding layer based on a residual network. Based on this, the main road traffic encoding layer first linearizes the main road historical traffic, and then encodes the linearized data to obtain the main road traffic representation.

[0086] The formula is:

[0087]

[0088] Among them, h main It is the main road flow characterization; X main is the historical traffic of the main road; Linear() represents the linear layer, Represents the residual network-based encoding layer corresponding to the main path.

[0089] For auxiliary road traffic characterization: the auxiliary road traffic characterization layer is used to encode the auxiliary road historical traffic to obtain the auxiliary road traffic characterization.

[0090] Specifically, Figure 4As shown, the auxiliary road traffic encoding layer includes a linear layer and an encoding layer, and the encoding layer can specifically adopt an encoding layer based on a residual network. Based on this, the auxiliary road traffic encoding layer first linearizes the auxiliary road historical traffic, and then encodes the linearized data to obtain the auxiliary road traffic representation.

[0091] The formula is:

[0092]

[0093] Among them, h r It is the characterization of auxiliary road flow; X ramp is the historical traffic of the auxiliary road; Linear() represents the linear layer; Represents the residual network-based encoding layer corresponding to the auxiliary channel.

[0094] For historical external representation: adopt the external data representation encoding layer to encode the historical external data to obtain the historical external representation.

[0095] Specifically, Figure 4 As shown, various historical external data can be embedded separately to obtain embedded representations; then the embedded representations are fused to obtain fused representations; and then the fused representations are encoded, specifically, the encoding layer based on the residual network can be used for encoding to obtain the historical external representation.

[0096] The formula is:

[0097] h time =Embedding time (X time )

[0098] h weekday =Embedding weekday (X weekday )

[0099] h line =Linear(X lane )

[0100] h=Linear(Concat(h time ;h weekday ;h lane ))

[0101]

[0102] Among them, h time is an embedded representation of the temporal information of a historical moment; h weekday is an embedded representation of the date information of a historical moment; h lane is the embedded representation of lane information at a historical moment; Xtime It is the time information of the historical moment; Embedding time is the embedding layer corresponding to the time information; X weekday It is the date information of the historical moment; Embedding weekday is the embedding layer corresponding to the date information; X lane is the lane information at the historical moment; Linear(X lane ) is the embedding layer corresponding to the lane information, specifically the linear layer; h is the fusion information; Concat() is the merging operation; h ex It is an external representation of history; It is the residual network-based encoding layer corresponding to the external data.

[0103] Regarding the weighted auxiliary road traffic characterization: the gating layer is used to process the historical control scheme to obtain the weighted value, and then the weighted auxiliary road traffic characterization is obtained according to the weighted value and the auxiliary road traffic characterization.

[0104] Specifically, Figure 4 As shown, the historical control scheme can be linearized by using a linear layer, and then the linearized value can be normalized, such as by using a sigmoid function to obtain a normalized value, and then a weighted value can be obtained according to the normalized value.

[0105] The formula is:

[0106] h gate =1+sigmoid(Linear(S))

[0107] h r '=h r *h gate =h r *(1+sigmoid(Linear(S)))

[0108] Among them, h r ' is the weighted auxiliary road flow characterization; h r is the characterization of auxiliary road flow; gate is the weight value; S is the historical control scheme; Linear() represents the linear layer.

[0109] After obtaining the above-mentioned main road flow representation and historical external representation, the auxiliary road representation decoding layer is used to decode the main road flow representation and the historical external representation to obtain the auxiliary road traffic representation.

[0110] Specifically, Figure 4As shown, the auxiliary road representation decoding layer includes: a decoding layer and a linear layer. The decoding layer can specifically adopt a residual network-based decoding layer. Based on this, the decoding layer is first used to decode the main road traffic representation and the historical external representation, and then the decoded representation is linearized to obtain the auxiliary road traffic representation.

[0111] The formula is:

[0112]

[0113] Among them, h rn It is the characterization of auxiliary road traffic; main is the main road flow characterization; h ex It is the external representation of history; Concat() represents the merging operation; It represents the residual network-based decoding layer corresponding to the auxiliary path; Linear() represents the linear layer.

[0114] In this embodiment, by obtaining the main road flow representation and the historical external representation, these representations are decoded to obtain the auxiliary road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the auxiliary road traffic representation and improving the control effect of the target ramp.

[0115] After obtaining the above-mentioned main road flow representation, weighted auxiliary road flow representation and historical external representation, the main road representation decoding layer is used to decode the above-mentioned representations to obtain the main road traffic representation.

[0116] Specifically, Figure 4 As shown, the main road representation decoding layer includes: a decoding layer and a linear layer. The decoding layer can specifically adopt a decoding layer based on a residual network. Based on this, the decoding layer is first used to decode the main road flow representation, the historical external representation and the weighted auxiliary road flow representation, and then the decoded representation is linearized to obtain the main road traffic representation.

[0117] The formula is:

[0118]

[0119] Among them, h mn It is the main road traffic characterization; main is the main road flow characterization; h ex is the external data representation; h r ' is the weighted auxiliary road traffic representation; Concat() represents the merging operation; Represents the residual network-based decoding layer corresponding to the main path; Linear() represents linearization.

[0120] The above-mentioned encoding layer or decoding layer based on the residual network includes N residual blocks, each of which is composed of the following modules:

[0121] h i '=f(Linear(h i ))

[0122] h i = Dropout(Linear(h i '))

[0123] h i+1 =h i ”+linear(h i )

[0124] Among them, h i is the input representation of a single residual block; h i+1 is the output representation of the single residual block; h i ',h i ” is the representation of the intermediate operation; Linear() is the linear layer; Dropout() is the discarding process, such as random discarding; f() is the activation function, such as the ReLu activation function, the expression is: f(x) = max(0,x).

[0125] In this embodiment, by obtaining the main road flow representation, the historical external representation and the weighted auxiliary road flow representation, these representations are decoded to obtain the main road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the main road traffic representation and improving the control effect of the target ramp.

[0126] Through the above processing, the historical traffic representation can be obtained, including the main road traffic representation h mn and auxiliary road traffic characterization h rn .

[0127] In addition, in addition to the above-mentioned historical popular representations, the target representation can also include target external representations.

[0128] The target external representation is the external data representation at the target moment, which can be specifically determined based on the external data at the target moment (moment t). The acquisition process is consistent with the processing principle of the historical external representation and will not be elaborated herein.

[0129] Afterwards, a target control scheme can be determined based on the target characterization.

[0130] Figure 5 It is a schematic diagram of generating a network according to the solution provided in an embodiment of the present disclosure.

[0131] like Figure 5 As shown, the input of the solution generation network is the target representation and the output is the target control solution.

[0132] The target control scheme at the target time is determined based on a preset number of historical traffic indicators and the target external representation.

[0133] Assuming that the preset number is represented by M (M is a positive integer) and the target time is represented by time t, then M main road traffic representations can form a main road traffic representation sequence, which is expressed as M auxiliary road traffic representations can form an auxiliary road traffic representation sequence, expressed as The target external representation can be used express.

