Game updating method and system based on game development data adjustment and storage medium

Abstract

The invention relates to the technical field of computers, in particular to a game updating method and system based on game development data adjustment and a storage medium. A state vector is formed by combining preset key indexes, a candidate scheme is generated by utilizing a deep reinforcement learning model, innovative contents are output by adopting generative adversarial network training, and an integrated updating scheme is formed through nonlinear weighted fusion; the integrated updating scheme is deployed on a digital twinning simulation platform, operation data are collected through a virtual player model simulation test, a verification scheme is generated through statistical analysis and multi-target optimization, and self-adaptive parameter adjustment is automatically triggered according to real-time feedback to form adjustment data; the adjustment data and the verification scheme data are integrated to generate a full-amount updating scheme, rolling deployment is implemented through an automatic deployment platform, and closed-loop feedback periodic dynamic iteration updating is achieved. Intelligentization, automation and dynamic optimization of game updating are achieved, and system stability and user experience are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a game update method and system based on game development data adjustment, and a storage medium. Background Art

[0002] With the increasing richness of game content and continuous expansion of functions, game updates have become a key means to improve user experience and extend the product life cycle. However, the existing technologies mainly rely on manual or semi-automated adjustment of the overall game code (Chinese invention patent, publication number: CN119002969B), and achieve updates through code partitioning, manual identification and replacement. The core means include code segmentation, error log analysis, static code replacement, etc. These technical means have many deficiencies: First, the large number and complexity of the code lead to a cumbersome adjustment process and an extended development cycle; Second, there is a lot of manual intervention, making it difficult to achieve real-time dynamic updates, and in the case of large amounts of data and frequent changes, it is easy to have omissions and errors; Third, the traditional method relies on fixed repair strategies for BUG handling and lacks a perfect automated response mechanism, which may cause some BUGs to not be repaired in time, thus affecting the overall running stability of the game.

[0003] Generally speaking, the existing technologies have obvious defects in terms of update efficiency, automation level, and dynamic response ability. There is an urgent need for an update method that uses real-time data-driven and intelligent decision-making closed-loop to improve game update efficiency and dynamic response ability. Summary of the Invention

[0004] In view of the above-mentioned many problems existing in the prior art, the present invention provides a game update method and system based on game development data adjustment, and a storage medium. The present invention collects and preprocesses multi-source game data to form fusion data, then combines preset key indicators to form a state vector, uses deep reinforcement learning to generate candidate solutions, and at the same time uses a generative adversarial network to train and output innovative content, and generates an integrated update solution through non-linear weighted fusion. The present invention is tested by a virtual player model simulation on a digital twin simulation platform, and automatically triggers adaptive parameter adjustment according to real-time feedback, and finally realizes the update dynamic iteration of the intelligent decision-making closed-loop, effectively improving the game update efficiency and dynamic response ability.

[0005] A game update method based on game development data adjustment includes the following steps:

[0006] Collect player behavior data, system performance data, community sentiment data, and BUG feedback data from the game environment. After preprocessing all the data, perform feature extraction and fusion processing to form fusion data;

[0007] Construct a state vector based on the fusion data and preset key indicators, use a deep reinforcement learning model to generate candidate solutions, and at the same time use a generative adversarial network to train historical scenario data and current game style data to generate innovative content, and fuse the candidate solutions and innovative content through a non-linear weighted fusion method to form an integrated update solution;

[0008] Deploy the integrated update solution on a digital twin simulation platform, conduct simulation tests on the integrated update solution through a virtual player model, collect operation data, and use statistical analysis and multi-objective optimization to generate a verification solution. At the same time, automatically trigger adaptive parameter adjustment based on real-time feedback to form adjustment data;

[0009] Integrate the adjustment data and the verification solution data to generate a full-scale update solution, implement rolling deployment of the full-scale update solution through an automated deployment platform, and use real-time monitoring data to form a closed-loop feedback to periodically trigger dynamic iterative updates.

[0010] Preferably, the preset key indicators are player retention rate, system response indicators, and economic balance indicators, and the fusion data and the key indicators are combined according to a predetermined weight to form a state vector.

[0011] Preferably, the deep reinforcement learning model adopts a multi-task transfer learning strategy, uses the state vector as input to generate candidate solutions, and the candidate solutions clearly specify the level difficulty, currency circulation, and character balance parameters.

[0012] Preferably, the generative adversarial network adopts a network structure composed of several convolutional layers, trains historical scenario data and current game style data to generate an innovative content output, and uses the feature vectors extracted by a variational autoencoder to enhance the innovative content output. Fuse the candidate solution and the enhanced innovative content through a non-linear weighted fusion method to form integrated update solution data.

[0013] Preferably, the candidate solution and the innovative content are fused through a non-linear weighted fusion method to form an integrated update solution, which specifically includes:

[0014] Fuse the features of the candidate solution and the innovative content by constructing an interaction matrix and applying a non-linear activation function to form an integrated update solution.

[0015] Preferably, the integrated update solution is deployed on a digital twin simulation platform, and the integrated update solution is simulated and tested through a virtual player model to collect operation data, which specifically includes:

[0016] Implement and deploy the integration and update solution on the digital twin simulation platform using containerization technology, and conduct simulation tests on the integration and update solution based on the virtual player model of Agent-Based Modeling, so as to collect system load, player behavior, economic balance, and BUG generation rate data as operation data.

