Track drawing method, server, client and system

Abstract

The embodiment of the invention provides a track drawing method, a server, clients and a system. The track drawing method has the following beneficial effects that global model parameters are randomly sent to a plurality of clients; receiving reference model parameters after iterative training and noise disturbance of the plurality of clients; aggregating the reference model parameters of the plurality of clients to obtain new global model parameters; and sending the new global model parameters to the plurality of clients, so that the plurality of clients carry out a new round of iterative training. According to the method, the model parameters are subjected to noise disturbance, and a stronger privacy protection effect is achieved by increasing the difference between the disturbed data and the original data.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a trajectory mapping method, a server, a client and a system. Background Art

[0002] There are three main types of crowdsourcing mapping methods based on trajectory data: clustering, trajectory merging, and kernel density estimation. Clustering-based methods first use a clustering algorithm based on geospatial distance and direction similarity to generate a set of road endpoints, and then use trajectories to connect these road endpoints into road segments; trajectory merging-based methods scan trajectories in sequence and for each trajectory, either merge it into an existing road segment or generate a new road segment when the merge operation is not possible; kernel density estimation-based methods perform kernel density estimation on the point cloud and then extract the map.

[0003] The above crowdsourcing mapping methods often publish and execute spatiotemporal tasks while exposing the information and physical location of the tasks and workers, which is easy to expose the real data of users. Privacy protection is an important factor in crowdsourcing mapping.

[0004] Therefore, how to effectively protect privacy is a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0005] In order to overcome the shortcomings of the existing trajectory mapping method, which has poor privacy protection, the present invention proposes a trajectory mapping method, a server, a client and a system.

[0006] In order to achieve the above object, according to a first aspect of the present invention, an embodiment of the present invention provides a trajectory mapping method, which is applied to a server, and the method comprises the following steps:

[0007] Randomly send global model parameters to multiple clients;

[0008] Receiving reference model parameters after iterative training and noise disturbance by the multiple clients;

[0009] Aggregate the reference model parameters of multiple clients to obtain new global model parameters;

[0010] The new global model parameters are sent to the multiple clients so that the multiple clients perform the next round of iterative training.

[0011] Optionally, the aggregating the reference model parameters of multiple clients to obtain new global model parameters is calculated by the following formula:

[0012]

[0013] Among them, W t+1represents the global model parameters of the t+1th round, is the new model parameter, W t represents the global model parameters of the tth round, m represents the number of clients, ΔW i Represents the reference model parameters returned after iterative training and noise perturbation of the i-th client.

[0014] Optionally, sending the new global model parameters to the multiple clients includes:

[0015] Calculate the validation metrics of the current global model;

[0016] Calculate the deviation of validation metrics over multiple rounds;

[0017] When the deviation is greater than a preset threshold, new global model parameters are sent to the multiple clients.

[0018] According to a second aspect of the present invention, an embodiment of the present invention also provides another trajectory mapping method, which is applied to a client and includes:

[0019] Update local model parameters according to the global model parameters received from the server;

[0020] Extracting features from the collected trajectory data and aligning them with the real road network map to obtain a trajectory grid feature map, and inputting the trajectory grid feature map into a model corresponding to the local model parameters for iterative training;

[0021] Calculate the difference between the local model parameters before and after iterative training, and perform pruning operations;

[0022] According to the privacy budget, the size of the data set, and the number of clients participating in the training, the clipped difference is subjected to noise perturbation to obtain reference model parameters, and the reference model parameters are sent to the server for the server to perform aggregate calculations.

[0023] Optionally, the difference between the local model parameters before and after the iterative training is calculated and a clipping operation is performed, which is calculated by the following formula:

[0024]

[0025] in, represents the difference after clipping, represents the difference before clipping, C represents the preset clipping threshold, The Euclidean norm of the difference before clipping.

[0026] Optionally, according to the privacy budget, the size of the data set, and the number of clients participating in the training, the clipped difference is subjected to noise perturbation to obtain reference model parameters, which are calculated using the following formula:

[0027]

[0028] σ=kLΔS / ε

[0029]

[0030] in, represents the reference model parameters, Represents the difference after clipping, N(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution of , k is the privacy constant, C is the preset trimming threshold, ε is the privacy budget, δ is ΔS is the sensitivity, L is the ratio of clients participating in training to the total clients, and D is the size of the training dataset held by the client.

