CN117077765A - Electroencephalogram signal identity recognition method based on personalized federal incremental learning - Google Patents

Abstract

The present invention provides an EEG signal identity recognition method based on personalized federated incremental learning, which includes the following steps: first, using a server-client based federated learning architecture, each client initializes a private local model, learning rate network and playback sample pool. Then, at each task increment, each client uses the incremental meta-learning method to continuously learn from the local continuous task stream. Next, the present invention designs a personalized federated incremental learning strategy to achieve learning of heterogeneous data for different incremental tasks in different clients through globally shared model parameters and local retention learning rate networks. Finally, the central server uses the global model trained during the communication process to identify the EEG signal data from different institutions. The personalized federated incremental learning method proposed by the present invention realizes the forward transfer of knowledge between clients in the federated communication process to generate a universal identity recognition template suitable for each terminal.

Description

A method for EEG signal identity recognition based on personalized federated incremental learning

Technical Field

The present invention relates to the field of federated incremental learning for electroencephalogram (EEG) signal identification, and more specifically, to an EEG signal identification method for privacy protection and continuous learning of distributed task incremental scenarios based on personalized federated incremental learning.

Background Art

With the rapid increase in the demand for high reliability of identity authentication systems and the rapid development of artificial intelligence technology, biometric recognition (including face recognition, speaker recognition, iris recognition, EEG signal recognition, etc.) has made significant progress and has been widely used in recent years. While people are constantly generating data, they are also paying more and more attention to privacy and security. Especially when the identity recognition system involves individual privacy data, people require more privacy and security of the system. EEG signals have become a biometric authentication feature with a higher security factor due to their universality, permanence, collectability and uniqueness, but there are still certain limitations. On the one hand, the emergence of deep neural networks has achieved rapid and correct recognition of complex biometric encodings such as EEG, and can obtain higher recognition accuracy and better generalization ability on large-scale data sets. However, with the popularization of hardware devices and the increase in online data streams, the traditional offline training method of deep neural network models has become ineffective. For example, in real-life identity authentication systems, new users register their EEG features every day, and the offline trained model cannot be applied to the newly registered identity data, which brings difficulties to efficient and reliable identity authentication systems. Traditional training uses the method of training the entire data set in a centralized manner, which can obtain an offline model with the best recognition performance and generalization effect. However, when the amount of data becomes larger and larger, this method undoubtedly puts tremendous pressure on storage and computing memory, and is quite time-consuming. At the same time, due to privacy issues or storage limitations, the data of old tasks are usually not fully accessed. In this case, traditional neural networks are trained only based on samples of new tasks each time, which will cause the model to be biased towards new data, thereby reducing the recognition performance of old data, which is called catastrophic forgetting. Therefore, in the face of a continuous stream of incremental biological data, a reliable identity authentication model must demonstrate the ability to continuously learn: that is, the ability to learn continuous tasks without forgetting to recognize tasks that have been trained in the past. On the other hand, real-life data comes from various distributed edge devices. Centralizing training with huge data sets can improve the accuracy and robustness of the model, but this places high demands on the storage and computing capabilities of cloud devices, and may also bring privacy risks and data leakage threats to edge devices. Each distributed organization trains its own data separately, which can protect the data privacy of each edge device, but will greatly reduce the accuracy and generalization of each distributed model.

In real scenarios, it is necessary to consider both the data increment problem and the privacy problem. The task data of each distributed authentication device is continuously coming, which undoubtedly brings difficulties to EEG signal identification. On the one hand, every time a new task stream arrives, each client needs to retrain a new model based on all training samples to apply to all tasks, which will lead to memory usage and time waste; especially when each edge device is lightweight, the continuous task stream will increase the storage pressure of each device. On the other hand, when the data between the clients are not independent and identically distributed, if only the current latest task data is used for training, the model gradient will be biased towards the new task, resulting in catastrophic forgetting of the old tasks in the past. In summary, designing an EEG signal identification method suitable for distributed task increment scenarios has guiding significance for biometric authentication systems in real scenarios.

