Blockchain-enabled trusted management system and method for aigc service life cycle

Abstract

The application provides a kind of blockchain empowerment AIGC service life cycle's credible management system and method, the system includes: cloud edge collaborative blockchain, including as full node's cloud server and as light node's deployment with AIGC model's edge server;AIGC model verification module is used for embedding model watermark for each AIGC model when pre-training starts, realizes AIGC model verification by extracting and comparing the model watermark contained in the training log of on-chain consensus with the training log stored in off-chain database;AIGC service provider selection module is used for selecting AIGC service provider for user;AIGC product transaction module is used for establishing state channel through cloud edge collaborative blockchain when transaction starts between AIGC service provider and user, and the state channel is used to realize the atomic transaction between AIGC service provider and user.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence content generation technology, and in particular to a trusted management system and method for enabling the lifecycle of AIGC services through blockchain. Background Technology

[0002] Artificial Intelligence Generated Content (AIGC) technology automates and customizes content using artificial intelligence algorithms. Compared to traditional Professionally Generated Content (PGC) and User-Generated Content (UGC), AIGC can quickly and personalizedly generate various forms of content, including text, images, videos, and audio. In recent years, AIGC has received widespread attention and application from academia and industry, experiencing rapid development and being widely used in fields such as gaming and video processing. AIGC model training requires substantial data, computational resources, and significant training time.

[0003] Figure 2 This is a lifecycle diagram of the existing cloud-edge collaborative AIGC service. Centralized cloud data center servers possess powerful computing capabilities; therefore, cloud servers (CS) are primarily used for AIGC model pre-training and fine-tuning. A cloud server (CS) can include a GAN network composed of a generator and discriminator, as well as a diffusion model. The CS, acting as an AIGC model provider, delivers the trained model to edge devices in the edge network. Due to the heterogeneity of model types, computing power, and reliability metrics, the edge network is used for edge-layer AIGC inference services. Deploying AIGC models on edge devices as AIGC service providers (ASPs) allows for the distribution of services to user devices closer to the edge layer, providing personalized, low-latency, and flexible services. Specifically, the entire AIGC service lifecycle includes: ① The cloud server provides the model to the AIGC service provider; ② Selecting an AIGC service provider (ASP), which, simply put, means choosing a suitable and reliable ASP for the user; ③ Transactions between user devices based on the selected ASP.

[0004] Current research largely focuses on improving the quality of AIGC services, but neglects how to ensure the reliability of the AIGC service lifecycle while ensuring service quality. For example... Figure 1As shown, AIGC services face trust challenges in the lifecycle of model provision, service provider selection, and product transactions. These challenges include: (1) Inconsistency in AIGC model provision. The deployment models provided by cloud servers to edge devices may deviate from the pre-agreed terms, such as differences in training duration and model type. This directly affects the service quality of the model and leads to inconsistencies between cost and performance; (2) Difficulty in ensuring the reliability of AIGC inference services. Edge servers exhibit heterogeneity in computing resources, model configuration, and service quality, making it difficult for users to select suitable and reliable AIGC service providers; (3) Lack of atomic transactions in AIGC products. There are two-way default risks in AIGC product and fee transactions between AIGC service providers and users, including the risk of mismatch between fees and resource costs under the pre-payment model and the risk of payment default under the post-payment model.

[0005] Blockchain, as a distributed ledger, offers a viable solution to these problems. It establishes trust among participants by maintaining the immutability and traceability of data. Therefore, blockchain can be used to establish a trusted mechanism for AIGC services between the cloud (cloud servers), edge (edge ​​networks), and users. However, traditional blockchains struggle to efficiently handle AIGC service requests from a large number of users.

[0006] Currently, attempts to use blockchain to address the trust issues of AIGC services mainly focus on solving trust or security problems in the process of generating digital content and applications. There is a lack of means to address the trust challenges that exist in the lifecycle of AIGC services in model provision, service provider selection, and product transaction. There is also a lack of construction of a trustworthy operation process throughout the entire lifecycle of AIGC services. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a trusted management system and method for AIGC service lifecycle empowered by blockchain, in order to eliminate or improve one or more defects existing in the prior art.

[0008] One aspect of the present invention provides a blockchain-enabled trusted management system for the AIGC service lifecycle, the system comprising:

[0009] The cloud-edge collaborative blockchain comprises cloud servers as full nodes and edge servers deployed with AIGC models as light nodes, where the edge servers with AIGC models act as AIGC service providers. The light nodes are used to reach consensus on AIGC product transactions between AIGC service providers and users, while the full nodes are used to reach consensus on training logs and the reputation of AIGC service providers. An AIGC model verification module is used to embed a model watermark for each AIGC model at the start of pre-training. The encoded vector of the model watermark is included in the training log of the corresponding AIGC model. During the training process of the AIGC model, the hashed training logs are stored on the cloud-edge collaborative blockchain, and the encrypted training logs are also stored in an off-chain database. The AIGC model verification module is used to extract and compare the training logs from the on-chain consensus. The AIGC model is validated using model watermarks contained in the training logs stored in the off-chain database. The model watermark includes a unique identifier for each AIGC model. An AIGC service provider selection module is used to select AIGC service providers for users. The AIGC service provider reputation construction based on full-node consensus considers a reward function that takes into account the reputation factor of the AIGC service provider. This reward function is used to select a multi-objective optimization problem model for the AIGC service provider. An AIGC product transaction module is used to establish a state channel through a cloud-edge collaborative blockchain when a transaction between the AIGC service provider and the user begins. Based on light node consensus, the AIGC product transaction locks encrypted AIGC products from the AIGC service provider and encrypted service fees from the user, using the state channel to achieve atomic transactions between the AIGC service provider and the user.

[0010] In some embodiments of the present invention, the cloud-edge collaborative blockchain is used to store relevant data generated throughout the entire lifecycle of the AIGC service, and the consensus form of full nodes and light nodes is a DAG ledger; the cloud-edge collaborative blockchain uses on-chain smart contracts to execute on-chain operations, and the off-chain database is the InterPlanetary File System (IPFS).

[0011] In some embodiments of the present invention, the model watermark further includes the version number of the AIGC model, a timestamp, and a randomly generated portion. The version number is used to verify the version of the AIGC model, the timestamp is used to calculate the training duration of the AIGC model, and the randomly generated portion is used to assist in verifying the AIGC model.

[0012] In some embodiments of the present invention, the reward function also considers user utility factors and energy consumption factors, and updates the reputation of AIGC service providers on the blockchain in real time in the form of a reputation table; the reward function is used for a multi-objective optimization problem model for selecting AIGC service providers for users, including: constructing a multi-objective optimization problem model for selecting AIGC service providers for users, using a preset AIGC service provider selection algorithm to determine the AIGC service provider selection method that maximizes the reward function, and using the AIGC service provider selection module to select the AIGC service provider for the user.

[0013] In some embodiments of the present invention, the reward function considers user utility factors by calculating the total user utility value; wherein, the total user utility value is calculated by combining the AIGC service latency and the user-perceived content quality evaluation function; and the reward function considers energy consumption factors by calculating the total energy consumption of all AIGC service providers.