[0031] According to a third aspect of the present invention, an embodiment of the present invention further provides a server, including:

[0032] A sending module, used to randomly send global model parameters to multiple clients;

[0033] A receiving module, used for receiving the reference model parameters after iterative training and noise disturbance of the multiple clients;

[0034] Aggregation module, used to aggregate the reference model parameters of multiple clients to obtain new global model parameters;

[0035] The iteration module is used to send new global model parameters to the multiple clients so that the multiple clients can perform a new round of iterative training.

[0036] According to a fourth aspect of the present invention, an embodiment of the present invention further provides a client, including:

[0037] An updating module, used for updating local model parameters according to the global model parameters received from the server;

[0038] A training module is used to extract features from the collected trajectory data and align it with the real road network map to obtain a trajectory grid feature map, and input the trajectory grid feature map into a model corresponding to the local model parameters for iterative training;

[0039] The trimming module is used to calculate the difference between the local model parameters before and after iterative training and perform trimming operations;

[0040] The noise perturbation module is used to perform noise perturbation on the trimmed difference according to the privacy budget, the data set size and the number of clients participating in the training, obtain reference model parameters, and send the reference model parameters to the server for the server to perform aggregation calculation.

[0041] Optionally, the noise disturbance module calculates the reference model parameters by the following formula:

[0042]

[0043] σ=kLΔS / ε

[0044]

[0045] in, represents the reference model parameters, Represents the difference after clipping, N(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution of , k is the privacy constant, C is the preset trimming threshold, ε is the privacy budget, δ is ΔS is the sensitivity, L is the ratio of clients participating in training to the total clients, and D is the size of the training dataset held by the client.

[0046] According to a fifth aspect of the present invention, an embodiment of the present invention further provides a trajectory mapping system, wherein the trajectory mapping system comprises the server described in the above embodiment and the client described in the above embodiment.

[0047] As described above, a trajectory mapping method, server, client, and system provided by an embodiment of the present invention have the following beneficial effects: randomly sending global model parameters to multiple clients; receiving reference model parameters after iterative training and noise disturbance of the multiple clients; aggregating the reference model parameters of multiple clients to obtain new global model parameters; sending the new global model parameters to the multiple clients so that the multiple clients can perform a new round of iterative training. The present invention performs noise disturbance on the model parameters, and achieves a stronger privacy protection effect by increasing the difference between the disturbed data and the original data. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of a flow chart of a trajectory mapping method provided by an embodiment of the present invention;

[0049] Figure 2 It is a flowchart of an iterative control method provided by an embodiment of the present invention;

[0050] Figure 3 is a schematic flow chart of another trajectory mapping method provided by an embodiment of the present invention;

[0051] Figure 4 It is a structural diagram of a server for executing a trajectory mapping method provided by an embodiment of the present invention;

[0052] Figure 5 It is a structural schematic diagram of a client of an execution trajectory mapping method provided by an embodiment of the present invention;

[0053] Figure 6 It is a structural schematic diagram of a trajectory mapping system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0055] See also Figures 1 to 6 It should be noted that the illustrations provided in this embodiment are only used to illustrate the basic concept of the present invention in a schematic manner, and the illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0056] The embodiment of the present invention can be based on the architecture of interconnected communication between the server and the client, and on the basis of deep learning, adopt federated learning technology, deploy a model for aggregation operation on the server side, and deploy a deep learning model on the client side to process and train trajectory data. The embodiment of the present invention will take the above scenario as an example to introduce in detail the method for the server and the client to perform trajectory mapping. Of course, it should be noted that the model deployed by the server and the client is not limited to the model based on federated learning technology, and any other model for constructing a trajectory map should fall within the scope of protection of the present invention.

[0057] See also Figure 1 , is a flow chart of a trajectory mapping method provided by an embodiment of the present invention, such as Figure 1 As shown, the embodiment of the present invention shows the process of the server executing the trajectory mapping method.