Summary of the invention

To solve the above problems, the present invention provides an EEG signal identity recognition method based on personalized federated incremental learning, constructs a personalized federated incremental learning framework based on privacy protection, realizes personalized incremental learning by globally sharing meta-parameters and locally retaining learning rates, and each client adopts an adaptive incremental meta-learning method based on sample playback and task distillation for local updates, thereby achieving adaptive learning of different tasks while avoiding forgetting old tasks.

The use process of the EEG signal identity recognition method based on personalized federated incremental learning provided by the present invention includes the following steps:

According to the method of EEG signal identification based on personalized federated incremental learning according to claim 1, it is characterized in that in step S101: a server-client based federated learning architecture is adopted to evenly distribute the motor imagery EEG signals collected by each incremental task to all K clients (federal communication involves a total of K clients), and each client k (1≤k≤K) initializes a private local model Learning Rate Network and playback sample pool

First, the server-client architecture used is a federated learning framework with the server as the center and the client as the distributed node. Each client is trained only based on local private data samples, and does not share data with other clients or servers, but shares model parameters with the server.

Then, according to different data distribution scenarios, the collected motor imagery EEG signals are evenly distributed to all clients. For EEG signal data with M categories and n samples in each category, under the IID (independent and identically distributed) setting, the training samples of all classes are evenly distributed on all clients, and each client will process data with the same category but non-overlapping samples: each client k processes M sample categories and n samples. In the non-IID setting, all categories are evenly distributed across all clients, and the categories processed by each client are disjoint: the number of sample categories processed by each client k is The number of samples is

Finally, each client k is initialized, including the model parameters of the local meta-learning model Model parameters of the local learning rate network And a playback sample pool of a certain capacity

According to the method for EEG signal identification based on personalized federated incremental learning of claim 1, it is characterized in that in step S102: for the kth client (1≤k≤K, the federated communication involves a total of K clients) described in step S101, set its private task flow When the cth task communicates (1≤c≤C, there are C tasks communicating in the federated process), enter the current incremental task Sample data in, based on local models and the current incremental task Conduct incremental learning;

First, p and i represent the number of external and internal loops, respectively. represents the model parameters obtained by performing i-time internal training in the p-th external cycle of the k-th client;

Then, in the inner loop of incremental meta-learning, meta-learning is combined with the recent class mean sample replay; based on the current incremental task The data of the local playback sample pool of each client k is replayed using a learnable learning rate network for meta-learning of multi-task small sample scenarios during the c-th task communication (2≤c≤C). Perform sample sampling to obtain the sampled old task samples Based on old task samples Calculate distillation loss; for the current incremental task Perform sample sampling to obtain a new task sample b = {(X ^c , Y ^c ), 2≤c≤C}, based on the new task sample b and the old task sample Calculate the classification loss; so the goal of meta-learning internal training is

Among them, X1 ^:c and Y1 ^:c represent the training data and corresponding labels processed by the client during the first c task communications, ^Xc and ^Yc represent the training data and corresponding labels processed by the client during the cth task communication, λ represents the relative weight of classification loss and distillation loss, _lmeta represents the loss function of the meta task, It means the input is X ^1:c and the model parameters are The output prediction result of , l _CE and l _KD represent the classification loss function and distillation loss function respectively; at the same time, the playback sample pool is updated according to the nearest class mean sample playback rule;

Finally, in the outer loop of incremental meta-learning, the learning rate network and the parameters of the meta-model are updated by gradient descent according to the meta-loss to find the learning rate and direction suitable for different tasks; for the next outer loop p+1 of k clients, the meta-parameters Gradient update is performed by the meta-loss based on classification loss and distillation loss:

Where β represents the learning rate parameter for updating the meta-model parameters; for the next external cycle p+1 of the k-th client, the meta-loss is based on the learning rate network of the p-th external cycle Calculate the gradient and adaptively update the learning rate network:

Among them, α _hyperlr represents the learning rate parameter updated by the learning rate network, is the learnable learning rate network for the k-th client p-th external loop, the architecture and metamodel of this network The architecture is the same.