[0134] The solution generation network may specifically include: an encoding layer and a decoding layer.

[0135] The input of the coding layer includes: the main road traffic characterization sequence and the auxiliary road traffic characterization sequence, and the output includes: the main road code at the target time, Indicated by; and the auxiliary road code of the target time, express.

[0136] The input of the decoding layer includes: the main channel code at the target time Auxiliary road code at target time and target external representation The output is the target control solution S t .

[0137] Based on this, the preset number of main road traffic representations can be encoded to obtain the main road coding at the target moment; the preset number of auxiliary road traffic representations can be encoded to obtain the auxiliary road coding at the target moment; the main road coding at the target moment, the auxiliary road coding at the target moment and the target external representation can be decoded to obtain the target control scheme.

[0138] Specifically, Figure 5 As shown:

[0139] Main road encoding for the target time: first encode the main road traffic representation sequence, such as encoding it using a residual network-based encoding layer, and then further encode the encoded representation, such as using a long short-term memory (LSTM) network to obtain the main road encoding at the target time. The LSTM network is a time-recurrent neural network.

[0140] The formula is:

[0141]

[0142] in, is the main path code at the target time (time t); It is a sequence of M (M is a preset value) main road traffic representations before time t; Represents the encoding layer based on the residual network; LSTM() represents the LSTM network.

[0143] Auxiliary road encoding at the target time: first encode the auxiliary road traffic representation sequence, such as encoding it using a residual network-based encoding layer, and then further encode the encoded representation, such as further encoding it using LSTM, to obtain the auxiliary road encoding at the target time.

[0144] The formula is:

[0145]

[0146] in, is the auxiliary road code at the target time (time t); is a sequence of M (M is a preset value) auxiliary road traffic representations before time t; Represents the encoding layer based on the residual network; LSTM() represents the LSTM network.

[0147] For the decoding layer: decode the main path coding at the target time, the auxiliary path coding at the target time and the target external representation to obtain the target control solution.

[0148] Specifically, Figure 5 As shown, the first linear layer is first used to linearize the main path encoding at the target moment, the auxiliary path encoding at the target moment, and the target external representation; then the decoding layer based on the residual network is used to decode the linearized representation; then the second linear layer is used to linearize the decoded representation to obtain the target control solution.

[0149] The formula is:

[0150]

[0151] in, is the main path code at the target time (time t); is the auxiliary road code at time t; It is the external representation of the target; Concat() represents the merging operation; Linear() represents linearization; Represents the decoding layer based on the residual network.

[0152] S p is the probability of the candidate control scheme (on or off), and then the candidate control scheme with the largest probability is taken as the target control scheme S at the target time (time t). t .

[0153] In this embodiment, the target control scheme is determined based on a preset number of historical traffic representations and target external representations, and a variety of information can be referred to to improve the accuracy of the target control scheme.

[0154] Based on the above processing, the target control scheme at the target time is obtained.

[0155] Afterwards, the target ramp is controlled using the target control scheme.

[0156] Based on the above application scenarios, the present disclosure also provides the following embodiments.

[0157] Figure 6 is a schematic diagram according to a second embodiment of the present disclosure, which provides a ramp control method.

[0158] In this embodiment, a ramp control model is adopted to determine a target control scheme based on target data, and then the target ramp is controlled according to the target control scheme.

[0159] The ramp control model includes: a representation extraction network and a solution generation network.

[0160] like Figure 6 As shown, the method of this embodiment includes:

[0161] 601. Using a representation extraction network, a target representation is obtained based on the target data.

[0162] Among them, reference Figure 4 , the input of the representation extraction network is the target data, and the output is the target representation.

[0163] The target data includes: historical traffic on the main road, historical traffic on the auxiliary road, historical control plans and historical external data. The historical external data includes time information, date information and lane information at historical moments.

[0164] The target representation includes historical traffic representation, specifically including: main road traffic representation and auxiliary road traffic representation.

[0165] The target representation also includes: target external representation, that is, the external data representation at the target moment.

[0166] Among them, the traffic characterization of the auxiliary road can be obtained based on the historical traffic of the main road and the historical external data; the traffic characterization of the main road can be obtained based on the historical traffic of the main road, the historical traffic of the auxiliary road, the historical control scheme and the historical external data.

[0167] Specifically, for the characterization of auxiliary road traffic: refer to Figure 4 , the external data representation encoding layer is used to encode the historical external data to obtain the historical external representation; the main road historical traffic encoding layer is used to encode the main road historical traffic to obtain the main road traffic representation; the auxiliary road representation decoding layer is used to decode the external data representation and the main road traffic representation to obtain the auxiliary road traffic representation.

[0168] For main road traffic characterization: Reference Figure 4 The auxiliary road historical traffic encoding layer is used to encode the auxiliary road historical traffic to obtain the auxiliary road traffic representation; the gating layer is used to process the historical control scheme to obtain the weighted value; according to the weighted value and the auxiliary road traffic representation, the weighted auxiliary road traffic representation is obtained; the main road representation decoding layer is used to decode the historical external representation, the main road traffic representation and the weighted auxiliary road traffic representation to obtain the main road traffic representation.

[0169] The number of historical traffic representations is a preset number, such as M (M is a positive integer), and each historical traffic representation is obtained based on the historical data of the corresponding historical moment. For example, for moment (t-1), the main road traffic representation at moment (t-1) is It is based on the historical traffic volume of the main road at time (t-1) Historical traffic volume of auxiliary roads at time (t-1) The historical control scheme S at time (t-1) t-1 and historical external data at time (t-1) Obtained; auxiliary road traffic representation at time (t-1) It is based on the historical traffic volume of the main road at time (t-1) and historical external data at time (t-1) obtained; Including time information at (t-1) Date Information and lane information

[0170] 602. Using a solution generation network, determine a target control solution for the target ramp at a target time based on the target representation.

[0171] Assuming that the target time is time t, the target control scheme S at time t is determined by using M (M is a preset value) historical traffic representations before time t and the external data representation at time t (target external representation). t .

[0172] Among them, the preset number of main road traffic representations can be encoded to obtain the main road coding at the target moment; the preset number of auxiliary road traffic representations can be encoded to obtain the auxiliary road coding at the target moment; the main road coding at the target moment, the auxiliary road coding at the target moment and the target external representation can be decoded to obtain the target control scheme.

[0173] Specifically, refer to Figure 5 ,The input of the solution generation network is the target representation, and the output is the target control solution.

[0174] The target representation includes: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations, and a target external representation, wherein the preset number of main road traffic representations constitute a main road traffic representation sequence, and the preset number of auxiliary road traffic representations constitute an auxiliary road traffic representation sequence.

[0175] The solution generation network includes: encoding layer and decoding layer.

[0176] The encoding layer is used to encode the main road traffic characterization sequence to obtain the main road code at the target time, and the auxiliary road traffic characterization sequence is encoded to obtain the auxiliary road code at the target time; the decoding layer is used to decode the main road code at the target time, the auxiliary road code at the target time and the target external representation to obtain the target control plan.