[0058] Step S101: Randomly send global model parameters to multiple clients.

[0059] In an exemplary embodiment, the server can be connected to multiple clients at the same time, that is, the server can send global model parameters to all connected clients at the same time, or send global model parameters to some clients. For the convenience of control, the present invention randomly selects m clients as participants in this round of federated learning. Specifically, the server initializes the global model for trajectory mapping and sends the initialized global model parameters to the client, so that the client performs model training based on the global model parameters and returns the corresponding results.

[0060] Step S102: receiving the reference model parameters after iterative training and noise disturbance of the multiple clients.

[0061] According to the description of the above steps, the server further receives reference model parameters returned by m clients, where the reference model parameters are model parameters generated after each client receives the global model parameters from the server, undergoes local training iterations and adds noise disturbance.

[0062] Step S103: Aggregate the reference model parameters of multiple clients to obtain new global model parameters.

[0063] Furthermore, the server processes the reference model parameters returned from the m clients. In a specific implementation, the reference model parameters are aggregated to obtain new global model parameters, which can be calculated by the following formula:

[0064]

[0065] Among them, W t+1 represents the global model parameters of the t+1th round, is the new model parameter, W t represents the global model parameters of the tth round, m represents the number of clients, ΔW i Represents the reference model parameters returned after iterative training and noise perturbation of the i-th client.

[0066] Step S104: sending the new global model parameters to the multiple clients so that the multiple clients perform the next round of iterative training.

[0067] After the aggregation operation in step S103, the server sends the new global model parameters back to the m clients for the next round of iterative training. It should be noted that the number of training rounds is not limited in the embodiment of the present invention, and any number of rounds can be limited according to the actual trajectory mapping needs.

[0068] In specific implementation, trajectory mapping usually requires multiple rounds of training. In order to ensure mapping efficiency, before executing step S104, a judgment step of terminating training may also be included. Figure 2, is a flow chart of an iterative control method provided by an embodiment of the present invention. As shown in the figure, the server performs the following process to control the number of iterative rounds of the global model:

[0069] Step S1041: Calculate the verification index of the current global model.

[0070] After each round of iteration, after the server receives the corresponding reference model parameters from m clients, it can further calculate the verification indicators of the new global model parameters obtained after aggregation. The verification indicators may include one or more of the average precision, recall rate and F1 score.

[0071] In the specific implementation, the specific definition of the verification indicator is as follows:

[0072] Precision=TP / (TP+FP)

[0073] Recall=TP / (TP+FN)

[0074] F1=(2×Precision×Recall) / (Precision+Recall)

[0075] Among them, Precision represents the average precision, Recall represents the recall rate, TP is the positive sample predicted by the model as the positive class, FN is the negative sample predicted by the model as the negative class, TN is the negative sample predicted by the model as the positive class, and FP is the positive sample predicted by the model as the negative class.

[0076] Step S1042: Calculate the deviation of verification indicators of multiple rounds.

[0077] In order to ensure the calculation accuracy, in the specific implementation, the deviation of the verification indicators of the five rounds can be further calculated, for example, the deviation of the average accuracy, recall rate and F1 score of the five rounds can be calculated. In this way, it can be further observed whether the verification indicators tend to be stable in the five rounds of iteration.

[0078] Step S1043: When the deviation is greater than a preset threshold, the new global model parameters are sent to the multiple clients.

[0079] When the deviation is greater than the preset threshold, it means that the verification index still changes significantly after 5 rounds of iteration, and the global model has not reached convergence, and further iteration is needed, so the new global model is sent to m clients for the next round of iteration. Of course, if the deviation is less than or equal to the threshold, it means that the verification index does not change significantly after 5 rounds of iteration, so it can be considered that the global model has stabilized and reached convergence, and the iteration can be further stopped, and the global model parameters at this time are output as the final result.