According to the method for EEG signal identification based on personalized federated incremental learning of claim 1, it is characterized in that in step S103: the personalized federated incremental learning framework is composed of multiple clients described in step S102 and a central server, and after each client completes the local update of the current task, the learning rate network and the playback sample pool are locally retained, and the parameters of the meta-model are globally shared to achieve communication with the server and retain the local personalized learning direction;

First, the central server aggregates the local metamodel parameters from all clients The amount of data processed by each client in the current round is d ₁ ,…,d _k ,…,d _K . Then, the aggregated model parameters are federated and weighted averaged according to the amount of data to obtain the global model parameters. Finally, the central server distributes the global model parameters Θ to all clients.

According to the method for EEG signal identity recognition based on personalized federated incremental learning according to claim 1, it is characterized in that in step S104: the personalized federated incremental learning method designed in S103 is iteratively trained until R communication rounds to ensure model convergence; the personalized federated incremental learning method is used to perform distributed incremental learning on the EEG data sample to be identified to determine the user label corresponding to the EEG data sample, and the specific process is that the central server identifies the input test EEG signal x based on the trained global parameter Θ, and the obtained label y is the predicted identity information.

The beneficial effects of the present invention are: the EEG signal identification method based on personalized federated incremental learning of the present invention constructs a federated learning framework based on a server-client architecture, in which each client adopts an incremental meta-learning method to perform local incremental learning; then, a personalized federal communication strategy for globally sharing model parameters and locally retaining learning rates is designed, and each client is set to communicate only its meta-learning parameters with the server instead of communicating a private learning rate network. On the one hand, it can strengthen the protection of the local model learning method, and on the other hand, it can realize personalized learning of different optimizer learning rates of different clients. At the same time, the local privatization of the learning rate network also reduces the communication cost of transmission.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of the structure of an EEG signal identity recognition method based on personalized federated incremental learning according to an embodiment of the present invention.

FIG2 is a schematic diagram of the architecture of an incremental learning method based on adaptive meta-learning in a client according to an embodiment of the present invention.

FIG3 is a schematic diagram of calculating the distillation loss and the classification loss in the meta-task of an embodiment of the present invention.

FIG4 is a schematic diagram of the updating process of the inner loop and the outer loop of the incremental meta-learning method according to an embodiment of the present invention.

FIG5 is a diagram showing the overall framework of EEG signal identity recognition using personalized federated incremental learning according to an embodiment of the present invention.

DETAILED DESCRIPTION

The present invention is described in detail below with reference to the accompanying drawings and specific embodiments: The method of the present invention is divided into four parts.

Part 1: Architecture Construction of Federated Learning

Part 2: Client local updates based on incremental meta-learning

Part III: Personalized communication between server and client

Part 4: Server performs EEG signal identification

According to these four parts, the EEG signal identity recognition method based on personalized federated incremental learning in the embodiment of the present invention, as shown in FIG1 , includes the following steps:

S101: Adopting a server-client based federated learning architecture, the motor imagery EEG signals collected by each incremental task are evenly distributed to all K clients (federated communication involves a total of K clients), and each client k (1≤k≤K) initializes a private local model Learning Rate Network and playback sample pool

First, the server-client architecture adopted is a federated learning framework with the server as the center and the client as a distributed node. Each client is trained only based on local private data samples, and does not share data with other clients or servers, but shares model parameters with the server.