[0177] The coding layer may specifically include a coding layer based on a residual network and an LSTM. Based on this, taking the main road as an example, the coding layer based on the residual network is used to encode the main road traffic representation sequence to obtain a coding representation, and then the LSTM is used to encode the coding representation to obtain the main road coding.

[0178] The decoding layer may specifically include: a first linear layer, a decoding layer based on a residual network, and a second linear layer. The first linear layer is used to linearize the main path coding, the auxiliary path coding, and the target external representation, and then the encoding layer based on the residual network is used to encode the representation output by the first linear layer, and then the second linear layer is used to linearize the representation output by the encoding layer based on the residual network to obtain the probability of each candidate control scheme (on or off), and the candidate control scheme with a large probability is used as the target control scheme.

[0179] 603. At the target time, control the target ramp based on the target control scheme.

[0180] For example, if the target control scheme is on, the traffic light shows green, indicating that vehicles in the target auxiliary road can enter the target main road through the target ramp; or, if the target control scheme is off, the traffic light shows red, indicating that vehicles in the target auxiliary road cannot enter the target main road through the target ramp.

[0181] Figure 7 is a schematic diagram according to the third embodiment of the present disclosure, which provides a ramp control model training method, the ramp control model includes: a representation extraction network and a solution generation network, such as Figure 7 As shown, the training method includes:

[0182] 701. Use the representation extraction network to obtain a target representation; the target representation includes: a historical traffic representation of a target road connected to a target ramp before a target time.

[0183] 702. Using the solution generation network, based on the target representation, determine a prediction control solution for the target ramp at the target time.

[0184] 703. Construct a target loss function based on the prediction control scheme and the label control scheme.

[0185] 704. Based on the target loss function, adjust the parameters of the solution generation network and fix the parameters of the representation extraction network.

[0186] In this embodiment, the target control scheme is determined based on the target representation. Compared with the method based on preset time or preset threshold, the accuracy of the target control scheme can be improved, the effect of the ramp control model can be improved, and then the control effect of the target ramp can be improved; the target representation includes a historical traffic representation, which can reflect the historical traffic conditions and further improve the model and control effects.

[0187] In some embodiments, the target road includes: a target main road and a target auxiliary road;

[0188] The historical traffic representation includes: main road traffic representation and auxiliary road traffic representation;

[0189] The representation extraction network includes: a main path representation extraction network and an auxiliary path representation extraction network;

[0190] The adopting the representation extraction network to obtain the target representation includes:

[0191] Using the auxiliary road representation extraction network, based on the main road historical traffic samples and historical external data samples of the target main road, the traffic representation of the auxiliary road is obtained;

[0192] The main road characterization extraction network is adopted to obtain the main road traffic characterization based on the main road historical traffic samples, the auxiliary road historical traffic samples of the target auxiliary road, the historical external data samples and the historical control scheme samples of the target ramp.

[0193] In this embodiment, the traffic characterization of the auxiliary road is obtained based on the historical traffic samples of the main road and the historical external data samples, and the influence of the main road and external data on the auxiliary road can be considered to improve the accuracy of the traffic characterization of the auxiliary road; the traffic characterization of the main road is obtained based on the historical traffic samples of the auxiliary road, the historical external data samples, the historical control scheme samples and the historical traffic samples of the main road, and the influence of various factors on the main road can be considered to improve the accuracy of the traffic characterization of the main road, thereby improving the ramp control effect.

[0194] In some embodiments, the obtaining of the auxiliary road traffic representation based on the main road historical traffic and historical external data of the target main road includes: encoding the main road historical traffic to obtain the main road traffic representation; encoding the historical external data to obtain the historical external representation; decoding the main road traffic representation and the historical external representation to obtain the auxiliary road traffic representation.

[0195] In this embodiment, by obtaining the main road flow representation and the historical external representation, these representations are decoded to obtain the auxiliary road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the auxiliary road traffic representation and improving the control effect of the target ramp.

[0196] In some embodiments, the main road traffic characterization is obtained based on the main road historical traffic, the auxiliary road historical traffic of the target auxiliary road, the historical external data and the historical control scheme of the target ramp, including: encoding the auxiliary road historical traffic to obtain an auxiliary road traffic characterization; obtaining a weighted value based on the historical control scheme; obtaining a weighted auxiliary road traffic characterization based on the weighted value and the auxiliary road traffic characterization; encoding the historical main road traffic data to obtain a main road traffic characterization; encoding the historical external data to obtain a historical external characterization; decoding the weighted auxiliary road traffic characterization, the main road traffic characterization and the historical external characterization to obtain the main road traffic characterization.

[0197] In this embodiment, by obtaining the main road flow representation, the historical external representation and the weighted auxiliary road flow representation, these representations are decoded to obtain the main road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the main road traffic representation and improving the control effect of the target ramp.

[0198] In some embodiments, the historical traffic representation includes: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations;

[0199] The target representation also includes: a target external representation, which is an external data representation of the target at the moment;

[0200] The scheme generation network includes: a coding layer and a decoding layer; the scheme generation network is used to determine the prediction control scheme of the target ramp at the target time based on the target representation, including: using the coding layer to encode the preset number of main road traffic representations to obtain the main road coding at the target time; and encoding the preset number of auxiliary road traffic representations to obtain the auxiliary road coding at the target time; using the decoding layer to decode the main road coding at the target time, the auxiliary road coding at the target time and the target external representation to obtain the prediction control scheme.

[0201] In this embodiment, a target control scheme is determined based on a preset number of historical traffic representations and target external representations, and a variety of information can be referenced to improve the accuracy of the predictive control scheme.

[0202] Furthermore, the training process of the ramp control model can utilize the additionally introduced solution evaluation network to train the final representation extraction network and solution generation network.

[0203] Figure 8 It is a schematic diagram of the network involved in the training process provided by the embodiment of the present disclosure.

[0204] like Figure 8 As shown, the ramp control model includes: a representation extraction network 801 and a solution generation network 802. During the training phase, a solution evaluation network 803 is also introduced.

[0205] During training, the target sample is input into the representation extraction network, and the output is the target representation; the target representation is input into the solution generation network, and the output is the predicted control solution.

[0206] The scheme evaluation network is used to determine the label control scheme based on the target representation, and then construct the target loss function based on the prediction control scheme and the label control scheme. The target loss function is then used to adjust the model parameters, specifically adjusting the parameters of the scheme generation network and fixing the parameters of the representation extraction network.