[0014] In some embodiments of the present invention, the AIGC service provider selection algorithm includes a global network and multiple local networks. The trajectories generated by the interaction between each local network and the environment are asynchronously stored in the experience replay pool of the global network. Each local network updates the policy gradient and value gradient using an A2C network and passes them to the global network. The global network uses a stable diffusion model to generate the optimal decision for AIGC service provider selection based on the asynchronous trajectories, policy gradients, and value gradients from each local network, thereby maximizing the reward function.

[0015] Corresponding to the aforementioned trusted management system, this invention also provides a blockchain-enabled trusted management method for the AIGC service lifecycle, comprising: an AIGC model verification module embedding a model watermark for each AIGC model at the start of pre-training; the encoded vector of the model watermark being included in the training log of the corresponding AIGC model; during the training process of the AIGC model, on the one hand, storing the hashed training log on the cloud-edge collaborative blockchain, and on the other hand, encrypting and storing the training log in an off-chain database; and verifying the AIGC model by extracting and comparing the model watermark contained in the on-chain consensus training log with the training log stored in the off-chain database; wherein, the model watermark includes a unique identifier for each AIGC model; and an AIGC service provider selection module selecting an AIGC service provider for the user; wherein, the AIGC service provider reputation construction based on full-node consensus considers the AIGC service... The AIGC product transaction module establishes a state channel through a cloud-edge collaborative blockchain at the start of a transaction between the AIGC service provider and the user. Based on light node consensus, the AIGC product transaction locks encrypted AIGC products from the AIGC service provider and encrypted service fees from the user, achieving atomic transactions between the AIGC service provider and the user using the state channel. The nodes of the cloud-edge collaborative blockchain include cloud servers as full nodes and edge servers deployed with AIGC models as light nodes. The edge servers with AIGC models act as AIGC service providers. The light nodes are used to reach consensus on AIGC product transactions between the AIGC service provider and the user, while the full nodes are used to reach consensus on training logs and the reputation of the AIGC service provider.

[0016] Corresponding to the aforementioned trusted management system, the present invention also provides an electronic device, including a processor, a memory, and a computer program / instructions stored in the memory, wherein the processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the electronic device implements the functions of any of the trusted management systems described in the above embodiments.

[0017] Corresponding to the aforementioned trusted management system, the present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the functions of the trusted management system as described in any of the above embodiments.

[0018] Corresponding to the aforementioned trusted management system, the present invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the functions of the trusted management system as described in any of the above embodiments.

[0019] The blockchain-enabled trusted management system and method for the AIGC service lifecycle proposed in this invention can empower the entire lifecycle of AIGC services using cloud-edge collaborative blockchain. During the AIGC model training phase, the AIGC model verification module verifies the AIGC model through embedded model watermarks. When selecting an AIGC service provider for a user, the AIGC service provider selection module makes the optimal choice based on the reputation of the AIGC service provider through full-node consensus. During the AIGC product transaction process, atomic transactions between the AIGC service provider and the user are realized through the state channel established by the cloud-edge collaborative blockchain. This helps to solve the trust challenges existing in the lifecycle of AIGC services in model provision, service provider selection, and product transaction, and facilitates the construction of a trusted operating process throughout the entire lifecycle of AIGC services.

[0020] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0021] Those skilled in the art will understand that the objectives and advantages achievable with this invention are not limited to those specifically described above, and that the above and other objectives achievable with this invention will become clearer from the following detailed description. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings:

[0023] Figure 1 This is an architecture diagram of a trusted management system for the lifecycle of AIGC services empowered by blockchain, as shown in one embodiment of the present invention.

[0024] Figure 2 This is a lifecycle diagram of the existing cloud-edge collaborative AIGC service.

[0025] Figure 3 This is a diagram of a trusted management system architecture for the lifecycle of AIGC services empowered by blockchain, as shown in another embodiment of the present invention.

[0026] Figure 4 This is a flowchart of the AIGC product transaction module in one embodiment of the present invention.

[0027] Figure 5 This is a diagram of the DMA3C algorithm architecture for AIGC service provider selection in one embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0029] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0030] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0031] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0032] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0033] To address the trust challenges in the AIGC service lifecycle, including model provision, service provider selection, and product transaction, the following issues need to be addressed: (1) Inconsistency risks exist in AIGC model provision. The deployment models provided by cloud servers to edge devices may deviate from pre-agreed standards, such as differences in training duration and model type. This directly affects the service quality of the models and leads to inconsistencies between cost and performance; (2) The reliability of AIGC inference services is difficult to guarantee. Edge servers exhibit heterogeneity in computing resources, model configuration, and service quality, making it difficult for users to select suitable and reliable AIGC service providers; (3) AIGC products lack atomic transactions. There are two-way default risks in AIGC product and fee transactions between AIGC service providers and users, including the risk of mismatch between fees and resource costs under the pre-payment model and the payment default risk under the post-payment model.

[0034] This invention proposes a trusted management system and method for the AIGC service lifecycle empowered by blockchain, aiming to build a trusted operating process throughout the entire lifecycle of the AIGC service.

[0035] Figure 1 This is an architecture diagram of a blockchain-enabled trusted management system for the AIGC service lifecycle, as shown in one embodiment of the present invention. The system includes:

[0036] The cloud-edge collaborative blockchain 110 consists of nodes including cloud servers as full nodes and edge servers deployed with an AIGC model as light nodes. These edge servers act as AIGC service providers. The light nodes are used to facilitate consensus on AIGC product transactions between AIGC service providers and users, while the full nodes are used for consensus training logs and maintaining the reputation of AIGC service providers.

[0037] In practical implementation, the content of the cloud-edge collaborative blockchain consensus is not limited to those listed above. The cloud-edge collaborative blockchain is responsible for storing relevant data generated throughout the entire lifecycle of the AIGC service. For example, light nodes can be responsible for consensus user services and transaction transactions, while full nodes can be responsible for consensus model training, providing model services, and cross-shard transactions. Users, as clients, access the AIGC service provider selection module, the AIGC product transaction module, and the cloud-edge collaborative blockchain.

[0038] The AIGC model verification module 120 embeds a model watermark for each AIGC model at the start of pre-training. The encoded vector of this watermark is included in the training log of the corresponding AIGC model. During AIGC model training, the hashed training logs are stored on the cloud-edge collaborative blockchain, while simultaneously being encrypted and stored in an off-chain database. The AIGC model verification module verifies the AIGC model by extracting and comparing the model watermark contained in the on-chain consensus training logs with that stored in the off-chain database. The model watermark includes at least a unique identifier for each AIGC model, which can be used to verify the identity of the AIGC model.

[0039] The AIGC service provider selection modulo 130 is used to select AIGC service providers for users. Specifically, the AIGC service provider reputation construction based on full-node consensus considers a reward function that takes into account the reputation factor of the AIGC service provider. This reward function is used in a multi-objective optimization problem model for selecting AIGC service providers for users. The method that optimizes the reward function for selecting AIGC service providers for users is taken as the determined AIGC service provider selection result.