[0080] From the description of the above embodiments, it can be seen that in a trajectory mapping method provided by an embodiment of the present invention, a server randomly sends global model parameters to multiple clients; receives reference model parameters after iterative training and noise disturbance of the multiple clients; aggregates the reference model parameters of the multiple clients to obtain new global model parameters; and sends the new global model parameters to the multiple clients so that the multiple clients can perform a new round of iterative training. The present invention performs noise disturbance on the model parameters, and achieves a stronger privacy protection effect by increasing the difference between the disturbed data and the original data.

[0081] See also Figure 3 , is a flow chart of another trajectory mapping method provided by an embodiment of the present invention. As shown in the figure, the process of executing the trajectory mapping method by the client is as follows:

[0082] Step S201: Update local model parameters according to the global model parameters received from the server.

[0083] According to the description of the above-mentioned server execution trajectory mapping method embodiment, the randomly selected m clients receive the global model parameters from the server. The embodiment of the present invention specifically describes the operation steps of the selected clients on the global model parameters.

[0084] The client also deploys a local model, corresponding to the local model parameters. After receiving the global model parameters sent by the server, the client further updates the local model parameters according to the global model parameters to ensure the consistency of the model data information.

[0085] Step S202: extracting features from the collected trajectory data and aligning them with the real road network map to obtain a trajectory grid feature map, and inputting the trajectory grid feature map into a model corresponding to the local model parameters for iterative training.

[0086] The client can also collect trajectory data locally. The trajectory data can be the location coordinates of the road network collected by data collection equipment at ordinary times, which will not be described in detail in the embodiments of the present invention.

[0087] Specifically, the client extracts features from the collected trajectory data, aligns the trajectory grid feature map with the real road network map through segmentation, inputs it into the local model for local training, and uses adaptive moment estimation for local optimization to solve the regularized optimization problem.

[0088] Step S203: Calculate the difference between the local model parameters before and after iterative training, and perform a clipping operation.

[0089] For each local model parameter, the client can calculate the difference between the data before and after the iteration, and perform a trimming operation on the difference.

[0090] In specific implementation, the client can calculate the difference by the following formula:

[0091]

[0092] in, represents the difference after clipping, represents the difference before clipping, C represents the preset clipping threshold, The Euclidean norm of the difference before clipping.

[0093] Step S204: According to the privacy budget, the size of the data set, and the number of clients participating in the training, the trimmed difference is subjected to noise perturbation to obtain reference model parameters, and the reference model parameters are sent to the server for the server to perform aggregate calculations.

[0094] In order to add noise disturbance to the difference, in an exemplary embodiment, the client performs noise disturbance on the difference, and the reference model parameters can be calculated by the following formula:

[0095]

[0096] σ=kLΔS / ε

[0097]

[0098] in, represents the reference model parameters, Represents the difference after clipping, N(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution of , k is the privacy constant, C is the preset trimming threshold, ε is the privacy budget, δ is ΔS is the sensitivity, L is the ratio of clients participating in training to the total clients, and D is the size of the training dataset held by the client.

[0099] The client sends the calculated reference model parameters back to the server, so that the server collects the reference model parameters of all m clients and then performs subsequent operations such as aggregation calculation.

[0100] As can be seen from the description of the above embodiments, an embodiment of the present invention provides a trajectory mapping method, in which the client updates the local model parameters according to the global model parameters sent by the server; extracts the features of the collected trajectory data and aligns it with the real road network map to obtain a trajectory grid feature map, and inputs the trajectory grid feature map into the model corresponding to the local model parameters for iterative training; calculates the difference between the local model parameters before and after the iterative training, and performs a clipping operation; according to the privacy budget, the size of the data set, and the number of clients participating in the training, the clipped difference is noise-perturbed to obtain the reference model parameters, and the reference model parameters are sent to the server for the server to perform aggregate calculations. The present invention performs noise protection on the sent intermediate results, prevents the problem of inferring the original data from the intermediate results, and ensures the privacy of crowdsourcing data; in addition, the present invention solves the problem of achieving a better balance between the privacy budget and the model performance, thereby improving the efficiency of trajectory mapping.

[0101] Through the description of the above method embodiments, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0102] Corresponding to the embodiment of the trajectory mapping method on the server side provided by the present invention, the present invention also provides a server.