[0207] The training process can be divided into three stages, namely the initial stage, the intermediate stage and the target stage. The initial stage is used for main and auxiliary road traffic prediction training, the intermediate stage is used for solution evaluation training, and the target stage is used for training the solution generation network. The loss functions corresponding to the above three stages are respectively called the initial loss function (loss1), the intermediate loss function (loss2) and the target loss function (loss3). In addition, the representation extraction network and the solution evaluation network are trained in the initial stage and the intermediate stage, that is, in the initial stage, the initial loss function is used to adjust the parameters of the representation extraction network and the parameters of the solution evaluation network; in the intermediate stage, based on the representation extraction network and the solution evaluation network obtained in the initial stage, the intermediate loss function is used to adjust the parameters of the representation extraction network and the parameters of the solution evaluation network again; in the target stage, based on the representation extraction network and the solution evaluation network obtained in the intermediate stage, the target loss function is used to adjust the parameters of the solution generation network, and the parameters of the representation extraction network are fixed to obtain the final representation extraction network and the solution generation network, which constitute the ramp control model.

[0208] Fig. 9 It is a schematic diagram of a solution evaluation network provided by an embodiment of the present disclosure.

[0209] The scheme evaluation network is used to predict the target flow at the target time based on the historical traffic representation, and to determine the score of each candidate control scheme based on the target flow, so that the candidate control scheme with a higher score can be used as the label control scheme.

[0210] Specifically, Fig. 9 As shown, the scheme evaluation network includes: a main road flow prediction module and an auxiliary road flow prediction module.

[0211] The input of the main road flow prediction module includes: the main road traffic characterization sequence, the auxiliary road traffic characterization sequence and the actual control plan at the target time; the output is the main road predicted flow at the target time.

[0212] The formula is:

[0213]

[0214]

[0215] Among them, Y mf is the predicted main road flow at the target time (time t); is the main road traffic characterization sequence; h rn-gate is the gate value; is the auxiliary road traffic characterization sequence; It is the real control solution at the target moment, which can be obtained through manual annotation; It represents the residual network-based decoding layer corresponding to the main path; LSTM() represents the LSTM network; Concat() represents the merging operation; Linear() represents the linear layer; sigmoid() represents the sigmoid operation, and sigmoid is a normalization function.

[0216] The input of the auxiliary road flow prediction module includes: the auxiliary road traffic characterization sequence and the actual control plan at the target time; the output is the auxiliary road predicted flow at the target time.

[0217] The formula is:

[0218]

[0219] Among them, Y rf is the auxiliary road predicted flow at the target time (time t); is the auxiliary road traffic characterization sequence; It is the real control solution at the target moment, which can be obtained through manual annotation; It represents the residual network-based decoding layer corresponding to the auxiliary path; LSTM() represents the LSTM network; Concat() represents the merging operation; Linear() represents the linear layer; sigmoid() represents the sigmoid operation, and sigmoid is a normalization function.

[0220] After obtaining the predicted flow of the main road and the predicted flow of the auxiliary road, the score of the candidate control scheme (on or off) can be determined according to the preset evaluation function.

[0221] The formula is:

[0222] Y sp =g(Y mf +Y rf )

[0223] Among them, Y sp is the score of the candidate control solution; Y mf is the predicted flow of the main road; Y rf is the predicted traffic of the auxiliary road; g() is the evaluation function, which can be set according to actual needs.

[0224] Based on the solution evaluation network, the present disclosure also provides the following training method.

[0225] Fig.10 is a schematic diagram according to a fourth embodiment of the present disclosure. This embodiment provides a ramp control model training method, the ramp control model comprising: a representation extraction network and a solution generation network, such as Fig.10 As shown, the training method includes:

[0226] 1001. Using a representation extraction network, the initial sample is processed to obtain an initial representation.

[0227] The initial samples are training samples in the initial stage, including: historical traffic samples corresponding to the initial stage (main road historical traffic samples and auxiliary road historical traffic samples), historical control scheme samples of the target ramp, and historical external data.

[0228] These initial samples are input into the representation extraction network, and the output is the initial representation.

[0229] The initial representation includes: the historical traffic representation corresponding to the initial stage (main road traffic representation and auxiliary road traffic representation) and the target external representation.

[0230] 1002. Using a solution evaluation network, determine an initial predicted flow based on the initial characterization.

[0231] The initial representation is input into the scheme evaluation network, and the output is the initial predicted traffic.

[0232] The initial predicted traffic includes: the main road predicted traffic and the auxiliary road predicted traffic at the target time corresponding to the initial stage.

[0233] 1003. Construct an initial loss function based on the initial predicted traffic and its corresponding initial real traffic.

[0234] The initial sample also includes the initial real traffic at the target time. The initial real traffic includes the real traffic on the main road and the real traffic on the auxiliary road. The initial real traffic can be obtained through manual labeling.

[0235] In this way, the initial loss function can be constructed based on the pre-labeled initial real traffic and the initial predicted traffic obtained by the scheme evaluation network.

[0236] The calculation formula of the initial loss function can be:

[0237] Loss1=L δ (Y m ,Y mf )+L δ (Y r ,Y rf )

[0238] Among them, Loss1 is the initial loss function; Y m is the real traffic volume of the main road; mf is the predicted flow of the main road; Y r is the actual flow of the auxiliary road; Y rf It is the predicted flow of auxiliary roads.

[0239] L δ () is the Huber loss function, which is expressed as follows:

[0240]

[0241] Among them, Y t is the true value; is the predicted value; δ is the preset parameter.

[0242] 1004. Based on the initial loss function, adjust parameters of the representation extraction network and the solution evaluation network.

[0243] For example, the parameters may be adjusted using a stochastic descent gradient method or an Adam optimization algorithm.

[0244] The Adam optimization algorithm is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process. It can iteratively update the neural network parameters based on training data.

[0245] The initial parameters of the representation extraction network and the solution evaluation network in the initial stage can be randomly initialized or adopt the parameters of the existing corresponding pre-trained models.

[0246] By adjusting the parameters, the final parameters of the initial stage can be obtained.

[0247] Through the above 1001 to 1004, the initial stage of training is completed.

[0248] In this embodiment, initial training can be performed based on predicted traffic, thereby improving model accuracy.

[0249] 1005. Use the representation extraction network to process the intermediate samples to obtain intermediate representations.

[0250] Among them, the intermediate samples are training samples of the intermediate stage, including: historical traffic samples corresponding to the intermediate stage (main road historical traffic samples and auxiliary road historical traffic samples), historical control plan samples of the target ramp and historical external data.

[0251] These intermediate samples are input into the representation extraction network, and the output is the intermediate representation.

[0252] The intermediate representation includes: the historical traffic representation corresponding to the intermediate stage (main road traffic representation and auxiliary road traffic representation) and the target external representation.

[0253] 1006. Using the scheme evaluation network, determine the intermediate predicted flow and the prediction scores of the multiple candidate control schemes based on the intermediate representation.

[0254] The intermediate representation is input into the scheme evaluation network, and the output includes the intermediate predicted flow and the prediction score of each candidate control scheme.

[0255] The intermediate predicted traffic includes: the main road predicted traffic and the auxiliary road predicted traffic at the target time corresponding to the intermediate stage.

[0256] The candidate control schemes include: on or off, and the prediction scores include: on score and off score.