[0040] AIGC product transaction module 140 is used to establish a state channel through cloud-edge collaborative blockchain when a transaction between AIGC service provider and user begins. Based on light node consensus, AIGC product transaction locks the encrypted AIGC product from AIGC service provider and the encrypted service fee from user, and uses this state channel to realize atomic transactions between AIGC service provider and user.

[0041] By employing this embodiment of the invention, cloud-edge collaborative blockchain can be used to empower the entire lifecycle of AIGC services. During the AIGC model training phase, the AIGC model verification module verifies the AIGC model through an embedded model watermark. When selecting an AIGC service provider for a user, the AIGC service provider selection module makes the optimal choice based on the reputation of the AIGC service provider through full-node consensus. During the AIGC product transaction process, atomic transactions between the AIGC service provider and the user are realized through the state channel established by the cloud-edge collaborative blockchain. This helps to solve the trust challenges existing in the lifecycle of AIGC services in model provision, service provider selection, and product transaction, and facilitates the construction of a trustworthy operating process throughout the entire lifecycle of AIGC services.

[0042] In some embodiments of the present invention, the cloud-edge collaborative blockchain is used to store relevant data generated throughout the entire lifecycle of the AIGC service, and the consensus form of full nodes and light nodes is a DAG (Directed Acyclic Graph Ledger) ledger.

[0043] Using this embodiment of the invention, cloud-edge collaborative blockchain can monitor and manage the entire lifecycle of AIGC services based on stored data.

[0044] In some embodiments of the present invention, the cloud-edge collaborative blockchain uses on-chain smart contracts to execute on-chain operations, and the off-chain database is the InterPlanetary File System (IPFS).

[0045] In some embodiments of the present invention, the model watermark further includes the AIGC model's version number, timestamp, and a randomly generated portion. The version number can be used to verify the AIGC model's version, the timestamp can be used to calculate the AIGC model's training duration, and the randomly generated portion can be used to assist in verifying the AIGC model.

[0046] Using this embodiment of the invention, model watermarking can achieve richer functions in addition to verifying AIGC models.

[0047] In some embodiments of the present invention, the reward function also considers user utility and energy consumption factors, and updates the reputation of the AIGC service provider on the blockchain in real time in the form of a reputation table. Accordingly, the reward function is a multi-objective optimization problem model for users selecting AIGC service providers. This step includes: constructing a multi-objective optimization problem model for selecting AIGC service providers for users; using a preset AIGC service provider selection algorithm to determine the AIGC service provider selection method that maximizes the reward function; and using this as the AIGC service provider selection result for the user by the AIGC service provider selection module.

[0048] Furthermore, in some embodiments of the present invention, the reward function can consider user utility factors by calculating the total user utility value and energy consumption factors by calculating the total energy consumption of all AIGC service providers. The total user utility value is calculated by integrating the AIGC service latency and the user-perceived content quality evaluation function.

[0049] By employing this embodiment of the invention, multiple dimensions of AIGC service provider selection can be comprehensively considered through various reward function approaches to determine the most suitable AIGC service provider selection method for users.

[0050] Furthermore, in some embodiments of the present invention, the AIGC service provider selection algorithm comprises a global network and multiple local networks. Trajectories generated by the interaction between each local network and the environment are asynchronously stored in the experience replay pool of the global network. Each local network updates its policy gradient and value gradient using an A2C (Advantage Actor-Critic) network and transmits them to the global network. The global network uses a stable diffusion model to generate the optimal AIGC service provider selection decision based on the asynchronous trajectory, policy gradient, and value gradient from each local network, thereby maximizing (or "optimizing") the reward function.

[0051] This invention incorporates a diffusion model to generate optimal decisions, and determines the optimal choice for a user's AIGC service provider through an A3C network that incorporates the diffusion model.

[0052] Figure 3 This is a diagram of a trusted management system architecture for the lifecycle of AIGC services empowered by blockchain, as shown in another embodiment of the present invention.

[0053] Figure 3 The diagram illustrates a detailed trusted management system architecture, which is named the CEAIGC architecture in this invention. It comprises a cloud-edge collaborative blockchain platform, an AIGC model verification module, an AIGC service provider selection module, and an AIGC product transaction module. The functions of each part of this architecture will be explained below in conjunction with a specific implementation process.

[0054] For cloud-edge collaborative blockchains, located in the cloud layer (see reference) Figure 2The blockchain's nodes consist of cloud servers (CS) and edge servers located at the edge layer. The edge servers pre-deploy AIGC models and are referred to as ASP (AIGC Server Provider) nodes or light nodes. The cloud servers, acting as full nodes, are responsible for model training, providing model services, and consensus across shards, while users act as clients accessing the blockchain. The edge servers with AIGC models, acting as light nodes, are responsible for user services and transaction consensus. Sharding technology divides the light nodes into multiple shards, enabling parallel processing of requests and achieving consensus, thereby improving the blockchain's throughput. All node consensus is achieved through a DAG ledger to accelerate consensus efficiency. The cloud-edge collaborative blockchain stores all data generated throughout the AIGC service's lifecycle, ensuring data verifiability and tamper-proof protection. On-chain operations can be executed through smart contracts on the blockchain for automated execution, with transparent and tamper-proof rules. To alleviate on-chain storage pressure, the InterPlanetary File System (IPFS) is used as an off-chain database to store some data. Blockchain serves as a foundation of trust to support the efficient and reliable operation of AIGC services.

[0055] For the AIGC model verification module, the designed embedded model watermark encoding rules can be used to achieve unique identification by utilizing the model information contained in the watermark. Alternatively, a unique identifier can be assigned to each AIGC model, and the generated watermark can be encrypted and stored in the blockchain, then embedded at the start of AIGC model pre-training. During model training, the hash value of the training log is stored on the blockchain, while the original log file is encrypted and stored in IPFS. The watermark in the model is extracted and matched with the on-chain watermark for verification, thereby achieving reliable verification of the basic information of the AIGC model.

[0056] For the AIGC service provider selection module, a reward function is constructed by comprehensively considering factors such as user utility, AIGC service provider reputation, and energy consumption. The DMA3C algorithm, which integrates the stable diffusion model and the A3C algorithm proposed in this invention, is used to select a suitable AIGC service provider for the user. The AIGC service provider reputation value is updated in real-time in the blockchain using a reputation table. Details of the reward function construction and the DMA3C algorithm implementation are provided below.

[0057] For the AIGC product transaction module: A state channel is established using smart contracts to transfer AIGC product transactions to an off-chain state channel for execution, thereby reducing the pressure on on-chain transaction processing. The AIGC service provider and user conduct fee and AIGC product ownership transactions within the state channel. Signature and encryption technologies ensure the security and atomicity of the transaction process. A transaction time threshold is set; if this time is exceeded, the transaction will roll back to its original state. After the transaction is completed, the results are updated on-chain and the state channel is closed.

[0058] Next, combined Figure 2 The AIGC model verification module is described, including four stages: ① generation, ② embedding, ③ extraction, and ④ verification of the model watermark.