Then, according to different data distribution scenarios, the collected motor imagery EEG signals are evenly distributed to all clients. For EEG signal data with M categories and n samples in each category, under the IID (independent and identically distributed) setting, the training samples of all classes are evenly distributed on all clients, and each client will process data with the same category but non-overlapping samples: each client k processes M sample categories and n samples. In the non-IID setting, all categories are evenly distributed across all clients, and the categories processed by each client are disjoint: the number of samples processed by each client k is The number of sample categories is

Finally, each client k is initialized, including the model parameters of the local meta-learning model Model parameters of the local learning rate network And a playback sample pool of a certain capacity

S102: At each task increment, each client initialized in step S101 performs incremental learning based on the local model and the current private incremental task, which is mainly achieved by the following three steps: 1) combining the inner loop of meta-learning with the playback of the nearest class mean sample; 2) meta-task distillation; 3) meta-learning outer loop based on a learnable learning rate network;

FIG2 is a schematic diagram of the architecture of the incremental learning method based on adaptive meta-learning local to each client.

The first step combines the inner loop of meta-learning with the replay of the nearest class mean sample. For the kth client (1≤k≤K, federated communication involves a total of K clients), the private task flow When the cth task communicates (1≤c≤C, there are C tasks communicating in the federated process), the private task processed by client k is During the cth task communication (1≤c≤C), each client k (1≤k≤K) based on its private task The local incremental model is trained with samples from Sampled from private dataset And includes a total Assuming that the sample space has N category labels, then Private Datasets middle represents sample data, It should be noted that for client k, the task sample categories processed by different task communication rounds do not overlap, that is, when i≠j For different clients p and q, the task sample categories processed by the same task communication round may overlap, that is, when p≠q

The incremental meta-learning training process can be divided into internal training and external training. For time c, this method is based on the current new task The new task sample b is sampled for internal training, based on b and from the sample pool Sample of old tasks Therefore, this method can ensure the alignment of gradients of new and old tasks while learning new data distribution.

First, in the inner loop of incremental meta-learning, based on the data of the current new task, a learnable learning rate network is used to quickly adapt to the multi-task small sample scenario. p and i represent the number of outer and inner loops, respectively. It represents the model parameters obtained by performing i internal trainings in the p-th external loop of the k-th client.

0033.

Then, let b _m = m _c-1 ∪b be the number of samples from the sample pool. Sample of old tasks Mixed data of new task sample b sampled by the current task, based on the model parameters obtained by internal training The local meta-parameters Θ ^c are obtained by external meta-training with the mixed data b _m . Therefore, at time c, the objective of formula (1) can be rewritten as:

Among them, {X ^1:c ,Y ^1:c } is the sample data of 1 to c tasks currently processed obtained by sampling from b _m .

The recent class mean sample replay method uses a fixed-capacity replay sample pool, which stores the sample data of the recent class mean in the old task for retrieval during the new task training process. is the playback sample pool obtained by sampling the nearest class mean in the cth round of communication, and is updated to after the cth round of communication The nearest class mean sample replay method includes two main stages: sample sampling and sample pool update. The sample sampling stage constructs a limited size replay sample m (the process is shown in Algorithm 1); the sample pool update stage reconstructs a fixed capacity replay sample pool at time c (The process is shown in Algorithm 2).

The second step is meta-task distillation: As mentioned above, the meta-task sampling _bm is composed of the current task batch b and the sample pool Sample playback sample Mixed composition. The present invention does not need to store the network parameters θ ^c corresponding to each task, but the prototype of the nearest class mean sample in the playback sample pool (the feature embedding extracted before the fully connected layer of the corresponding task) as a soft label. For each old task τ, the soft label in the playback sample pool is generated by the classifier θ ^{τ when the task τ} is just trained, so as to ensure that θ ^τ can learn the sample distribution of the task τ most accurately.

As shown in Figure 3, this is a method for calculating the distillation loss of old samples and the classification loss of new and old samples in the meta-task. When calculating the distillation loss, let the playback sample The number of class labels is |m|. Sample data in (where label Let the output of the previous old classifier and the output of the current new classifier be and O ^|m| (x)＝[ ^o1 (x),…,o ^|m| (x)]. Then the distillation loss can be expressed as:

in T represents the temperature scale. When calculating the classification loss, let the number of category labels of the new task samples be |n|. Mixed data of |m| replay samples in b and |n| samples in current task b (where label Let the output of the current new classifier be O ^|m|+|n| (x)=[ ^o1 (x),…,o ^|m| (x),o ^|m|+1 (x),…,o ^|m|+|n| (x)]. Then the cross entropy classification loss can be expressed as:

in, If the true category y of the sample is k, it takes 1, otherwise it takes 0. p _k (x) represents the probability of the classifier outputting the kth category (such as the logits value of the softmax layer).