[0257] 1007. Construct an intermediate loss function based on the intermediate predicted traffic and its corresponding intermediate real traffic, and the predicted score and its corresponding real score.

[0258] The intermediate samples also include the intermediate real traffic at the target time. The intermediate real traffic includes the real traffic on the main road and the real traffic on the auxiliary road. The intermediate real traffic can be obtained through manual labeling.

[0259] The intermediate samples also include the true scores of the candidate solutions, which can also be obtained through manual annotation.

[0260] In this way, an intermediate loss function can be constructed based on the pre-labeled intermediate true traffic and the intermediate predicted traffic obtained by the solution evaluation network, as well as the pre-labeled true scores and the predicted scores obtained by the solution evaluation network.

[0261] The calculation formula of the intermediate loss function can be:

[0262] Loss2=L δ (Y m ,Y mf )+L δ (Y r ,Y rf )+L ce (Y s ,Y sp )

[0263] Among them, Loss2 is the intermediate loss function; Y m is the real traffic volume of the main road; mf is the predicted flow of the main road; Y r is the actual flow of the auxiliary road; Y rf is the predicted flow of the auxiliary road; Y s is the true score, Y sp is the prediction score; L ce () represents the cross entropy loss function.

[0264] L δ () is the Huber loss function, which is expressed as follows:

[0265]

[0266] Among them, Y t is the true value; is the predicted value; δ is the preset parameter.

[0267] 1008. Based on the intermediate loss function, adjust parameters of the representation extraction network and the solution evaluation network.

[0268] The initial parameters of the representation extraction network and the solution evaluation network in the intermediate stage are the final parameters obtained in the initial stage. By adjusting the parameters, the final parameters of the intermediate stage can be obtained.

[0269] Through the above steps 1005 to 1008, the intermediate stage of training is completed.

[0270] In this embodiment, intermediate training can be used to perform training based on the predicted flow rate and the predicted scores of the candidate control solutions, thereby improving the accuracy of the model.

[0271] 1009. Use the representation extraction network to process the target sample to obtain the target representation.

[0272] The target samples are training samples of the target stage, including: historical traffic samples corresponding to the target stage (main road historical traffic samples and auxiliary road historical traffic samples), historical control scheme samples of the target ramp, and historical external data.

[0273] These target samples are input into the representation extraction network, and the output is the target representation.

[0274] The target representation includes: the historical traffic representation corresponding to the target stage (main road traffic representation and auxiliary road traffic representation) and the target external representation.

[0275] 1010. Using the scheme evaluation network, determine target scores of the multiple candidate control schemes based on the target representation.

[0276] The target representation is input into the scheme evaluation network, and the output includes the target score of each candidate control scheme.

[0277] 1011. Determine a label control scheme from among the multiple candidate control schemes based on the target score.

[0278] Among them, the candidate control scheme with a higher target score is determined as the label control scheme. For example, if the score of the opening scheme is greater than the score of the closing scheme, then the label control scheme is opening.

[0279] In this embodiment, the target scores of the candidate control schemes obtained through the scheme evaluation network are used to determine the label control schemes, which can improve the accuracy of the label control schemes and improve efficiency compared to manual labeling, thereby enabling accurate and efficient model training.

[0280] 1012. Use the scheme generation network to determine a predictive control scheme for the target ramp at the target time based on the target representation.

[0281] Among them, the target representation is input into the solution generation network, and the output is the predictive control solution.

[0282] 1013. Based on the prediction control scheme and the label control scheme, construct a target loss function.

[0283] Among them, the target loss function can adopt the cross entropy loss function, and the calculation formula is:

[0284] Loss3=L ce (S j ,S p )

[0285] Among them, Loss3 is the target loss function; S j is a label control scheme; S p is the predictive control scheme; L ce () represents the cross entropy loss function.

[0286] 1014. Based on the target loss function, adjust the parameters of the solution generation network and fix the parameters of the representation extraction network.

[0287] Through the above steps 1009 to 1014, the training of the target stage is completed.

[0288] Afterwards, the representation extraction network and the solution generation network obtained in the target stage are combined into a ramp control model, and the ramp control model is used to determine the target control solution in the reasoning stage so as to control the target ramp according to the target control solution.

[0289] Fig.11 1 is a schematic diagram according to the fifth embodiment of the present disclosure. This embodiment provides a ramp control device, the device 1100 includes: an acquisition module 1101 , a determination module 1102 and a control module 1103 .

[0290] The acquisition module 1101 is used to obtain a target representation; the target representation includes: a historical traffic representation of a target road connected to the target ramp before a target time; the determination module 1102 is used to determine a target control scheme for the target ramp at the target time based on the target representation; and the control module 1103 is used to control the target ramp at the target time based on the target control scheme.

[0291] In this embodiment, the target control scheme is determined based on the target representation. Compared with the method based on preset time or preset threshold, the accuracy of the target control scheme can be improved and the control effect of the target ramp can be improved; the target representation includes a historical traffic representation, which can reflect the historical traffic conditions and further improve the control effect.

[0292] In some embodiments, the target road includes: a target main road and a target secondary road; the historical traffic representation includes: a main road traffic representation and a secondary road traffic representation; the acquisition of the target representation includes: acquiring the secondary road traffic representation based on the main road historical traffic and historical external data of the target main road; acquiring the main road traffic representation based on the main road historical traffic, the secondary road historical traffic of the target secondary road, the historical external data and the historical control plan of the target ramp.

[0293] In this embodiment, the traffic characterization of the auxiliary road is obtained based on the historical traffic flow of the main road and the historical external data, and the influence of the main road and external data on the auxiliary road can be considered to improve the accuracy of the traffic characterization of the auxiliary road; the traffic characterization of the main road is obtained based on the historical traffic flow of the auxiliary road, the historical external data, the historical control scheme and the historical traffic flow of the main road, and the influence of various factors on the main road can be considered to improve the accuracy of the traffic characterization of the main road, thereby improving the ramp control effect.

[0294] In some embodiments, the historical traffic representation includes: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations; the target representation also includes: a target external representation, which is an external data representation of the target moment; determining the target control scheme of the target ramp at the target moment based on the target traffic representation includes: encoding the preset number of main road traffic representations to obtain the main road coding at the target moment; encoding the preset number of auxiliary road traffic representations to obtain the auxiliary road coding at the target moment; decoding the main road coding at the target moment, the auxiliary road coding at the target moment and the target external representation to obtain the target control scheme.

[0295] In this embodiment, the target control scheme is determined based on a preset number of historical traffic representations and target external representations, and a variety of information can be referred to to improve the accuracy of the target control scheme.

[0296] In some embodiments, the obtaining of the auxiliary road traffic representation based on the main road historical traffic and historical external data of the target main road includes: encoding the main road historical traffic to obtain the main road traffic representation; encoding the historical external data to obtain the historical external representation; decoding the main road traffic representation and the historical external representation to obtain the auxiliary road traffic representation.

[0297] In this embodiment, by obtaining the main road flow representation and the historical external representation, these representations are decoded to obtain the auxiliary road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the auxiliary road traffic representation and improving the control effect of the target ramp.