[0059] Model watermark generation: The model watermark is generated by T bits {0, 1} T The composition, for example, could be a 64-bit model watermark b∈{0,1} T Its structure is as follows: the first 16 bits are the model identifier, which can be obtained from the AIGC model encoding library. Different watermark codes can be used to uniquely identify the corresponding model type. The next 8 bits represent the AIGC model version number, the 32 bits are the AIGC model timestamp, and the last 8 bits are a randomly generated part. This encoding method not only achieves a unique identifier for the model but also conveys basic model information. The generated model watermark 'b' will be used on a cloud server. ci private key sk csi It is then encrypted and stored in the blockchain.

[0060] Model watermark embedding: The model watermark is embedded at the beginning of AIGC model pre-training, so that the presence of the watermark does not impair the model's performance on the original task. The watermark is embedded in the convolutional layer because fully connected layers are often discarded during fine-tuning. Here, (S,S) represents the size of the convolutional filter, D represents the input depth of the convolutional layer, and L represents the number of filters in the convolutional layer. The convolutional layer parameters are then determined by W∈R. S*S*D*L R belongs to the real number domain. To eliminate the arbitrariness in the filter ordering, the average value of W over the L filters is calculated, i.e. Using w∈R S*S*D express Set the watermark b∈{0,1} T Embedded into w. E o (w) represents the general cost function. The cost function E(w) with a parameterized regularizer is defined as follows:

[0061] E(w)=E o (w)+λE R (w); (1)

[0062] The loss function E of the embedded regularizer is defined using binary cross-entropy. R (w):

[0063]

[0064] Where y j =σ(∑ i X ji w i The σ(*) function is the sigmoid function:

[0065]

[0066] During model training, a log file is used to record key information about the training process. This log file records training parameters and configurations, training epochs, training timestamps, loss functions, and related information, including the generated model watermark encoding vector b. The generated log file is then hashed and stored on the blockchain, with the original log file using the private key sk. csi It is encrypted and stored in an off-chain IPFS database.

[0067] Model watermark extraction: Watermarks need to be extracted from the convolutional layer parameters w, where the average vector parameters w∈R S*S*D and embedding parameter X∈R T*S*S*D The X needs to be projected onto w, and the process is performed at the defined threshold of 0. Therefore, the j-th bit of the model watermark b′ is extracted as follows:

[0068] b′ j =s(∑ i X ji w i (4)

[0069] The s(x) function is as follows:

[0070]

[0071] Model watermark verification: Based on the extracted watermark b′∈{0,1} T Obtain the watermark from the blockchain and use the public key to PK. csi Decryption yields the watermark b∈{0,1} T If the model is compared with b′ and they match, then the model is indeed the target model, and ownership of the model is confirmed. The extracted watermark is parsed to obtain the first 16 characters of the AIGC model identifier, which is then compared with the identifier recorded in the model encoding database to verify and determine the AIGC model type. The model log file is retrieved from the blockchain, and the public key is obtained and used for PK (penetrating) from off-chain. csiThe original log file is decrypted and hashed against the blockchain to verify data integrity. Finally, the timestamp field of the AIGC model training is extracted from the log file and compared with the watermark timestamp field to calculate the training duration of the AIGC model.

[0072] By adopting this implementation method and integrating blockchain and model watermarking technology, the intellectual property rights, model type and training time of AIGC models can be effectively verified, thereby achieving consistency verification of basic model information.

[0073] Next, combined Figure 2 The article elaborates on the selection module for AIGC service providers, taking into account three factors: user utility, AIGC service provider reputation, and energy consumption.

[0074] AIGC service provider selection involves choosing a suitable and reliable AIGC service provider (ASP) for users. This problem is similar to a resource-constrained allocation problem, where the goal is to allocate incoming tasks to available resources to maximize overall utility within resource constraints. Therefore, this invention will comprehensively consider user utility, AIGC service provider reputation, and energy consumption to construct an incentive function with a multi-faceted objective. Define users MU = {mu1, mu2, ... mu}. M Then the task is defined as Tasks = {task1, task2, ..., task...} J A set of available ASPs is defined as ASPs = {asp1, asp2, ..., asp...} I}, and M≥I. The goal is to assign user-initiated tasks to ASP, i.e., A={a1,...,a j , ..., a J}, to maximize its activation function Obj. Where asp i The total available resources are represented as res i (i=1,...,I), then b ij Represented as i th ASP execution j th The amount of resources consumed by the task. This indicates that when assigning the current task, i... th A set of tasks running on an ASP. The AIGC service provider selection problem can then be represented as decision variables in A. For integer programming problems, the activation function can be expressed as:

[0075]

[0076] Where U represents the utility value of the total users, C represents the total reputation value of all edge servers, and E represents the total energy consumption of all edge servers, where αk＝1，2，3 It is a weighting factor used to balance maximizing user utility, ASP reputation, and minimizing energy consumption.

[0077] Regarding user utility factors, the overall evaluation of ASP generation by users mainly consists of two parts: service latency T and the perceived content evaluation function. Considering service latency, a longer latency may indicate congestion problems in the ASP system. On servers with stronger computing power, the time required to process the same content is shorter. Therefore, service latency is an important indicator for evaluating user utility.

[0078] 1) Service latency: The service latency mainly considers the task queuing latency T. que Task processing latency T pro And the latency T for uploading AIGC content to the blockchain. net Among them, Task λ For task completion rate, Task μ If the task processing rate is T, then the task queue is T. que for:

[0079]

[0080] S AIGC Indicates the size of the AIGC output. This represents the average bandwidth in the network. Task Diff Indicates the difficulty of the AIGC task. Representing the edge of ASP i The average amount of computing resources invested per second, then Then process task a j The total service latency is as follows:

[0081]

[0082] 2) Perceived content evaluation function: then use Let i represent the user-perceived content quality evaluation function, then the i-th... th Execute j on an ASP th The perceived content quality assessment value for a task can be expressed as:

[0083] Taking AIGC services that generate images as an example, in the absence of an original image as a reference, image quality evaluation metrics include: ① Total Variation (TV), used to measure image smoothness and denoising effect; a low TV value indicates smoothness, while a high TV value may indicate excessive noise or detail. ② Blind / Referenceless Image Spatial Quality Evaluator (BRISQUE), used to evaluate the degree of natural image distortion; a lower score indicates better image quality, and a higher score indicates more severe distortion. Therefore, the user-perceived content quality evaluation function... The function can be a TV or a BRISQUE function. Because different ASPs differ in computing resources, computing power, and model deployment, the user utility values ​​when executing tasks vary significantly, highlighting the importance of choosing the right ASP.

[0084] To eliminate the magnitude differences between different indicators, the above indicators are normalized using the function Θ(*). Where x max and x min These are the upper and lower bounds of x, respectively.

[0085]

[0086] Then the user initiates a task j The utility value is:

[0087]

[0088] Where μ∈[0,1].