[0017] Preferably, the verification solution is generated by using statistical analysis and multi-objective optimization, specifically including:

[0018] The operation data is processed by standard statistical analysis and parallel multi-objective optimization based on the population search strategy to generate a verification solution.

[0019] Preferably, the adjustment data and the verification solution data are integrated in a predetermined data format to form a full-scale update solution, and the full-scale update solution is implemented in a rolling deployment manner on the automated deployment platform. The rolling deployment method includes first implementing the update on at least 10% of the production nodes, and gradually expanding to all production nodes after confirming stable operation, and real-time monitoring and collecting player behavior data, system performance data, community sentiment data, and BUG feedback data to form a closed-loop feedback, and the closed-loop feedback triggers dynamic iterative updates at a set time interval.

[0020] A game update system based on game development data adjustment, used to implement the game update method based on game development data adjustment. The system includes:

[0021] A collection and preprocessing module, which is used to collect player behavior data, system performance data, community sentiment data, and BUG feedback data in the game environment, and perform noise filtering, time correction, format standardization, feature extraction, and fusion processing on the data to form fusion data;

[0022] A core generation module, which is used to form a state vector based on the fusion data and preset key indicators, generate a candidate solution using a deep reinforcement learning model, and use a generative adversarial network to train and generate innovative content from historical scenario data and current game style data, and then fuse the candidate solution and the innovative content through a non-linear weighted fusion method to form an integration and update solution;

[0023] A simulation and adjustment module, which is used to deploy the integration and update solution on the digital twin simulation platform, conduct simulation tests on the integration and update solution using a virtual player model, collect operation data, generate a verification solution by using statistical analysis and multi-objective optimization, and automatically trigger adaptive parameter adjustment based on real-time feedback to form adjustment data;

[0024] A deployment feedback module is used to integrate the adjustment data and the verification scheme data according to a predetermined data format to generate a full-scale update scheme, implement rolling deployment of the full-scale update scheme through an automated deployment platform, and form a closed-loop feedback using real-time monitoring data to periodically trigger dynamic iterative updates.

[0025] A storage medium stores a computer program which, when executed by a processor, implements the steps of the game update method based on game development data adjustment.

[0026] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0027] The present invention realizes the automatic generation of candidate update schemes based on state vectors by introducing deep reinforcement learning technology;

[0028] The present invention realizes the training of innovative content for historical scenarios and current game styles by adopting generative adversarial network technology, thus enriching the creative dimension of the update scheme;

[0029] The present invention realizes the deep feature integration of candidate schemes and innovative content by using non-linear weighted fusion technology, thus forming an integrated update scheme with high decision-making accuracy and adaptive adjustment ability;

[0030] The present invention realizes simulation testing and real-time feedback closed-loop control by using a digital twin simulation platform and a virtual player model, thus automatically triggering adaptive parameter adjustment and forming the effect of dynamic iterative updates.

[0031] The comprehensive application of the above technical means solves the defects in the prior art such as manual updates, large amounts of code adjustment, imperfect BUG handling, and insufficient real-time performance. It not only realizes the dynamic iteration of intelligent decision-making closed-loop updates, but also effectively improves the game update efficiency and dynamic response ability, and further improves the overall game performance and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flowchart of the method of the present invention;

[0033] Figure 2 is a flowchart of forming an integrated update scheme in the present invention;

[0034] Figure 3 is a schematic diagram of simulation testing and adaptive adjustment in the present invention;

[0035] Figure 4 is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0036] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments can be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present disclosure.

[0037] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0038] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0039] As Figure 1 shown, a game update method based on game development data adjustment includes the following steps:

[0040] Collect player behavior data, system performance data, community sentiment data, and BUG feedback data from the game environment. After preprocessing all the data, perform feature extraction and fusion processing to form fusion data;

[0041] The present invention aims to capture the game running state in real time and provide reliable data support for subsequent intelligent update decisions by implementing a data-driven update mechanism in the game development environment. To this end, the system first collects player behavior data, system performance data, community sentiment data, and BUG feedback data in the game environment. Specifically, player behavior data includes click records, movement trajectories, operation paths, and residence durations generated by players during the game; system performance data covers server response time, memory usage, CPU utilization, network latency, etc.; community sentiment data mainly comes from game forums, social media, and internal feedback, and its content involves player comments, emotion identifiers, and related multimedia information; BUG feedback data captures system exceptions and fault logs through a preset error reporting mechanism.

[0042] All kinds of collected data are transmitted in a unified format through a standardized interface, and preliminary processing is carried out on the player behavior data, system performance data, community sentiment data, and BUG feedback data collected from the game environment. Specifically, for player behavior data, a time alignment algorithm is adopted and combined with a time synchronization protocol to convert data in different time zones into a unified standard time format to ensure the consistency of data timestamps; for system performance data, an outlier detection method based on statistics (such as mean, standard deviation) is used to remove outliers and smooth the data, and at the same time, each monitoring index is normalized to ensure that numerical data can be compared on a unified scale; for community sentiment data, a pre-trained natural language processing model is used to tokenize, sentiment score, and extract keywords from the text data, and the sentiment scores are normalized to achieve numerical consistency between different text data; for BUG feedback data, the system parses the fixed format of the error log, extracts key attributes (such as error code, error description, and timestamp), and converts them into a standard format compatible with other data.