[0298] In some embodiments, the main road traffic characterization is obtained based on the main road historical traffic, the auxiliary road historical traffic of the target auxiliary road, the historical external data and the historical control scheme of the target ramp, including: encoding the auxiliary road historical traffic to obtain an auxiliary road traffic characterization; obtaining a weighted value based on the historical control scheme; obtaining a weighted auxiliary road traffic characterization based on the weighted value and the auxiliary road traffic characterization; encoding the historical main road traffic data to obtain a main road traffic characterization; encoding the historical external data to obtain a historical external characterization; decoding the weighted auxiliary road traffic characterization, the main road traffic characterization and the historical external characterization to obtain the main road traffic characterization.

[0299] In this embodiment, by obtaining the main road flow representation, the historical external representation and the weighted auxiliary road flow representation, these representations are decoded to obtain the main road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the main road traffic representation and improving the control effect of the target ramp.

[0300] Fig.12 This is a schematic diagram according to the sixth embodiment of the present disclosure. This embodiment provides a ramp control model training device, the ramp control model includes: a characterization extraction network and a solution generation network, the device 1200 includes: an acquisition module 1201, a determination module 1202, a construction module 1203 and an adjustment module 1204.

[0301] The acquisition module 1201 is used to adopt the representation extraction network to obtain the target representation; the target representation includes: the historical traffic representation of the target road connected to the target ramp before the target time; the determination module 1202 is used to adopt the solution generation network, and determine the prediction control solution of the target ramp at the target time based on the target representation; the construction module 1203 is used to construct a target loss function according to the prediction control solution and the label control solution; the adjustment module 1204 is used to adjust the parameters of the solution generation network according to the target loss function, and fix the parameters of the representation extraction network.

[0302] In this embodiment, the target control scheme is determined based on the target representation. Compared with the method based on preset time or preset threshold, the accuracy of the target control scheme can be improved, the effect of the ramp control model can be improved, and then the control effect of the target ramp can be improved; the target representation includes a historical traffic representation, which can reflect the historical traffic conditions and further improve the model and control effects.

[0303] In some embodiments, the apparatus 1200 further includes:

[0304] The label module is used to determine target scores of multiple candidate control schemes using a scheme evaluation network; and determine the label control scheme among the multiple candidate control schemes based on the target scores.

[0305] In this embodiment, the target scores of the candidate control schemes obtained through the scheme evaluation network are used to determine the label control schemes, which can improve the accuracy of the label control schemes and improve efficiency compared to manual labeling, thereby enabling accurate and efficient model training.

[0306] In some embodiments, the apparatus 1200 further includes:

[0307] An intermediate training module is used to use the representation extraction network to process intermediate samples to obtain intermediate representations; use the solution evaluation network to determine the intermediate predicted flow and the predicted scores of the multiple candidate control solutions based on the intermediate representations; construct an intermediate loss function based on the intermediate predicted flow and its corresponding intermediate true flow, as well as the predicted score and its corresponding true score; and adjust the parameters of the representation extraction network and the solution evaluation network based on the intermediate loss function.

[0308] In this embodiment, intermediate training can be used to perform training based on the predicted flow rate and the predicted scores of the candidate control solutions, thereby improving the accuracy of the model.

[0309] In some embodiments, the apparatus 1200 further includes:

[0310] The initial training module is used to use the representation extraction network to process the initial sample to obtain the initial representation; use the solution evaluation network to determine the initial predicted flow based on the initial representation; construct an initial loss function based on the initial predicted flow and its corresponding initial true flow; and adjust the parameters of the representation extraction network and the solution evaluation network based on the initial loss function.

[0311] In this embodiment, initial training can be performed based on predicted traffic, thereby improving model accuracy.

[0312] In some embodiments, the target road includes: a target main road and a target secondary road; the historical traffic representation includes: a main road traffic representation and a secondary road traffic representation; the representation extraction network includes: a main road representation extraction network and a secondary road representation extraction network; the acquisition module is further used to: adopt the secondary road representation extraction network to obtain the secondary road traffic representation based on the main road historical traffic samples and historical external data samples of the target main road; adopt the main road representation extraction network to obtain the main road traffic representation based on the main road historical traffic samples, the secondary road historical traffic samples of the target secondary road, the historical external data samples and the historical control scheme samples of the target ramp.

[0313] In this embodiment, the traffic characterization of the auxiliary road is obtained based on the historical traffic samples of the main road and the historical external data samples, and the influence of the main road and external data on the auxiliary road can be considered to improve the accuracy of the traffic characterization of the auxiliary road; the traffic characterization of the main road is obtained based on the historical traffic samples of the auxiliary road, the historical external data samples, the historical control scheme samples and the historical traffic samples of the main road, and the influence of various factors on the main road can be considered to improve the accuracy of the traffic characterization of the main road, thereby improving the ramp control effect.

[0314] In some embodiments, the historical traffic representation includes: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations; the target representation also includes: a target external representation, which is an external data representation of the target moment; the scheme generation network includes: a coding layer and a decoding layer; the determination module is further used to: use the coding layer to encode the preset number of main road traffic representations to obtain the main road coding at the target moment; and, encode the preset number of auxiliary road traffic representations to obtain the auxiliary road coding at the target moment; use the decoding layer to decode the main road coding at the target moment, the auxiliary road coding at the target moment and the target external representation to obtain the predictive control scheme.

[0315] In this embodiment, by obtaining the main road flow representation and the historical external representation, these representations are decoded to obtain the auxiliary road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the auxiliary road traffic representation and improving the control effect of the target ramp.

[0316] In some embodiments, the acquisition module is further used to: encode the main road historical traffic to obtain the main road traffic representation; encode the historical external data to obtain the historical external representation; decode the main road traffic representation and the historical external representation to obtain the auxiliary road traffic representation.

[0317] In this embodiment, by obtaining the main road flow representation, the historical external representation and the weighted auxiliary road flow representation, these representations are decoded to obtain the main road traffic representation, and the relevant historical data can be processed using a deep neural network to obtain the corresponding representation, thereby improving the accuracy of the main road traffic representation and improving the control effect of the target ramp.

[0318] In some embodiments, the acquisition module is further used to: encode the historical traffic of the auxiliary road to obtain an auxiliary road traffic representation; obtain a weighted value based on the historical control scheme; obtain a weighted auxiliary road traffic representation based on the weighted value and the auxiliary road traffic representation; encode the historical main road traffic data to obtain a main road traffic representation; encode the historical external data to obtain a historical external representation; decode the weighted auxiliary road traffic representation, the main road traffic representation and the historical external representation to obtain the main road traffic representation.

[0319] In this embodiment, a target control scheme is determined based on a preset number of historical traffic representations and target external representations, and a variety of information can be referenced to improve the accuracy of the predictive control scheme.

[0320] It can be understood that in the embodiments of the present disclosure, the same or similar contents in different embodiments can be referenced to each other.