[0089] The user's total utility value is: Regarding the reputation factor of AIGC service providers (i.e., ASPs), the server's reputation score is also an important consideration when selecting an ASP. We consider the ASP's reputation index in the node reputation metrics. i Task completion rate, available time rate (percentage), and user ratings of ASP history.

[0090] This invention constructs a reputation index based on ASP. i Task completion rate but in It's ASP. i Total number of tasks received, com taski It's ASP. i The number of successfully processed tasks is the percentage of tasks successfully completed by the server out of the total tasks. The higher the task completion rate, the higher the server's reliability.

[0091] This invention constructs a reputation index considering ASP. iPercentage of available time for reputation metrics but For ASP i Available time (i.e., the time during which tasks can be processed normally). For ASP i The longer the uptime, the more reliable the server.

[0092] User ratings of ASP history: These are user ratings based on ASP history. i The results of the inference service are used to assign a satisfaction score, with the score ranging from [0,1]. Let N be the value of the inference service. mi For user mu m to asp i The number of times a score has been given, its Mu represents m For ASP i The score of the nth evaluation. For asp i The rating is:

[0093]

[0094] However, for user rating scores, the weight of low scores (i.e., low scores indicating dissatisfaction) should be greater than that of relatively satisfactory scores. In this invention, the range of dissatisfaction scores is set to [0, 0.6], and the weight is set to ζ. u Then set For mu m For ASP i The rating is the number of times you are dissatisfied. Indicates to mu m For ASP i The unsatisfactory score in the nth evaluation. The relatively satisfactory score ranges from (0.6, 1], and the weight is set to ζ. s ,set up For mu m For ASP i The number of times the rating was relatively satisfactory. Indicates to mu m For ASP i If the relative satisfaction score of the nth evaluation is , then the corrected asp i The rating is:

[0095]

[0096] s.tζ u +ζ s =1

[0097] ζ s <ζ u

[0098]

[0099] In edge network environments, due to the rapid iteration of AI-generated models, ASP performance can change significantly over time, thus giving more weight to the latest evaluation results than older ones. However, to prevent malicious users from submitting a large number of tasks in a short period and manipulating ASP scores through extreme ratings, this invention introduces a weighting factor:

[0100]

[0101] Where λ > 0 is the time decay factor, which controls the decay rate. This indicates asp i The timestamp for receiving the latest evaluation Indicates its mu m For ASP i The timestamp of the evaluation Mu represents m For ASP i The amount of service fees paid for performing AIGC inference services.

[0102] The final score for ASP is:

[0103]

[0104] User ratings of ASPs are stored in the blockchain in the form of a rating table, ensuring that the rating table data cannot be tampered with. This allows users to obtain the latest and most accurate data ratings when making a comprehensive selection of ASPs.

[0105] The reputation metric value of ASP is updated in real time and consensus is reached on the blockchain. Then the i-th... th The reputation value of an ASP is:

[0106]

[0107] The total reputation value of ASP is then:

[0108]

[0109] Regarding energy consumption factors, in the ASP selection problem, the energy consumption function of the AIGC service provider's edge server is defined as follows:

[0110] Where E i It is an edge server ESP i Total energy consumption, P idle This refers to the server's energy consumption when it is idle, P. max It is the energy consumption of the server when it is fully loaded, L i It's ASP. iThe current server load, L max It is the server's maximum load capacity, T i It's ASP. i Total time for the server to process tasks.

[0111] The total energy consumption of all AIGC edge service providers is:

[0112]

[0113] By constructing an incentive function that integrates multiple objectives, the optimal decision is generated using a diffusion model. The Markov chain decision process is modeled, and the DMA3C algorithm (Section IV) is used to select the ASP (AIGC service provider) that maximizes the incentive function for the user.

[0114] Next, combined Figure 2 and Figure 4 This section describes the AIGC product transaction module. Figure 4 This is a flowchart of the AIGC product transaction module in one embodiment of the present invention, including the process of establishing a status channel, the transaction process, and the process of closing the channel.

[0115] When user mu m Determine the selected AIGC service provider (represented as asp) i After that, when asp i Output AIGC product p k Subsequently, due to the large storage space required for AIGC products, they were stored in an off-chain IPFS database, and a hash function was used to output Hash(p). k Its off-chain storage address Address k and retrieval index k , forming p k On-chain storage tuples Will use ASP i private key After signing It is stored on the blockchain. Among them, state channels are an off-chain scaling solution designed to address the problems of slow blockchain transaction speeds and high costs.

[0116] Regarding the process of establishing a state channel, when mu m and ASP i Upon establishing a transaction relationship, public keys are first exchanged to ensure secure subsequent communication. Then, both parties submit a state channel establishment request via an on-chain smart contract and confirm the transaction initiation. The smart contract verifies the legality of the service conditions and checks whether the user's on-chain account balance meets at least the agreed service fee f. iIf the balance is insufficient, channel creation fails; once verification is successful, the smart contract locks the corresponding funds and officially opens the channel.

[0117] To prevent stalls caused by malicious delays by either party during the transaction process, a time threshold T is set for the transaction. tran If the transaction process is not completed within T... tran If the transaction is completed within the specified time, the smart contract will automatically terminate the transaction and roll back to its initial state. All locked fees and data will be released, ensuring that the rights of both parties are not harmed. The transaction steps will be detailed below, as follows: Figure 4 As shown:

[0118] ① When the transaction begins, user mu m First, generate a symmetric key Dec. m and using Dec m Service fees for the agreement f j Encryption is performed, and an encryption fee is generated. The encrypted fee was then locked in an off-chain channel.

[0119] ②mu m Encrypt Dec using its own public key. m ,generate Send it to the service provider ASP i .

[0120] ③asp i Encrypting (signing) the AIGC tuple information yields use Encrypt and lock the data in the off-chain channel to ensure that both the tuple information and the fees are locked.

[0121] ④ When user mu m Need to obtain AIGC tuple information When using PriKey via off-chain channel num and Dec m Unlock the AIGC tuple information to complete the data access operation. During this process, Dec... m Will be exposed to ASP i At this time, asp i The received Dec can be used m Unlock fee balance f j The corresponding transaction amount f j Complete the payment transfer.

[0122] This design ensures that while the user successfully unlocks and obtains the AIGC tuple information, the service provider unlocks the corresponding service fee, achieving atomic transfer of transaction fees.

[0123] The above process allows transactions within an off-chain channel to be repeated multiple times, until user mu m The balance in the off-chain channel is exhausted. After each transaction, both parties must synchronize and sign to update the channel status to ensure that subsequent transactions can be based on the latest status. Users can securely and efficiently obtain the necessary AIGC tuple information, while service providers can reliably collect fees.

[0124] For the channel closure process, when both parties in the state channel decide to terminate the transaction or the transaction times out, both parties must sign a final state including the balance allocation and submit it to the on-chain smart contract. The smart contract verifies the validity of the signature and confirms the latest state agreed upon by both parties. In case of dispute, the contract can arbitrate based on the submitted historical states or transaction records, and complete the fund settlement with the last valid signature state. After settlement, funds are transferred to the on-chain account, the channel is closed and resources are released, ensuring the transparency and finality of on-chain settlement. Figure 3 As shown, the AIGC product transaction module provides four services: atomic transaction, timeout rollback, signature identification, and status update.