[0043] After preprocessing, the data is stored in a standard format, and machine learning algorithms (such as convolutional neural networks, recurrent neural networks, or Transformer models) are used to extract high-dimensional feature representations. Finally, according to the predetermined rules, the features from each source are weighted and fused to generate a unified fused data vector. Through the above processing flow, the data is effectively unified in terms of time and value, providing a reliable and unified data foundation for the subsequent construction of the state vector and intelligent decision-making.

[0044] After the above preprocessing, the data is further subjected to feature extraction. The system uses convolutional neural networks (CNNs), recurrent neural networks (RNNs), and Transformer models to extract high-dimensional feature representations of each data source respectively. After the extracted feature representations are normalized, they are fused through the multi-head attention mechanism based on Transformer to dynamically adjust the weights of the features from each data source, forming a fused data that comprehensively describes the game running state. This fused data not only retains the temporal information but also integrates multi-modal features such as text, images, and numerical values, and can comprehensively reflect player behavior, system operation, user emotions, and fault conditions. The entire process of data collection, preprocessing, and fusion, while ensuring data consistency and real-time performance, realizes seamless docking with subsequent modules through a standardized data interface (such as using RESTful API), providing a solid foundation for intelligent updates based on game development data. Through this overall process, the system can obtain a large amount of multi-dimensional data in a short time and preprocess and feature fusion in an efficient and accurate manner, thus laying a data foundation for the next state evaluation and intelligent decision-making, and at the same time providing data guarantee for the real-time response and dynamic update of the system.

[0045] Such asFigure 2 As shown, a state vector is constructed based on the fusion data and preset key indicators, and a candidate solution is generated using a deep reinforcement learning model. At the same time, a generative adversarial network is used to train historical scenario data and current game style data to generate innovative content, and the candidate solution and the innovative content are fused through a non-linear weighted fusion method to form an integrated update solution.

[0046] The update method based on game development data adjustment in the present invention aims to achieve intelligent dynamic update of the game system through data-driven. The overall solution first collects information from multiple data sources in the game environment, and these data sources include player behavior data, system performance data, community sentiment data, and BUG feedback data. After the collected data is processed by the preprocessing module for noise filtering, time correction, format standardization, and feature extraction, respective feature representations are generated and integrated into a unified fusion data vector through a multi-head attention fusion algorithm. The fusion data vector fully reflects the information of each dimension in the game running state, including both the interaction behavior of players in the game and the system operation indicators and player feedback emotions, etc.

[0047] On this basis, the system combines the fusion data with preset key indicators (these key indicators are defined as player retention rate, system response indicator, and economic balance indicator in the preferred embodiment) according to a predetermined weight to form a state vector. The state vector, as the input of the deep reinforcement learning model, is an important data expression describing the overall state of the current game. Based on the constructed state vector, the model generates a candidate update solution through deep reinforcement learning technology. The candidate solution mainly gives suggestions for adjusting key parameters in the game, such as changes in level difficulty, regulation of currency circulation, and optimization of character balance parameters.

[0048] At the same time, the system uses a generative adversarial network (GAN) to train historical scenario data and current game style data to generate content output with innovative features. Here, the historical scenario data provides visual information of previous scenarios in the game, while the current game style data reflects the current design style of the game. Through the joint training of these two types of data, the innovative content output by the generative adversarial network can capture and display new trends in the game visually and emotionally. The generated innovative content and the candidate update solution are then integrated through a non-linear weighted fusion method. This fusion method forms an interaction matrix and applies a non-linear activation function to deeply fuse the feature information in the candidate solution and the innovative content, thereby forming an integrated update solution that integrates intelligent decision-making and creative output.

[0049] The integrated update solution combines the innovative advantages of data-driven intelligent decision-making and visual creativity, enabling game updates to not only be scientifically adjusted in numerical parameters but also achieve breakthroughs in game content and experience. The entire process, from data collection, preprocessing, feature fusion, state vector construction, and generating candidate solutions through deep reinforcement learning, to generating innovative content using a generative adversarial network, and then non-linearly weighted fusing the candidate solutions with the innovative content, forms a complete data-driven closed-loop update link. Through real-time data collection and efficient fusion, the present invention provides an intelligent update mechanism for the game system based on real-time data feedback, which helps improve the user experience and system stability of the game, and at the same time provides a solid foundation for subsequent automated deployment and dynamic iterative updates. This update method can effectively reduce decision-making biases caused by data latency and information asymmetry in practical applications, and achieve adaptive optimization of the game system under multi-dimensional indicators, with strong operability and promotion value.

[0050] Preferably, the preset key indicators are the player retention rate, system response indicator, and economic balance indicator, and the fused data and the key indicators are combined according to a predetermined weight to form a state vector.

[0051] In the present invention, the preferred settings of the preset key indicators are the player retention rate, system response indicator, and economic balance indicator, and these indicators and the fused data are combined according to a predetermined weight to form a state vector. The player retention rate reflects the attractiveness and continuous participation of the game to players, and its calculation formula is usually expressed as:

[0052]

[0053] where, R represents the player retention rate, N active represents the number of active players within the set observation period, and N init represents the number of players who initially registered or first participated in the game. This indicator dynamically reflects the mobility of the player group through daily or weekly data statistics, providing important feedback data for game operation.