Therefore, formula (4) can be rewritten as:

Among them, l _CE is the cross entropy loss function for correct classification on b _m , l _KD is The above distillation loss function is used for network regularization. The meta-parameters of the next outer cycle p+1 are updated by the gradient of the meta-loss based on the classification loss and the distillation loss:

Among them, β represents the learning rate parameter for meta-model parameter update.

The third step is to use a meta-learning outer loop based on a learnable learning rate network: At the p+1th outer loop of the kth client, the meta-loss is based on the learning rate network of the pth outer loop. Calculate the gradient and adaptively update the learning rate network:

in, is the learnable learning rate network for the pth outer cycle, the architecture and metamodel of this network The architecture is the same.

Figure 4 depicts and The update process, and All are updated by gradient descent, where c represents the number of tasks being processed, and i and p represent the number of inner and outer loops, respectively. The purpose is to adaptively adjust the learning rate and direction of model updates while reducing dependence on learning rate initialization.

S103: The client updated in step S102 retains the local rate network and playback sample pool, and shares the local meta-learning model parameters with the server; the central server obtains the global model by aggregating the shared model parameters of the client updated in step S102; the central server distributes the global model obtained by federation averaging to all clients;

As shown in Figure 5, each client k maintains its own private playback sample pool locally. and the learnable learning rate network parameter The central server communicates its meta-learning model parameters Θ _k with the central server; the central server weights and aggregates the meta-learning model parameters uploaded by each participating client and distributes them to each client as global model parameters. Unlike traditional federated systems that share the optimizer's learning rate during communication or reinitialize the optimizer's learning rate during each round of communication, in the framework constructed in this chapter, each client k only initializes the learning rate network at the first task flow. During subsequent federated communication, task flows are maintained and trained as they arrive. To achieve local personalized learning. This method allows the client to learn the local data distribution in a personalized way and realizes the forward transfer of knowledge between clients during federated communication, while reducing the communication cost. Algorithm 4 describes the specific process of the framework. The personalized framework designed enables the model parameters of meta-learning to be shared globally between clients, and each client retains the adaptive learning rate network parameters locally.

S104: The server performs identity recognition of the EEG signal based on the global model after federated averaging obtained in step S103; uses the personalized federated incremental learning method to perform distributed incremental learning on the EEG data sample to be recognized, and determines the user label corresponding to the EEG data sample. The specific process is that the central server recognizes the input test EEG signal x based on the trained global parameter Θ, and the obtained label y is the predicted identity information.

Experimental design

The experimental object of the training model of the present invention is a large-scale standard EEG Motor Movement/Imagery Dataset. In this embodiment, the data set contains more than 1,500 1-2 minute EEG signal records, and the EEG signal samples come from 109 healthy subjects with a sampling frequency of 160Hz; each subject performs a different movement/imagination task, and uses the BCI2000 system to record 64-channel EEG signals; each subject performs 14 experiments: 2 1-minute baseline exercises (the first time with eyes open, the second time with eyes closed), and 3 2-minute exercises for each of the following 4 tasks:

Task1: The target appears on the left or right side of the screen. The subject opens and closes the corresponding fist until the target disappears. Then the subject relaxes.

Task2: The target appears on the left or right side of the screen. The subject imagines opening and closing the corresponding fist until the target disappears. The subject then relaxes.

Task 3: The target appears at the top or bottom of the screen. The subject opens or closes both fists (if the target is at the top) or both feet (if the target is at the bottom) until the target disappears. The subject then relaxes.

Task 4: The target appears at the top or bottom of the screen. The subject imagines opening or closing two fists (if the target is at the top) or two feet (if the target is at the bottom) until the target disappears. The subject then relaxes.