[0125] Figure 5 This is a diagram of the DMA3C algorithm architecture for AIGC service provider selection in one embodiment of the present invention. The DMA3C algorithm is an ASP selection algorithm model based on diffusion reinforcement learning. The DMA3C algorithm is used to determine the selection method of AIGC service providers, and includes three parts: diffusion model to generate optimal decision, Markov decision process modeling process, and DMA3C algorithm architecture.

[0126] For diffusion models to generate optimal decisions, the Denoising Diffusion Probability Model (DDPM) comprises two steps: the forward process gradually adds noise to the data until it becomes entirely Gaussian noise; the backward diffusion process gradually removes noise until the original data is restored. The solution set for the ASP (Application Policy Optimization) choice problem is finite and discrete, therefore, the decision scheme can be represented as a discrete probability distribution for each decision. This problem can be solved by combining diffusion models with reinforcement learning. Specifically, the forward process continuously adds noise to the optimal policy solution, gradually transforming it into Gaussian noise; the backward diffusion process considers the RL (Reinforcement Learning) policy network π... θ (*) represents a denoiser that starts with Gaussian noise and, combined with the current environmental state s, gradually removes noise to ultimately generate the optimal decision solution x0. The following section details the process of generating the optimal decision using the diffusion model, including the forward and backward diffusion processes. The role of the critic network in generating the optimal decision is similar to that elsewhere.

[0127] The forward process refers to adding Gaussian noise to the data until the data becomes completely noisy. The optimal decision solution is then output. This is the probability distribution of each decision made under the observed environmental state s. Through step x t-1 Adding Gaussian noise yields x t The Gaussian distribution is defined by its mean and variance, where the mean is . The variance is β t I, then the process is represented as:

[0128]

[0129] The two-dimensional standard Gaussian distribution ∈ ~N(0,1) is obtained through the parameter β. t By x t-1 Get x t , where x t It can be represented as The entire forward process of the diffusion model can be represented as x1, x2, ..., x T A Markov chain, the formula for a Markov chain is expressed as:

[0130]

[0131] definition Define α t =1-β t By reparameterizing, we can obtain a distribution q(x) that follows the distribution. t Any sample x in |x0) t The formula is expressed as:

[0132]

[0133] The backdiffusion process is essentially a denoising inference process. At time T, the Gaussian noise is progressively denoised to obtain the decision set x0. This process is then analyzed using q(x... t |x t-1 Sampling can be performed from Gaussian noise x T ~N(0,I) reconstructs the x0~q(x) distribution, and then uses a deep learning model to predict the inverse distribution:

[0134]

[0135] The mean and variance are calculated based on the model, and the backward process of the diffusion model is obtained through a Markov chain:

[0136] Where p(x) t )=N(x T ;0,I) is randomly sampled Gaussian noise, p θ (x t-1 |x t ) is a Gaussian distribution whose mean and variance need to be calculated.

[0137] The mean value is calculated based on the forward process reasoning:

[0138]

[0139] Calculate according to formula (20) Where ∈ θ (x t , t, s) is a depth model parameterized by θ, which generates denoised noise based on the observation value s. Substituting x0 into formula (4.19) yields,

[0140]

[0141] The variance is:

[0142]

[0143] Using formula (22), we obtained x. t-1 ~p θ (x t-1 |x t Then we can further obtain from x t To x t-1 The calculation formula is as follows:

[0144]

[0145] Z ~ N(0, I) is a two-dimensional standard Gaussian distribution.

[0146] Finally, the softmax function is applied to x0 to transform it into a probability distribution, i.e.

[0147]

[0148] π θ (s) represents the probability of choosing each action corresponding to the current environment s. Through the above DDPM process, the probability distribution of the optimal decision x0 is derived, the decision value network in DRL is constructed, and the probability distribution of the optimal decision is output.

[0149] For Markov Decision Process (MDP) modeling, J tasks are assigned to I ASPs to maximize their incentive function Obj. This function considers user utility, server reputation, and energy consumption for task processing, taking into account the resource constraints of each ASP. Each ASP has a limited number of tasks it can handle, which must be within its resource capacity; otherwise, congestion or task failures may occur. MDP provides a robust structure for decision-making in dynamic and uncertain environments, effectively addressing this problem. An MDP is defined as (S, A, P, R), where S is a finite set of states, A is a finite set of actions, P is the state transition probability (the probability of changing from state s to s′), and R is the reward value after the action execution and state transition. Given an initial state s0, in each step 1 = 0, 1, ..., L, the action a1 ∈ A is executed, changing the environment from state s0 to state s0. l Transform ∈S to s l+1 ∈S, and obtain a reward value r from the corresponding environment. l ∈R.

[0150] State space: A state is a specific set of discrete or continuous variables. An agent can perceive these changes from interactions in the environment and make decisions. The state space S contains the environmental information for making decisions. The state of the agent s∈S consists of three feature vectors, representing the current arrival of the user task s. J′ The current resource status of all ASPs. Res′ The current reputation values ​​of all ASPs. C′ .

[0151] Vector s J′ Including the current task's arrival requirements and J′ and its expected completion time t, the vector s J′ =[J′,t].

[0152] Vector s Res′ It records the total available resources (res) for each ASP. i and current available resources Then vector

[0153] Vector s C′ The initial reputation value of each ASP was recorded. Compared to the current reputation score C i Then the vector

[0154] Then, from the state vector s, from the vector s J′ s Res′ s C′ Combining all the components, the vector S = [s J′ s Res′ sC′ Relevant parameters J′, t, res i , The value of π in the calculation strategy network θ Before training, the training function needs to be normalized so that its value is in the range of (0,1). This is to stabilize the training process and improve generalization ability.

[0155] Action Space: The action space A is defined as the set of all possible decisions that an agent can make, where each action a ∈ A is to assign the user task to one of I available ASPs. The action space is discrete, consisting of integers from [1, I]. This entire operation is controlled by the policy network π. θ (s) is determined by using the current state s as the input vector of elements I, where each element represents the probability of selecting a specific ASPs, i.e., a ~ π. θ (s), the next action to be executed will select the ASP with the highest output probability. i ,Right now:

[0156]

[0157] Reward Function: The reward function provides numerical feedback by quantifying the quality of actions performed by the agent. The agent learns to execute actions that maximize cumulative rewards. Let r∈R represent the immediate benefit gained from performing action a in state s. The reward function r(s,a) consists of three parts: user utility reward r... U ASP Reputation Rewards C and energy consumption penalty r E User utility rewards measure the quality of content generated by ASPs that are successfully assigned to tasks, thereby improving user satisfaction; reputation rewards reflect the success rate and response efficiency of ASPs in executing tasks; and energy consumption penalties are used to constrain the energy consumption of ASPs when executing tasks, in order to reduce operating costs and alleviate network congestion.