[0054] The system response indicator mainly reflects the stability and performance level of the server and network in the face of high-concurrency requests. This indicator not only involves the response time but also covers multiple parameters such as server CPU utilization rate, memory occupancy rate, and network latency. The response time can be obtained by calculating the time difference between the request sending and response receiving, and its calculation expression is:

[0055] T = t resp -t req

[0056] where, T represents the response time, t resp and t reqThey represent the time of request sending and response receiving respectively. By monitoring and statistically analyzing these parameters in real time, the system can give early warnings in a timely manner when performance fluctuations occur and assist in decision-making for updates.

[0057] The economic balance indicator is used to monitor the health of the in-game virtual economy, and its core lies in evaluating the balance relationship between the circulation of virtual currency and player consumption. Specifically, the system calculates the economic balance ratio by statistically analyzing the total supply, circulation speed of virtual currency in the game, and player consumption data, and preset reasonable thresholds to trigger economic regulation strategies when the values are abnormal. The above key indicators are all normalized to ensure the consistency of data on the numerical scale, and then weighted and fused through predetermined weights.

[0058] After normalizing the fused data vector F and the key indicator vector K respectively, the system adopts the weighted fusion formula:

[0059] S = w f ·F′ + w k ·K

[0060] where F′ and K′ are the fused data and key indicator vectors after normalization respectively, and w f and w k are predetermined weight coefficients, satisfying the condition of w f + w k = 1.

[0061] In actual operation, the initial values of the weights w f and w k can be set to 0.7 and 0.3, but through methods such as historical data backtesting, cross-validation, and Bayesian optimization, the system can dynamically adjust these weight values to make the state vector more accurately reflect the running state of the current game system. By precisely selecting and calculating the preset key indicators, the system ensures that the acquisition, normalization, and weighting processes of each indicator have clear technical implementation steps, which can not only fully capture the changes in various operation data in the game but also provide scientific and accurate input information for the subsequent deep reinforcement learning model. The data interface and message transmission mechanism in this process adopt standardized protocols to ensure seamless data transfer between each processing module and achieve real-time data feedback. Through strict quantification and optimized design of the key indicators, the system can effectively improve the accuracy and reliability of data fusion when constructing the state vector, provide solid data support for the intelligent generation of game update plans, and play a warning and regulation role in actual operation to ensure that the game system meets the expected goals in terms of user experience, performance stability, and economic health.

[0062] Preferably, the deep reinforcement learning model adopts a multi-task transfer learning strategy, uses the state vector as input to generate candidate solutions, and the candidate solutions clearly specify the level difficulty, currency circulation volume, and character balance parameters.

[0063] In the present invention, the deep reinforcement learning model is used as the key technology for generating candidate solutions, and its specific implementation adopts a multi-task transfer learning strategy. The system first uses the previously constructed state vector S as the input, which integrates the fusion data and key metric information and comprehensively reflects the game running status. Based on this, the deep reinforcement learning model generates a series of candidate update solutions through methods such as policy evaluation and value function approximation. The candidate solutions mainly involve the adjustment of the level difficulty, currency circulation, and character balance parameters in the game.

[0064] In the actual implementation, the deep reinforcement learning model adopts a deep Q-network (DQN) or an Actor-Critic structure, and its network structure usually includes multiple fully connected layers and convolutional layers for processing the high-dimensional data in the state vector. The training objective of the model is to minimize the preset loss function L, which can be expressed as:

[0065]

[0066] where Q(S, a; θ) is the expected return of the current network for taking action a in state S under parameter θ, γ is the discount factor, r represents the immediate reward, S′ is the next state, and θ′ is the target network parameter. Through continuous iterative optimization, this formula enables the model to accurately predict the long-term benefits brought by each update solution in the complex state space.

[0067] The adoption of the multi-task transfer learning strategy aims to utilize the common information between different game subsystems to achieve cross-task knowledge sharing. For example, in the three tasks of level difficulty, currency circulation, and character balance, although the specific control objectives are different, they are all affected by player behavior data and system performance data. By sharing some network layers, the model can transfer the existing feature representations between tasks, thereby improving the overall learning efficiency and decision-making accuracy. In the embodiment, the model can adopt a shared convolutional layer to extract common features and then predict the parameters of each task in the task-specific layer. Specifically, during the training process, each task generates a separate loss function, and the system sums up the weighted losses of each task as the overall loss and jointly updates the weights of the shared layer through backpropagation to ensure the balanced performance of the model on different tasks.

[0068] The output results after generating candidate solutions clearly define the specific recommended values of each key game parameter. For example, the level difficulty is adjusted to 1.1 times the original value, the currency circulation is increased by 15% or decreased by 10%, and the character balance parameter is fine-tuned according to the player behavior distribution. After being optimized by the deep reinforcement learning model, these candidate solutions can provide a quantitative decision-making basis for game updates. The entire process realizes seamless connection of input and output through a standardized data interface, ensuring that the model has a high response speed and decision-making accuracy under real-time data updates. Through experimental verification, the deep reinforcement learning model shows high robustness and convergence speed in a multi-task environment, and can generate targeted and operable candidate solutions in different operation scenarios, thus providing a solid data and strategy foundation for the overall generation of subsequent update plans.