The abbreviation for open eyes is EO (Eye Open), closed eyes is EC (Eye Close), movement state is PHY (Physical), and imagined movement state is IMA (Image).

The present invention uses inter-task datasets to train the identity recognition model: the EO and EC resting state data are used as training and testing. In the federated incremental scenario, the number of clients is set to 5 and 10, the communication rounds are set to 1, 2, 5, 10, and 20, and the local playback sample pool size of each client is 109. Experiments are designed under IID settings and non-IID settings respectively:

IID setting, that is, independent and identically distributed setting. The training samples of all classes are evenly distributed on all clients, and each client will process the same category. Specifically, each client will process the task sequence with the same category label as other clients in a different order, and different clients process different samples of different categories. In this setting, the dataset is divided into 11 tasks consisting of 10 non-overlapping incremental categories (the last task contains 9 classes). When each task is incremented, the samples of all incremental categories of the task are evenly distributed to all clients;

Non-IID setting: that is, non-IID setting. All categories are evenly distributed among all clients, and the categories processed by each client are disjoint. Specifically, each client processes a task sequence containing different category labels, and the categories processed by different clients are different. In this setting, when there are 5 clients in the federated framework, 109 classes are divided into 9 tasks with 5 increments and 16 tasks with 4 increments; when there are 10 clients, 109 classes are divided into 11 tasks with 3 increments and 19 tasks with 4 increments.

Experimental Results

Table 1 and Table 2 show the global performance of the server and the local performance of the client under 5 clients, 10 clients and 1, 2, 5, 10, and 20 rounds of communication respectively under IID and non-IID settings. From the perspective of data distribution, the proposed method of the present invention is effective for federated incremental scenarios under IID and non-IID settings; in particular, the performance under the non-IID setting is better when the number of communication rounds is more, which shows that when the data is not independent and identically distributed, the method proposed in this chapter can enable each client to learn more migration knowledge from other clients in more communication rounds. From the perspective of the number of clients and communication rounds, the present invention is effective under different parameter settings, and multi-scenario experiments can verify the generalization and robustness of the present invention.

Table 1. IID setup: federated incremental performance under different clients and different communication rounds. Evaluated by global performance (%) on the server and local performance (%) on the client. The average of 3 random seeds is taken for each experiment.

Table 2. Non-IID setting: federated incremental performance under different clients and different communication rounds. Evaluated by global performance (%) on the server and local performance (%) on the client. The average of 3 random seeds is taken for each experiment.

Claims (5)