[0321] It can be understood that the “first”, “second”, etc. in the embodiments of the present disclosure are only used for distinction and do not indicate the degree of importance, time sequence, etc.

[0322] It is understandable that, unless otherwise specified, the order of the steps in the process indicates that the timing relationship between these steps is not limited.

[0323] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0324] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0325] Fig.13A schematic block diagram of an example electronic device 1300 that can be used to implement an embodiment of the present disclosure is shown. The electronic device 1300 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0326] like Fig.13 As shown, the electronic device 1300 includes a computing unit 1301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1302 or a computer program loaded from a storage unit 1308 into a random access memory (RAM) 1303. In the RAM 1303, various programs and data required for the operation of the electronic device 1300 can also be stored. The computing unit 1301, the ROM 1302, and the RAM 1303 are connected to each other via a bus 1304. An input / output (I / O) interface 1305 is also connected to the bus 1304.

[0327] Multiple components in the electronic device 1300 are connected to the I / O interface 1305, including: an input unit 1306, such as a keyboard, a mouse, etc.; an output unit 1307, such as various types of displays, speakers, etc.; a storage unit 1308, such as a disk, an optical disk, etc.; and a communication unit 1309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1309 allows the electronic device 1300 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0328] The computing unit 1301 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1301 performs the various methods and processes described above, such as a ramp control method or a ramp control model training method. For example, in some embodiments, the ramp control method or the ramp control model training method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1300 via the ROM 1302 and / or the communication unit 1307. When the computer program is loaded into the RAM 1303 and executed by the computing unit 1301, one or more steps of the ramp control method or the ramp control model training method described above may be performed. Alternatively, in other embodiments, the computing unit 1301 may be configured to execute the ramp control method or the ramp control model training method in any other appropriate manner (eg, by means of firmware).

[0329] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0330] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable information recommendation device, so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0331] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0332] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0333] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0334] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0335] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0336] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A ramp control method, comprising: Acquire target representation; The target representation includes: a historical traffic representation of a target road connected to the target ramp before the target time; Determining a target control scheme for the target ramp at the target time based on the target representation; At the target time, the target ramp is controlled based on the target control scheme.

2. The method according to claim 1, wherein: The target road includes: a target main road and a target auxiliary road; The historical traffic representation includes: main road traffic representation and auxiliary road traffic representation; The obtaining of the target representation comprises: Based on the historical main road traffic volume and historical external data of the target main road, obtaining the traffic characterization of the auxiliary road; The main road traffic characterization is acquired based on the main road historical traffic, the target auxiliary road historical traffic of the target auxiliary road, the historical external data and the historical control scheme of the target ramp.

3. The method according to claim 2, wherein: The acquiring of the auxiliary road traffic characterization based on the main road historical traffic volume and historical external data of the target main road includes: Encoding the main road historical traffic to obtain a main road traffic representation; encoding the historical external data to obtain a historical external representation; The main road traffic representation and the historical external representation are decoded to obtain the auxiliary road traffic representation.

4. The method according to claim 2, wherein: The acquiring the main road traffic characterization based on the main road historical traffic, the auxiliary road historical traffic of the target auxiliary road, the historical external data and the historical control scheme of the target ramp includes: Encoding the auxiliary road historical traffic to obtain an auxiliary road traffic representation; obtaining a weighted value based on the historical control scheme; Based on the weighted value and the auxiliary road flow representation, obtaining a weighted auxiliary road flow representation; Encoding the historical main road flow data to obtain a main road flow representation; encoding the historical external data to obtain a historical external representation; The weighted auxiliary road flow representation, the main road flow representation and the historical external representation are decoded to obtain the main road traffic representation.

5. The method according to claim 1, wherein: The historical traffic representations include: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations; The target representation also includes: a target external representation, which is an external data representation of the target at the moment; The determining, based on the target flow characterization, a target control scheme for the target ramp at a target time, comprises: Encoding the preset number of main road traffic representations to obtain the main road code at the target time; Encoding the preset number of auxiliary road traffic representations to obtain the auxiliary road code at the target time; The main path code at the target moment, the auxiliary path code at the target moment and the target external representation are decoded to obtain the target control solution.

6. A ramp control model training method, the ramp control model comprising: Characterizing an extraction network and a solution generation network, the method comprising: The target representation is obtained by using the representation extraction network; the target representation includes: the historical traffic representation of the target road connected to the target ramp before the target time; Using the scheme generation network, based on the target representation, determine a predictive control scheme for the target ramp at the target time; Based on the prediction control scheme and the label control scheme, constructing a target loss function; Based on the objective loss function, parameters of the solution generation network are adjusted, and parameters of the representation extraction network are fixed.

7. The method according to claim 6, further comprising: Using a solution evaluation network to determine target scores for multiple candidate control solutions; The labeled control solution is determined among the plurality of candidate control solutions based on the target score.

8. The method according to claim 7, further comprising: Using the representation extraction network, processing the intermediate samples to obtain intermediate representations; Using the scheme evaluation network, based on the intermediate representation, determine an intermediate predicted flow rate and prediction scores of the plurality of candidate control schemes; Based on the intermediate predicted traffic and its corresponding intermediate real traffic, and the predicted score and its corresponding real score, construct an intermediate loss function; Based on the intermediate loss function, parameters of the representation extraction network and the solution evaluation network are adjusted.

9. The method according to claim 8, further comprising: Using the representation extraction network, processing the initial sample to obtain an initial representation; Using the scheme to evaluate the network, based on the initial characterization, determine an initial predicted flow; Based on the initial predicted flow and its corresponding initial real flow, construct an initial loss function; Based on the initial loss function, parameters of the representation extraction network and the solution evaluation network are adjusted.

10. The method according to any one of claims 6 to 9, wherein: The target road includes: a target main road and a target auxiliary road; The historical traffic representation includes: main road traffic representation and auxiliary road traffic representation; The representation extraction network includes: a main path representation extraction network and an auxiliary path representation extraction network; The adopting the representation extraction network to obtain the target representation includes: Using the auxiliary road representation extraction network, based on the main road historical traffic samples and historical external data samples of the target main road, the traffic representation of the auxiliary road is obtained; The main road characterization extraction network is adopted to obtain the main road traffic characterization based on the main road historical traffic samples, the auxiliary road historical traffic samples of the target auxiliary road, the historical external data samples and the historical control scheme samples of the target ramp.

11. The method according to claim 10, wherein: The obtaining of the auxiliary road traffic characterization based on the main road historical traffic and historical external data of the target main road includes: encoding the main road historical traffic to obtain the main road traffic characterization; encoding the historical external data to obtain the historical external characterization; decoding the main road traffic characterization and the historical external characterization to obtain the auxiliary road traffic characterization; and / or, The method of obtaining the main road traffic characterization based on the main road historical traffic, the auxiliary road historical traffic of the target auxiliary road, the historical external data and the historical control scheme of the target ramp includes: encoding the auxiliary road historical traffic to obtain the auxiliary road traffic characterization; obtaining a weighted value based on the historical control scheme; obtaining a weighted auxiliary road traffic characterization based on the weighted value and the auxiliary road traffic characterization; encoding the historical main road traffic data to obtain the main road traffic characterization; encoding the historical external data to obtain the historical external characterization; decoding the weighted auxiliary road traffic characterization, the main road traffic characterization and the historical external characterization to obtain the main road traffic characterization.