[0069] Preferably, the generative adversarial network adopts a network structure composed of several convolutional layers, trains the historical scenario data and the current game style data to generate an innovative content output, enhances the innovative content output by using the feature vectors extracted by the variational autoencoder, and fuses the candidate solution with the enhanced innovative content through a non-linear weighted fusion method to form the integrated update plan data.

[0070] The generative adversarial network (GAN) is used in the present invention to generate innovative content. Its design adopts a network structure composed of multiple convolutional layers. By training the historical scenario data and the current game style data, it outputs new content that can reflect the game's visual and emotional creativity. Specifically, the GAN model includes two parts: a generator and a discriminator. The generator is composed of several convolutional layers, deconvolutional layers and normalization layers, and its purpose is to map the random noise and conditional vectors to the image space matching the game scenario; the discriminator is composed of multiple convolutional layers and fully connected layers, and is used to distinguish the generated image from the real image. During the training process, the generator and the discriminator are continuously iterated through adversarial training. The goal of the generator is to make the discriminator unable to distinguish the generated image from the real image, so as to achieve the goal of generating high-quality innovative content.

[0071] Specifically, the input of the generator includes a random noise vector z and a current game style conditional vector c; the generator generates an image output G(z, c) through multi-layer convolutional operations. The discriminator receives the real historical scenario image x or the generated image G(z, c) as input, and calculates the discrimination probability through multi-layer convolution and activation functions. The training objective is to minimize the generator loss function L G and maximize the discriminator loss function L D , and its standard adversarial training objective can be expressed as:

[0072]

[0073] where p data(x) represents the true image data distribution, p(z) represents the random noise distribution, and p(c) represents the conditional vector distribution. Through this training process, the generator can generate new images with game scene features and matching the current game style.

[0074] To further integrate the innovative content of the generative adversarial network with the candidate solutions generated by deep reinforcement learning, the present invention uses a variational autoencoder (VAE) to extract innovative features. The VAE model compresses the input image into a low-dimensional latent vector z through an encoder and then reconstructs the image by a decoder, thereby learning the key features of the image. The extracted feature vector and the innovative content output by the GAN are concatenated or weighted and fused through a feature fusion layer to form a unified description of the innovative content. The fusion process uses simple concatenation or weighted averaging, supplemented by a non-linear activation function (such as ReLU), to ensure that the fusion result has a richer semantic expression.

[0075] In practical applications, the innovative content output by the generative adversarial network not only provides visual novelty but also plays an important role in inspiring creativity and optimizing the user experience in game update decisions. In the embodiment, by combining the generated images with the candidate solution parameters, the system can achieve innovative updates of game scenes and interaction methods, thereby enhancing the attractiveness of the game and the player experience. The whole process is seamlessly connected to the subsequent update decision module through a standardized data interface and continuously optimizes the model parameters through a large amount of training data to ensure that the generated content has high stability and diversity in different scenarios.

[0076] In summary, the present invention combines the generative adversarial network and the variational autoencoder, specifically trains the historical scene and current game style data through a multi-layer convolutional network to generate innovative content, and integrates it with the candidate update solution through a feature fusion method, providing an innovative content output that combines data-driven and creative expression for game updates, effectively enhancing the intelligence and foresight of game update solutions.

[0077] Preferably, the candidate solution and the innovative content are fused by a non-linear weighted fusion method to form an integrated update solution, specifically including:

[0078] The features of the candidate solution and the innovative content are fused by constructing an interaction matrix and applying a non-linear activation function to form an integrated update solution.

[0079] The non - linear weighted fusion of candidate solutions and innovative content is one of the core technologies of the present invention. Its purpose is to effectively integrate the candidate solutions generated by the deep reinforcement learning model and the innovative content output by the generative adversarial network to form a comprehensive and creative integrated update solution. This fusion process uses the method of constructing an interaction matrix and applying a non - linear activation function to deeply interact the features of the candidate solutions and the innovative content. Specifically, let the candidate solution feature vector be C and the innovative content feature vector be I. The system first normalizes C and I to ensure that their numerical ranges are the same. Subsequently, an interaction matrix M is constructed to capture the interaction relationship between C and I. The calculation formula of the interaction matrix can be expressed as:

[0080] ″T

[0081] M = C′×I

[0082] where C′ and I′ are the normalized candidate solution feature vector and innovative content feature vector respectively, × represents matrix multiplication, and T represents matrix transpose operation. After constructing the interaction matrix, the system uses a non - linear activation function (such as ReLU or Swish function) to activate matrix M. The activated matrix M act represents the non - linear interaction result of the candidate solution and the innovative content on specific features.

[0083] Subsequently, the system weights and fuses M act with the original candidate solution feature vector C and the innovative content feature vector I. The fusion formula can be expressed as:

[0084] U = α·C + β·I + γ·M act

[0085] where U represents the final integrated update solution feature vector, α, β, and γ are predetermined weight coefficients, and satisfy the normalization condition of α + β + γ = 1. Through the above - mentioned weighted fusion, the system can retain the respective advantages when integrating the candidate solution and the innovative content, and supplement the complementary information between each other through the interaction matrix, so as to form an integrated update solution that combines data - driven and creative innovation.