1. A method for EEG signal identity recognition based on personalized federated incremental learning, characterized by comprising the following steps: S101: Adopting a server-client based federated learning architecture, the motor imagery EEG signals collected by each incremental task are evenly distributed to all K clients (federated communication involves a total of K clients), and each client k (1≤k≤K) initializes a private local model Learning Rate Network and playback sample pool S102: At each task increment, each client initialized in step S101 performs incremental learning based on the local model and the current private incremental task; S103: The client updated in step S102 retains the local rate network and playback sample pool, and shares the local meta-learning model parameters with the server; the central server obtains the global model by aggregating the shared model parameters of the client updated in step S102; the central server distributes the global model obtained by federation averaging to all clients; S104: The server performs identity recognition of the EEG signal based on the global model after federated averaging obtained in step S103. 2. According to the method of personalized federated incremental learning based on EEG signal identification in claim 1, it is characterized in that in step S101: adopting a server-client based federated learning architecture, the motor imagery EEG signals collected by each incremental task are evenly distributed to all K clients (federated communication involves a total of K clients), and each client k (1≤k≤K) initializes a private local model Learning Rate Network and playback sample pool First, the server-client architecture used is a federated learning framework with the server as the center and the client as a distributed node. Each client is trained only based on local private data samples, and does not share data with other clients or servers, but shares model parameters with the server. Then, according to different data distribution scenarios, the collected motor imagery EEG signals are evenly distributed to all clients. For EEG signal data with M categories and n samples in each category, under the IID (independent and identically distributed) setting, the training samples of all classes are evenly distributed on all clients, and each client will process data with the same category but non-overlapping samples: each client k processes M sample categories and n samples. In the non-IID setting, all categories are evenly distributed across all clients, and the categories processed by each client are disjoint: the number of sample categories processed by each client k is The number of samples is Finally, each client k is initialized, including the model parameters of the local meta-learning model Model parameters of the local learning rate network And a playback sample pool of a certain capacity 3. According to the method of personalized federated incremental learning based on EEG signal identification in claim 1, it is characterized in that in step S102: for the kth client (1≤k≤K, the federated communication involves a total of K clients) described in step S101, set its private task flow When the cth task communicates (1≤c≤C, there are C tasks communicating in the federated process), enter the current incremental task Sample data in, based on local models and the current incremental task Conduct incremental learning; First, p and i represent the number of external and internal loops, respectively. represents the model parameters obtained by performing i-time internal training in the p-th external cycle of the k-th client; Then, in the inner loop of incremental meta-learning, meta-learning is combined with the recent class mean sample replay; based on the current incremental task The data of the local playback sample pool of each client k is replayed using a learnable learning rate network for meta-learning of multi-task small sample scenarios during the c-th task communication (2≤c≤C). Perform sample sampling to obtain the sampled old task samples Based on old task samples Calculate distillation loss; for the current incremental task Perform sample sampling to obtain a new task sample b = {(X ^c , Y ^c ), 2≤c≤C}, based on the new task sample b and the old task sample Calculate the classification loss; so the goal of meta-learning internal training is Among them, X1 ^:c and Y1 ^:c represent the training data and corresponding labels processed by the client during the first c task communications, ^Xc and ^Yc represent the training data and corresponding labels processed by the client during the cth task communication, λ represents the relative weight of classification loss and distillation loss, _lmeta represents the loss function of the meta task, It means the input is X1 ^:c and the model parameters are The output prediction result of , l _CE and l _KD represent the classification loss function and distillation loss function respectively; at the same time, the playback sample pool is updated according to the nearest class mean sample playback rule; Finally, in the outer loop of incremental meta-learning, the learning rate network and the parameters of the meta-model are updated by gradient descent according to the meta-loss to find the learning rate and direction suitable for different tasks; for the next outer loop p+1 of k clients, the meta-parameters Gradient update is performed by the meta-loss based on classification loss and distillation loss: Where β represents the learning rate parameter for updating the meta-model parameters; for the next external cycle p+1 of the k-th client, the meta-loss is based on the learning rate network of the p-th external cycle Calculate the gradient and adaptively update the learning rate network: Among them, α _hyperlr represents the learning rate parameter updated by the learning rate network, is the learnable learning rate network for the k-th client p-th external loop, the architecture and metamodel of this network The architecture is the same. 4. According to claim 1, a method for EEG signal identification based on personalized federated incremental learning is characterized in that in step S103: the personalized federated incremental learning framework is composed of multiple clients described in step S102 and a central server. After each client completes the local update of the current task, it locally retains the learning rate network and the playback sample pool, and globally shares the parameters of the meta-model to achieve communication with the server and retain the local personalized learning direction; First, the central server aggregates the local metamodel parameters from all clients The amount of data processed by each client in the current round is d ₁ , ..., d _k , ..., d _K . Then, the aggregated model parameters are federated and weighted averaged according to the amount of data to obtain the global model parameters. Finally, the central server distributes the global model parameters Θ to all clients. 5. According to claim 1, a method for EEG signal identity recognition based on personalized federated incremental learning is characterized in that in step S104: the personalized federated incremental learning method designed in S103 is iteratively trained until R communication rounds to ensure model convergence; the personalized federated incremental learning method is used to perform distributed incremental learning on the EEG data sample to be identified to determine the user label corresponding to the EEG data sample, and the specific process is that the central server identifies the input test EEG signal x based on the trained global parameter Θ, and the obtained label y is the predicted identity information.