12. The method according to any one of claims 6 to 9, wherein: The historical traffic representations include: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations; The target representation also includes: a target external representation, which is an external data representation of the target at the moment; The scheme generation network includes: an encoding layer and a decoding layer; The adopting the scheme generation network to determine the prediction control scheme of the target ramp at the target time based on the target representation includes: Using the coding layer, encoding the preset number of main road traffic representations to obtain the main road code at the target time; and encoding the preset number of auxiliary road traffic representations to obtain the auxiliary road code at the target time; The decoding layer is used to decode the main path encoding at the target moment, the auxiliary path encoding at the target moment and the target external representation to obtain the predictive control scheme.

13. A ramp control device, comprising: An acquisition module, used to acquire target representation; The target representation includes: a historical traffic representation of a target road connected to the target ramp before the target time; A determination module, configured to determine a target control scheme for the target ramp at the target time according to the target representation; A control module is used to control the target ramp based on the target control scheme at the target time.

14. The device according to claim 13, wherein: The target road includes: a target main road and a target secondary road; the historical traffic representation includes: a main road traffic representation and a secondary road traffic representation; the acquisition of the target representation includes: acquiring the secondary road traffic representation based on the historical main road traffic volume and historical external data of the target main road; acquiring the main road traffic representation based on the historical main road traffic volume, the historical secondary road traffic volume of the target secondary road, the historical external data and the historical control scheme of the target ramp; and / or, The historical traffic representation includes: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations; the target representation also includes: a target external representation, which is an external data representation at the target moment; determining the target control scheme of the target ramp at the target moment based on the target flow representation includes: encoding the preset number of main road traffic representations to obtain the main road coding at the target moment; encoding the preset number of auxiliary road traffic representations to obtain the auxiliary road coding at the target moment; decoding the main road coding at the target moment, the auxiliary road coding at the target moment and the target external representation to obtain the target control scheme.

15. The device according to claim 14, wherein: The obtaining of the auxiliary road traffic characterization based on the main road historical traffic and historical external data of the target main road includes: encoding the main road historical traffic to obtain the main road traffic characterization; encoding the historical external data to obtain the historical external characterization; decoding the main road traffic characterization and the historical external characterization to obtain the auxiliary road traffic characterization; and / or, The method of obtaining the main road traffic characterization based on the main road historical traffic, the auxiliary road historical traffic of the target auxiliary road, the historical external data and the historical control scheme of the target ramp includes: encoding the auxiliary road historical traffic to obtain the auxiliary road traffic characterization; obtaining a weighted value based on the historical control scheme; obtaining a weighted auxiliary road traffic characterization based on the weighted value and the auxiliary road traffic characterization; encoding the historical main road traffic data to obtain the main road traffic characterization; encoding the historical external data to obtain the historical external characterization; decoding the weighted auxiliary road traffic characterization, the main road traffic characterization and the historical external characterization to obtain the main road traffic characterization.

16. A ramp control model training device, the ramp control model comprising: Characterization extraction network and solution generation network, the device comprises: An acquisition module, used to adopt the representation extraction network to acquire a target representation; the target representation includes: a historical traffic representation of a target road connected to the target ramp before a target time; A determination module, configured to adopt the scheme generation network and determine the prediction control scheme of the target ramp at the target time based on the target representation; A construction module, used to construct a target loss function according to the prediction control scheme and the label control scheme; An adjustment module is used to adjust the parameters of the solution generation network according to the target loss function and fix the parameters of the representation extraction network.

17. The apparatus according to claim 16, further comprising: A labeling module for determining target scores for multiple candidate control solutions using a solution evaluation network; And, based on the target score, determining the label control scheme among the plurality of candidate control schemes.

18. The apparatus according to claim 17, further comprising: An intermediate training module, used to process intermediate samples using the representation extraction network to obtain intermediate representations; Using the scheme evaluation network, based on the intermediate representation, determine an intermediate predicted flow rate and prediction scores of the plurality of candidate control schemes; Based on the intermediate predicted traffic and its corresponding intermediate real traffic, as well as the predicted score and its corresponding real score, an intermediate loss function is constructed; based on the intermediate loss function, parameters of the representation extraction network and the solution evaluation network are adjusted.

19. The apparatus according to claim 18, further comprising: An initial training module, used to process the initial sample using the representation extraction network to obtain an initial representation; Using the scheme to evaluate the network, based on the initial characterization, determine an initial predicted flow; Based on the initial predicted flow and its corresponding initial real flow, construct an initial loss function; Based on the initial loss function, parameters of the representation extraction network and the solution evaluation network are adjusted.

20. The device according to any one of claims 16 to 19, wherein: The target road includes: a target main road and a target secondary road; the historical traffic representation includes: a main road traffic representation and a secondary road traffic representation; the representation extraction network includes: a main road representation extraction network and a secondary road representation extraction network; the acquisition module is further used to: adopt the secondary road representation extraction network to obtain the secondary road traffic representation based on the main road historical traffic samples and historical external data samples of the target main road; adopt the main road representation extraction network to obtain the main road traffic representation based on the main road historical traffic samples, the secondary road historical traffic samples of the target secondary road, the historical external data samples and the historical control plan samples of the target ramp; and / or, The historical traffic representation includes: a preset number of main road traffic representations and a preset number of auxiliary road traffic representations; the target representation also includes: a target external representation, which is an external data representation of the target moment; the scheme generation network includes: a coding layer and a decoding layer; the determination module is further used to: use the coding layer to encode the preset number of main road traffic representations to obtain the main road coding at the target moment; and encode the preset number of auxiliary road traffic representations to obtain the auxiliary road coding at the target moment; use the decoding layer to decode the main road coding at the target moment, the auxiliary road coding at the target moment and the target external representation to obtain the predictive control scheme.

21. The device according to claim 20, wherein: The acquisition module is further used to: encode the historical traffic volume of the main road to obtain the main road traffic representation; encode the historical external data to obtain the historical external representation; decode the main road traffic representation and the historical external representation to obtain the auxiliary road traffic representation; and / or, The acquisition module is further used to: encode the auxiliary road historical traffic to obtain the auxiliary road traffic representation; Acquire a weighted value based on the historical control scheme; acquire a weighted auxiliary road flow representation based on the weighted value and the auxiliary road flow representation; Encoding the historical main road flow data to obtain a main road flow representation; encoding the historical external data to obtain a historical external representation; The weighted auxiliary road flow representation, the main road flow representation and the historical external representation are decoded to obtain the main road traffic representation.

22. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 12.

23. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-12.

24. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 12.

Images