[0086] When implementing this fusion process, strict control must be exercised over the data format and numerical precision in each link of collecting, normalizing, matrix constructing, and weighted calculating the candidate solutions and the characteristics of the innovative content. For example, in specific operations, if the dimension of the candidate solution feature vector C is n and the dimension of the innovative content feature vector I is m, then the size of the interaction matrix M is n×m; the non-linear activation function is applied to each element to ensure that all numerical values are positive, thus avoiding the influence of negative values on the fusion effect. This process has been backtested and verified with a large amount of data in the laboratory through a preset algorithm, proving that the integrated update solution after fusion has high accuracy and stability in predicting the game running effect and optimizing the user experience.

[0087] Finally, the integrated update solution is output in the form of a high-dimensional feature vector and transmitted to the subsequent deployment module through a unified data interface to realize the application of the update solution in the actual game environment. The advantage of the entire non-linear weighted fusion method is that it not only realizes the simple superposition of the candidate solution and the innovative content, but also captures the deep-seated correlation information between the two by constructing an interaction matrix, thereby generating a more creative and targeted update solution, providing a strong technical support for the dynamic update of the game system.

[0088] As Figure 3 shown, the integrated update solution is deployed on the digital twin simulation platform, the integrated update solution is simulated and tested through a virtual player model, the running data is collected, and a verification solution is generated by using statistical analysis and multi-objective optimization. At the same time, the adaptive parameter adjustment is automatically triggered according to the real-time feedback to form adjustment data;

[0089] The present invention proposes an overall method of deploying an integrated update solution on a digital twin simulation platform for how to utilize real-time data to drive intelligent decision-making during the game update process. The present invention first preprocesses and fuses multi-source data generated in the game environment to form a highly comprehensive fused data vector. Then, the fused data and the preset key indicators together constitute a state vector, which serves as the basic input for generating the subsequent intelligent update solution. Next, a deep reinforcement learning model is used to generate candidate update solutions, and a generative adversarial network (GAN) is used to train the historical scenario data and the current game style data, and output content with innovative features. The candidate solution and the innovative content are non-linearly weighted and fused to achieve the deep interaction between the two at the feature level, thereby generating an integrated update solution that integrates data-driven decision-making and visual creativity.

[0090] After that, the integrated update solution is deployed in the digital twin simulation platform. This platform simulates the operating state in the real game environment through virtual player models and collects operation data including system load, player behavior, economic balance, and BUG generation rate. The collected operation data is subjected to statistical analysis and parallel multi-objective optimization processing based on a population search strategy to generate a verification solution for evaluating and verifying the effect of the update solution. At the same time, the system automatically triggers adaptive parameter adjustment based on real-time feedback data to form the final adjustment data. The overall method constructs a complete closed-loop mechanism through data collection, fusion, intelligent decision-making, simulation testing, and feedback adaptive adjustment, which not only realizes the automatic generation and intelligent optimization of the game update solution but also provides a scientific basis and real-time data support for the dynamic iterative update of the system. The present invention makes full use of the advantages of multi-source data and uses core technical means such as advanced deep reinforcement learning, generative adversarial network, non-linear weighted fusion, and parallel multi-objective optimization to ensure the adaptability and accuracy of the update solution in different operating environments, thereby effectively improving the overall performance and user experience of the game system. Through the present invention, game developers can capture key indicators of the game operating state in real time, achieve dynamic regulation based on data feedback, avoid the failure of strategies caused by data delay and information asymmetry in traditional updates, and realize the continuous and stable operation of the system under high load and complex environments.

[0091] Preferably, the integrated update solution is deployed on the digital twin simulation platform, and the integrated update solution is subjected to simulation testing through virtual player models, and operation data is collected, specifically including:

[0092] The integrated update solution is implemented and deployed on the digital twin simulation platform by using containerization technology, and the integrated update solution is subjected to simulation testing based on the virtual player model of Agent-Based Modeling, so as to collect system load, player behavior, economic balance, and BUG generation rate data as operation data.

[0093] In the present invention, when the integrated update solution is deployed on the digital twin simulation platform, containerization technology and a virtual player model based on Agent-Based Modeling are adopted to ensure the consistency and high fidelity of the simulation environment with the actual production environment. Specifically, the digital twin simulation platform uses containerization technologies such as Docker to deploy various services, and through the Kubernetes scheduling mechanism, automatic resource allocation and load balancing are carried out, so as to ensure that the system can operate stably under high concurrency and resource constraints. A virtual player model is built inside the platform. This model is based on the principle of Agent-Based Modeling (ABM) and simulates the behavior of players. The virtual player model simulates the behavior decision-making process, path selection, interaction patterns and emotional fluctuations of players in the game, so that the reaction of the system in the real game scenario can be comprehensively captured during the simulation test. During the simulation test, the data collected by the system includes but is not limited to the system load of the server, CPU utilization rate, memory occupancy rate, network latency, as well as the behavior patterns, residence time, task completion rate and interaction frequency of players in the game. At the same time, the circulation of virtual currency and the BUG generation rate in the economic system are recorded. All these data are stored in a structured format and uploaded to the data processing center in real time.