[0103] See also Figure 4 , is a schematic diagram of the structure of a server for executing a trajectory mapping method provided by an embodiment of the present invention. As shown in the figure, the server includes:

[0104] A sending module 11 is used to randomly send global model parameters to multiple clients;

[0105] A receiving module 12 is used to receive the reference model parameters after iterative training and noise disturbance of the multiple clients;

[0106] Aggregation module 13, used for aggregating reference model parameters of multiple clients to obtain new global model parameters;

[0107] The iteration module 14 is used to send new global model parameters to the multiple clients so that the multiple clients can perform a new round of iterative training.

[0108] In an exemplary embodiment, the aggregation module 13 aggregates the reference model parameters of multiple clients to obtain new global model parameters, and the formula is as follows:

[0109]

[0110] Among them, W t+1 represents the global model parameters of the t+1th round, is the new model parameter, W t represents the global model parameters of the tth round, m represents the number of clients, ΔW i Represents the reference model parameters returned after iterative training and noise perturbation of the i-th client.

[0111] In an exemplary embodiment, the iteration module 14 can also control the number of iteration rounds, specifically, to calculate the verification index of the current global model; calculate the deviation of the verification index of multiple rounds; when the deviation is greater than a preset threshold, send the new global model parameters to the multiple clients.

[0112] Corresponding to the embodiment of the trajectory mapping method on the client side provided by the present invention, the present invention also provides a client.

[0113] See also Figure 5 , is a schematic diagram of the structure of a client for executing a trajectory mapping method provided by an embodiment of the present invention. As shown in the figure, the client includes:

[0114] An updating module 21, used for updating local model parameters according to the global model parameters received from the server;

[0115] The training module 22 is used to extract features from the collected trajectory data and align it with the real road network map to obtain a trajectory grid feature map, and input the trajectory grid feature map into a model corresponding to the local model parameters for iterative training;

[0116] A clipping module 23, used to calculate the difference between the local model parameters before and after iterative training, and perform clipping operations;

[0117] The noise perturbation module 24 is used to perform noise perturbation on the trimmed difference according to the privacy budget, the data set size and the number of clients participating in the training, obtain reference model parameters, and send the reference model parameters to the server for the server to perform aggregation calculation.

[0118] In an exemplary embodiment, the clipping module 23 calculates the difference between the local model parameters before and after the iteration and performs a clipping operation, which is calculated by the following formula:

[0119]

[0120] in, represents the difference after clipping, represents the difference before clipping, C represents the preset clipping threshold, The Euclidean norm of the difference before clipping.

[0121] In an exemplary embodiment, the noise perturbation module performs noise perturbation on the clipped difference to obtain reference model parameters, which are calculated by the following formula:

[0122]

[0123] σ=kLΔS / ε

[0124]

[0125] in, represents the reference model parameters, Represents the difference after clipping, N(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution of , k is the privacy constant, C is the preset trimming threshold, ε is the privacy budget, δ is ΔS is the sensitivity, L is the ratio of clients participating in training to the total clients, and D is the size of the training dataset held by the client.

[0126] Figure 6 is a schematic diagram of the structure of a trajectory mapping system provided by an embodiment of the present invention, such as Figure 6 As shown, the system includes a server 31 and multiple clients 32, and the client 32 is connected to the server 31 in communication, so as to realize information interaction between the server 31 and the client 32. Specifically, the server 31 executes the steps of executing trajectory mapping on the server side described in the above method embodiment, and the client 32 executes the steps of executing trajectory mapping on the client side described in the above method embodiment, which will not be repeated here.

[0127] The above product can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0128] The electronic device of the embodiment of the present invention exists in various forms, including but not limited to:

[0129] (1) Mobile communication devices: These devices are characterized by their mobile communication functions and their main purpose is to provide voice and data communications. These terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.

[0130] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, such as iPad.

[0131] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players (such as iPods), handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0132] (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0133] (5) Other electronic devices with data interaction functions.