[0158] The reward function is then:

[0159] r = α1r U +α2r C -α 3 r E (30)

[0160] Its purpose is to make the cumulative reward function If the value is maximized, then the expected value of the cumulative discount reward is:

[0161]

[0162] Where γ is the reward discount factor, with a value range of γ∈[0,1], is a parameter relating future rewards to present rewards, and θ is a parameter of the diffusion policy network π in DRL.

[0163] For the DMA3C (Diffusion Model-based Asynchronous Advantage Actor-Critic) algorithm architecture, such as Figure 5 As shown, the architecture consists of a global network and multiple local networks (LocalWorkers), each running an A2C network. The global policy network employs a stable diffusion model combined with experience replay. Due to the fast computation speed of the "Actor network" within the A2C network, sampling and update frequencies are increased. The stable diffusion model optimizes the convergence of long-term policies in the global policy network, generating higher-quality policies. Experience replay mitigates fluctuations in the data distribution of local networks, preventing overfitting and enhancing policy reliability. In the "Critic network," experience replay reduces the variance of value estimation, making asynchronous updates more stable and accurate. The asynchronous nature of the A3C (Asynchronous Advantage Actor-Critic) algorithm allows for rapid adaptation to changes in tasks and resources in dynamic environments. For the constantly changing task arrivals, service resources, and network conditions in the ASP selection problem, A3C's asynchronous exploration and real-time updates can promptly find policies suitable for the current environment. Multi-threaded parallel computation accelerates the convergence process, and asynchronous updates reduce sample correlation and the risk of local optima and overfitting, making the model more stable in dynamic environments. Therefore, this invention applies the A3C algorithm to the ASP selection problem.

[0164] In each parallel worker thread of the Local Workers (LW) network, an actor-critic network is utilized. The actor-critic network is based on local states (e.g., Figure 5 Extracting the value function V(s, θ′) from local states 1-N in the model v The policy function π(a|s;θ′) is combined with the value function V(s,θ′). v The action value function π(a|s; θ′) is used by the critic network to estimate the value of generated actions, while the actor network uses this value to optimize the policy, aiming to maximize future discounted rewards in the long run. The action value function Q(s, a) evaluates the long-term payoff for each current action. Specifically, the action value function is:

[0165] Q(s l a l ) = r l +γV(s l+1 , θ′ v(32)

[0166] The dominance function is:

[0167] A(s l a l )=Q(s l a l )-V(s l ) = r l +γV(s l+1 , θ′ v )-V(s l , θ′ v (33)

[0168] The value network is updated by calculating the value gradient through its mean squared error loss function:

[0169]

[0170] Then calculate the gradient of the policy function. Then update the policy network:

[0171]

[0172] Where, ρ π It is the state distribution, which depends on the policy parameters, and is also known as the policy gradient theorem.

[0173] After each worker thread in the local network performs gradient updates, it will store the policy gradient θ′ and the value gradient θ′. v This data is passed to the global network for asynchronous parameter updates. While parameters are being updated in the worker thread, trajectories are generated through interactions between the various local networks and the environment. The samples will be asynchronously stored in the global network's experience replay pool, which will be updated periodically to retain the latest high-quality samples.

[0174] The Global Network (GN) randomly samples data from the experience replay pool to update the policy network and value network. The data in the replay pool originates from interactions between multiple Local Worker Networks (LWs) and the environment, reducing data correlation and improving update stability. Simultaneously, the local networks asynchronously upload gradients θ′ and θ′. v In the global network, GN combines these gradients with the replay pool data for asynchronous updates to global parameters. The global network uses a diffusion model in the policy network to update policy parameters and a Critic network in the value network to update value parameters. After the update is complete, the global network will update the latest parameters θ and θ'. v Synchronize to all local networks, ensuring they are up-to-date with the latest policies and environment (e.g., ...). Figure 5Interactions within the environment (1-N).

[0175] Extensive experimental verification shows that the CEAIGC proposed in this invention can increase system throughput by 2.38 times, and the proposed DMA3C algorithm improves performance by 11.4%-28.6% compared to other DRL algorithms, thus supporting the efficient and reliable operation of AIGC services.

[0176] In summary, the blockchain-enabled trusted management system and method for the AIGC service lifecycle proposed in this invention adopts a newly designed AIGC service architecture (CEAIGC) that includes a cloud-edge collaborative blockchain, an AIGC model verification module, an AIGC service provider selection module, and an AIGC product transaction module, which can support the efficient and trusted operation of AIGC services.

[0177] Furthermore, the embedded AIGC model watermark encoding rule designed in this invention provides a watermark that can encompass basic model information and uniquely identify the model. By embedding the model watermark into the AIGC model and encrypting it in the blockchain, and then verifying the watermark extracted from the model against the watermark in the blockchain, the trusted verification of the AIGC model is achieved.

[0178] Furthermore, this invention comprehensively considers user utility, ASP reputation, and energy consumption factors to construct a multi-objective optimization problem model, and proposes a diffusion deep reinforcement learning (DRL) AIGC SP optimization selection algorithm (DMA3C). The diffusion model is used to replace the Actor network in the global network to generate the optimal decision, thereby comprehensively selecting a suitable and reliable ASP for the user. This is conducive to achieving accurate matching between the ASP and user needs, and providing users with reliable low-latency AIGC inference services at the edge.

[0179] Furthermore, this invention proposes a method for establishing state channels using a smart contract engine, transferring transactions to off-chain state channels. During the transaction process, signature and encryption technologies are used, and transaction time thresholds are set to achieve efficient, secure, and atomic ownership transfer of AIGC products and fees. This method breaks through the performance bottleneck of blockchain by constructing state channels between transaction users through a smart contract engine, enabling efficient, secure, and atomic ownership transfer of AIGC products and fees off-chain.

[0180] Corresponding to the aforementioned trusted management system, this invention also provides a blockchain-enabled trusted management method for the AIGC service lifecycle, comprising: an AIGC model verification module embedding a model watermark for each AIGC model at the start of pre-training; the encoded vector of the model watermark being included in the training log of the corresponding AIGC model; during the training process of the AIGC model, on the one hand, storing the hashed training log on the cloud-edge collaborative blockchain, and on the other hand, encrypting and storing the training log in an off-chain database; and verifying the AIGC model by extracting and comparing the model watermark contained in the on-chain consensus training log with the training log stored in the off-chain database; wherein, the model watermark includes a unique identifier for each AIGC model. The identifier is 1; the AIGC service provider selection module selects an AIGC service provider for the user; wherein, the AIGC service provider reputation construction based on full node consensus considers the reward function of AIGC service provider reputation factors, and the reward function is used for a multi-objective optimization problem model for selecting AIGC service providers for the user; the AIGC product transaction module establishes a state channel through cloud-edge collaborative blockchain when a transaction between the AIGC service provider and the user begins, and the AIGC product transaction based on light node consensus locks the encrypted AIGC product from the AIGC service provider and the encrypted service fee from the user, and uses the state channel to realize atomic transactions between the AIGC service provider and the user.