[0094] To ensure the accuracy and real-time of data collection, the digital twin simulation platform adopts high-frequency data collection technology and time-series database storage method. Each virtual player model collects key data at a preset time interval (such as every 30 seconds or every minute). The data transmission is realized by using the standard RESTful API and Kafka message queue for low-latency transmission. The containerization technology enables the platform to have good scalability and fault tolerance. Once a certain node fails, its service can be automatically taken over by other nodes to ensure uninterrupted data collection. At the same time, by configuring the load balancer, the system can dynamically adjust the task allocation of each node to ensure that the collected data is evenly distributed. The implementation of containerization technology not only improves the deployment efficiency but also reduces the system maintenance cost, providing a solid foundation for subsequent simulation tests and data analysis. In this way, the digital twin simulation platform can accurately simulate the game environment and generate data for subsequent verification of the solution with high fidelity, thus providing data support for the scientificity and reliability of the overall update solution. The implementation details of this platform include the image construction of each container, network configuration, log collection and monitoring mechanism, as well as the setting of behavior parameters and the definition of simulation rules of the virtual player model. All these have been verified and debugged many times to ensure that they can truly reflect the game operation state in actual application, so as to provide accurate and timely data feedback for game update decisions.

[0095] Preferably, the verification solution is generated by using statistical analysis and multi-objective optimization, specifically including:

[0096] The operation data is processed by standard statistical analysis and parallel multi-objective optimization based on a population search strategy to generate a verification plan.

[0097] In the present invention, the collected operation data is processed by standard statistical analysis and parallel multi-objective optimization based on a population search strategy, and is used to generate a verification plan, and further automatically triggers adaptive parameter adjustment according to real-time feedback to form adjustment data. Specifically, the operation data collected by the system from the digital twin platform after simulation testing includes server system load, player behavior data, economic balance indicators, and BUG generation rate data. These data enter the data analysis module after preprocessing. The data analysis module first uses conventional statistical methods to calculate statistics such as mean, variance, and standard deviation for each item of data, so as to obtain the basic distribution characteristics of the data. To further extract the deep information in the data, the system uses a population search strategy (such as particle swarm optimization or genetic algorithm) to process the data under a parallel multi-objective optimization framework. The parallel multi-objective optimization algorithm takes multiple indicators as objectives and constructs an objective function, and the form of the objective function is:

[0098]

[0099] where N represents the number of monitoring indicators, Error i represents the deviation of the i-th indicator, and α i is a preset weight. Through the population search strategy, the system continuously iteratively screens among a large number of candidate solutions, and finally selects a verification plan that minimizes the objective function. This verification plan reflects the operation effect of the current integration and update plan in the simulation environment, including indicators such as predicted system stability, player satisfaction, and economic balance.

[0100] After the verification plan is generated, the system further adaptively adjusts the verification plan according to real-time feedback data. The real-time feedback data is continuously collected by the monitoring system in the deployment stage and transmitted to the adaptive adjustment module through a data interface. This module uses preset rules to automatically trigger parameter adjustment when it detects that a certain indicator exceeds the preset threshold. The parameter adjustment algorithm constructs a new objective function based on the feedback data and re-optimizes the parameters of the update plan through the gradient descent method or the evolutionary algorithm. For example, when the system monitors that the server response time T exceeds the preset upper limit T max , the adjustment module will immediately calculate the response time deviation ΔT = T - T max and reallocate the system resource parameters according to this deviation, such as adjusting the task scheduling priority or reconfiguring the load balancing weight. During the whole process, the system records the parameter changes before and after the adjustment and the corresponding feedback effects, and incorporates the adjustment results into the overall data analysis again through a closed-loop feedback mechanism to achieve dynamic iterative update.

[0101] In addition, strict regulations are imposed on the algorithm parameters of statistical analysis and multi-objective optimization during the data processing to ensure that each operation has a clear technical implementation plan. All calculations are carried out through a standardized data interface, and the data transmission and processing in each link adopt a unified data format (such as JSON format) to ensure the consistency and reliability of the data. After the above process, the system can generate a verification plan reflecting the current operating status and prediction effect, and automatically optimize and update the plan according to the real-time data feedback, finally forming adjusted data, which provides an accurate basis for the generation of the full-scale update plan. This process not only improves the scientific nature of the update plan but also verifies the efficiency and self-adaptability of the data-driven decision-making model in practical applications, laying a solid foundation for the intelligent and dynamic update of the game system.

[0102] After generating the integrated update plan, the present invention conducts simulation tests on the present invention through a digital twin simulation platform and automatically triggers adaptive parameter adjustment based on the real-time feedback data, thereby generating the final adjusted data. The deployment of the integrated update plan on the digital twin platform adopts a standardized operation process to ensure that the operation of the plan in the simulation environment is highly consistent with the real production environment. During the deployment process, the system uses containerization technology to package the integrated update plan into a standard container image and realizes automatic scheduling and elastic expansion through Kubernetes. After the deployment is completed, the platform simulates the actual player behavior in the game through a virtual player model based on Agent-Based Modeling, covering various interaction scenarios, task execution, and economic transaction processes. The parameter settings of the virtual player model are calibrated according to historical player behavior data and real-time collected behavior statistical data to ensure that its simulation effect is representative.