[0134] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0135] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0136] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0137] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A trajectory mapping method, characterized in that: Applied to a server, the method comprises: randomly sending global model parameters to a plurality of clients; Receiving reference model parameters after iterative training and noise disturbance by the multiple clients; Aggregate the reference model parameters of multiple clients to obtain new global model parameters; The new global model parameters are sent to the multiple clients so that the multiple clients perform the next round of iterative training.

2. The trajectory mapping method according to claim 1, characterized in that: The new global model parameters are obtained by aggregating the reference model parameters of multiple clients, and are calculated by the following formula: Among them, W t+1 represents the global model parameters of the t+1th round, is the new model parameter, W t represents the global model parameters of the tth round, m represents the number of clients, ΔW i Represents the reference model parameters returned after iterative training and noise perturbation of the i-th client.

3. The trajectory mapping method according to claim 1, characterized in that: The sending of the new global model parameters to the multiple clients comprises: Calculate the validation metrics of the current global model; Calculate the deviation of validation metrics over multiple rounds; When the deviation is greater than a preset threshold, new global model parameters are sent to the multiple clients.

4. A trajectory mapping method, characterized in that: Applied to a client, the method comprises: Update local model parameters according to the global model parameters received from the server; Extracting features from the collected trajectory data and aligning them with the real road network map to obtain a trajectory grid feature map, and inputting the trajectory grid feature map into a model corresponding to the local model parameters for iterative training; Calculate the difference between the local model parameters before and after iterative training, and perform pruning operations; According to the privacy budget, the size of the data set, and the number of clients participating in the training, the clipped difference is subjected to noise perturbation to obtain reference model parameters, and the reference model parameters are sent to the server for the server to perform aggregate calculations.

5. The trajectory mapping method according to claim 4, characterized in that: The difference between the local model parameters before and after the iterative training is calculated and the clipping operation is performed, which is calculated by the following formula: in, represents the difference after clipping, represents the difference before clipping, C represents the preset clipping threshold, The Euclidean norm of the difference before clipping.

6. The trajectory mapping method according to claim 5, characterized in that: According to the privacy budget, the size of the data set, and the number of clients participating in the training, the clipped difference is perturbed with noise to obtain the reference model parameters, which are calculated using the following formula: σ=kLΔS / ε in, represents the reference model parameters, Represents the difference after clipping, N(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution of , k is the privacy constant, C is the preset trimming threshold, ε is the privacy budget, δ is ΔS is the sensitivity, L is the ratio of clients participating in training to the total clients, and D is the size of the training dataset held by the client.

7. A server, characterized in that: include: A sending module, used to randomly send global model parameters to multiple clients; A receiving module, used for receiving the reference model parameters after iterative training and noise disturbance of the multiple clients; Aggregation module, used to aggregate the reference model parameters of multiple clients to obtain new global model parameters; The iteration module is used to send new global model parameters to the multiple clients so that the multiple clients can perform a new round of iterative training.

8. A client, characterized in that: include: An updating module, used for updating local model parameters according to the global model parameters received from the server; A training module is used to extract features from the collected trajectory data and align it with the real road network map to obtain a trajectory grid feature map, and input the trajectory grid feature map into a model corresponding to the local model parameters for iterative training; The trimming module is used to calculate the difference between the local model parameters before and after iterative training and perform trimming operations; The noise perturbation module is used to perform noise perturbation on the trimmed difference according to the privacy budget, the data set size and the number of clients participating in the training, obtain reference model parameters, and send the reference model parameters to the server for the server to perform aggregation calculation.

9. The client according to claim 8, characterized in that: The noise disturbance module calculates the reference model parameters using the following formula: σ=kLΔS / ε in, represents the reference model parameters, Represents the difference after clipping, N(0,σ 2 ) means the mean is 0 and the variance is σ 2 Gaussian distribution of , k is the privacy constant, C is the preset trimming threshold, ε is the privacy budget, δ is ΔS is the sensitivity, L is the ratio of clients participating in training to the total clients, and D is the size of the training dataset held by the client.

10. A trajectory mapping system, characterized in that: The trajectory mapping system comprises the server as claimed in claim 7 and the client as claimed in claim 8 or 9.

Images