[0181] The cloud-edge collaborative blockchain includes cloud servers as full nodes and edge servers with AIGC models deployed as light nodes. The edge servers with AIGC models serve as AIGC service providers. The light nodes are used to reach consensus on AIGC product transactions between AIGC service providers and users. The full nodes are used to reach consensus on training logs and AIGC service provider reputation.

[0182] Corresponding to the aforementioned trusted management system, the present invention also provides an electronic device, which includes a computer device, the computer device including a processor and a memory, the memory storing computer instructions, and the processor executing the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the functions of the trusted management system as described above.

[0183] Corresponding to the aforementioned trusted management system, the present invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the functions of the trusted management system as described in any of the above embodiments. The computer-readable storage medium may be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

[0184] Corresponding to the aforementioned trusted management system, the present invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the functions of the trusted management system as described in any of the above embodiments.

[0185] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0186] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0187] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0188] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A trusted management system of a blockchain empowered AIGC service life cycle, characterized in that, The cloud edge collaborative blockchain includes a cloud server as a full node and an edge server deployed with an AIGC model as a light node, and the edge server deployed with the AIGC model serves as an AIGC service provider; wherein the light node is used for consensus of AIGC product transactions between the AIGC service provider and the user, and the full node is used for consensus of training logs and AIGC service provider reputation; An AIGC model verification module is configured to embed a model watermark in each AIGC model at the beginning of pre-training, and the encoding vector of the model watermark is contained in the training log of the corresponding AIGC model. During the training process of the AIGC model, on one hand, the training log after hash operation is stored on the cloud edge collaborative blockchain, and on the other hand, the training log is stored in an off-chain database in an encrypted manner. The AIGC model verification module is configured to realize AIGC model verification by extracting and comparing the model watermark contained in the training log stored in the off-chain database and the training log stored on the chain. An AIGC service provider selection module is configured to select an AIGC service provider for the user; wherein a reward function considering AIGC service provider reputation factors is constructed based on the AIGC service provider reputation consensus of the full node, and the reward function is used for a multi-objective optimization problem model of selecting an AIGC service provider for the user. An AIGC product transaction module is configured to establish a state channel through the cloud edge collaborative blockchain at the beginning of the transaction between the AIGC service provider and the user, and the AIGC product transaction is locked based on the consensus of the light node, the AIGC product encrypted from the AIGC service provider and the service fee encrypted from the user, and the state channel is used to realize the atomic transaction between the AIGC service provider and the user. The reward function is used for a multi-objective optimization problem model of selecting an AIGC service provider for the user, including: constructing a multi-objective optimization problem model for selecting an AIGC service provider for the user, determining an AIGC service provider selection method that maximizes the reward function by using a preset AIGC service provider selection algorithm, and taking the AIGC service provider selection result of the AIGC service provider selection module for the user as the result of the AIGC service provider selection module for the user. The AIGC service provider selection algorithm includes a global network and multiple local networks, the trajectories generated by each local network interacting with the environment are asynchronously stored in the experience replay pool of the global network, each local network updates the policy gradient and value gradient and passes them to the global network, and the global network generates an optimal decision of the AIGC service provider selection based on the trajectories, policy gradient and value gradient from each local network asynchronously using a stable diffusion model, so as to maximize the reward function. The cloud edge collaborative blockchain is used to store the related data generated in the whole life cycle of the AIGC service, and the consensus form of the full node and the light node is a DAG ledger.

2. The trust management system of claim 1, wherein, ​ The cloud-edge collaborative blockchain performs on-chain operations using on-chain smart contracts, and the off-chain database is an Interplanetary File System (IPFS).

3. The trust management system of claim 1, wherein, The model watermark further includes a version number, a timestamp, and a randomly generated part of the AIGC model, the version number is used to verify the version of the AIGC model, the timestamp is used to calculate the training duration of the AIGC model, and the randomly generated part is used to assist in verifying the AIGC model.

4. The trust management system of claim 1, wherein, The reward function also considers user utility factors and energy consumption factors, and updates the AIGC service provider reputation in real time on the blockchain in the form of a reputation table.

5. The trust management system of claim 4, wherein, The user utility factor is considered in the reward function by calculating the total user utility value, wherein the total user utility value is calculated by combining the AIGC service delay and the user perceived content quality evaluation function; The energy consumption factor is considered in the reward function by calculating the total energy consumption of all AIGC service providers.

6. A method for trusted management of a blockchain empowered AIGC service life cycle, characterized in that, It includes: The AIGC model verification module embeds a model watermark in each AIGC model at the beginning of pre-training, the encoding vector of the model watermark is included in the training log of the corresponding AIGC model, and during the training process of the AIGC model, on the one hand, the training log after the hash operation is stored on the cloud-edge collaborative blockchain, and on the other hand, the training log is stored in the off-chain database, and the AIGC model verification is realized by extracting and comparing the model watermark contained in the training log of the on-chain consensus and the training log stored in the off-chain database; wherein the model watermark includes an identifier unique to each AIGC model; The AIGC service provider selection module selects an AIGC service provider for the user; wherein the reward function considering the AIGC service provider reputation factor is used to build an AIGC service provider reputation based on full node consensus, and the reward function is used for the multi-objective optimization problem model of selecting an AIGC service provider for the user; the reward function for the multi-objective optimization problem model of selecting an AIGC service provider for the user includes: constructing a multi-objective optimization problem model for selecting an AIGC service provider for the user, using a preset AIGC service provider selection algorithm to determine the AIGC service provider selection method that maximizes the reward function, as the AIGC service provider selection result of the AIGC service provider selection module for the user; the AIGC service provider selection algorithm contains a global network and multiple local networks, each local network asynchronously stores the trajectory generated by interacting with the environment into the experience replay pool of the global network, each local network updates the policy gradient and value gradient using the A2C network and passes it to the global network, and the global network generates the optimal decision of the AIGC service provider selection based on the trajectory, policy gradient and value gradient from each local network using the stable diffusion model, so as to maximize the reward function. The AIGC product transaction module establishes a state channel through a cloud-edge collaborative blockchain when a transaction between an AIGC service provider and a user starts, locks AIGC products from the AIGC service provider and service fees from the user based on light node consensus, and realizes atomic transaction between the AIGC service provider and the user by using the state channel. The cloud-edge collaborative blockchain includes a cloud server as a full node and an edge server with an AIGC model as a light node, the edge server with the AIGC model is an AIGC service provider, the light node is used for consensus of AIGC product transactions between the AIGC service provider and the user, and the full node is used for consensus of training logs and AIGC service provider reputation.

7. An electronic device comprising a processor, a memory, and a computer program / instructions stored on the memory, wherein, The processor is configured to execute the computer programs / instructions, and the electronic device implements the functions of the trusted management system according to any one of claims 1 to 5 when the computer programs / instructions are executed.

8. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that, The computer programs / instructions are executed by the processor to implement the functions of the trusted management system according to any one of claims 1 to 5.

9. A computer program product comprising computer programs / instructions, characterized in that, The computer programs / instructions are executed by the processor to implement the functions of the trusted management system according to any one of claims 1 to 5.

Images