[0103] During the simulation test process, the system collects various operation data in real time, including but not limited to server system load (such as CPU utilization rate, memory occupancy rate, and network latency), player behavior (such as click-through rate, stay duration, and task completion rate), economic balance data (such as virtual currency circulation and consumption behavior), and BUG generation rate. The collected data is uniformly stored by the data storage module and transmitted to the data analysis module through the real-time data interface. The data analysis module uses standard statistical methods to calculate statistics such as mean, variance, and percentile of the operation data, and uses a parallel multi-objective optimization algorithm based on a population search strategy to jointly optimize each index. By constructing a multi-objective optimization problem, the system sets the objective function to minimize the deviation of each index, ensuring that the verification plan can comprehensively reflect the overall performance of the integrated update plan in the simulation environment.

[0104] After the verification plan is generated, the system implements adaptive parameter adjustment based on real-time feedback data. This adaptive module calculates the parameter adjustment amount automatically by comparing the differences between the real-time data and the verification plan, and immediately triggers an update. For example, when the real-time monitoring data indicates that the player task completion rate decreases and the BUG generation rate increases, the adaptive module will analyze the system performance data, calculate various deviation values, and re-optimize the parameter settings of the candidate update plan through a preset adjustment strategy. The adaptive parameter adjustment algorithm can use the gradient descent method to solve the objective function, and its specific calculation process includes:

[0105]

[0106] where θ represents the parameter vector in the integrated update plan, η is the learning rate, is the gradient of the objective function. In this way, the system can automatically correct inappropriate parameters in the update plan during operation, so as to generate the final adjustment data.

[0107] The entire simulation test and adaptive adjustment process realizes seamless data transmission and real-time processing through a unified data interface and message queue, ensuring the consistency and reliability of data among various processing modules. The system can not only pre-verify the effect of the integrated update plan in the simulation environment, but also achieve closed-loop control through real-time monitoring feedback, so as to continuously optimize and iterate the update plan, providing the game system with the ability of dynamic adjustment and intelligent optimization. Through this series of strict technical processes, the finally generated adjustment data fully reflects the real-time changes in the game operation state, providing a scientific basis and reliable support for the final deployment of the full-scale update plan.

[0108] Integrate the adjustment data and the verification plan data according to a predetermined data format to generate a full-scale update plan, implement rolling deployment of the full-scale update plan through an automated deployment platform, and use real-time monitoring data to form a closed-loop feedback to periodically trigger dynamic iterative updates.

[0109] The present invention proposes an innovative method for how to utilize multi-source data to construct an intelligent decision-making closed loop during the game update process. With data-driven as the core, the present invention integrates multi-dimensional data that has undergone preliminary preprocessing, feature extraction, and fusion, as well as adjusted data and verification scheme data formed through further processing, in accordance with a predetermined data format to generate a full-scale update plan. The full-scale update plan is then implemented through a rolling deployment on an automated deployment platform, and a closed-loop feedback mechanism is constructed using real-time monitoring data to periodically trigger dynamic iterative updates, thereby achieving intelligent adaptive regulation and continuous optimization of the game system. The overall solution realizes the whole-process closed-loop control from data collection, preprocessing, feature fusion to state assessment, candidate plan generation, simulation verification, automatic adjustment, and full-scale deployment. First, in the data collection stage, the system obtains data such as player behavior, system performance, community sentiment, and BUG feedback in real time from the game environment to ensure that each data source can timely reflect the game operation state in terms of time and space. Next, after the preprocessing module filters out noise, corrects time, and standardizes the format of this data, the key information of each data source is extracted through the feature extraction module, and then the data from each source is integrated into unified fusion data using the multi-head attention fusion technology. This fusion data serves as the basis for subsequent construction of the state vector and generation of the intelligent update plan, and its features include both user behavior patterns, system load conditions, as well as sentiment tendencies and fault frequencies, etc.

[0110] On this basis, the system combines the integrated data with preset key indicators according to a predetermined weight to form a state vector, which reflects the overall operation state of the game and provides high-quality input for the deep reinforcement learning model to generate candidate update plans. During the generation of candidate plans, the system uses the deep reinforcement learning model and the generative adversarial network to generate candidate plans and innovative content respectively, and adopts a non-linear weighted fusion method to integrate the two, forming an integrated update plan that combines numerical regulation and creative expression. Subsequently, the integrated update plan is deployed on the digital twin simulation platform, which uses containerization technology and virtual player models to simulate the real game environment and collect operation data including system load, player behavior, economic balance, and BUG generation rate. After statistical analysis and parallel multi-objective optimization processing of the operation data, a verification plan is generated to scientifically evaluate the actual effect of the update plan. At the same time, the system collects feedback data through the real-time monitoring module, and automatically triggers adaptive parameter adjustment when the monitoring indicators exceed the preset threshold, thereby forming the final adjustment data. Finally, the adjustment data and the verification plan data are integrated in a predetermined data format to generate a full-scale update plan, which is implemented through rolling deployment on the automated deployment platform. At the same time, a closed-loop feedback mechanism is constructed using real-time monitoring data to achieve periodic dynamic iterative updates. The entire process from data collection to full-scale update forms a closed-loop control system, ensuring that the update plan can adapt to the continuous changes in the game environment, realizing the dynamic regulation and continuous optimization of the system, and thus improving the player experience and system stability. The present invention has good operability and scalability in practical applications, can accurately identify various abnormal states during operation, and can timely adjust the update plan to ensure that the game system is always in the best operating state.

[0111] Preferably, the adjustment data and the verification plan data are integrated in a predetermined data format to form a full-scale update plan, and the full-scale update plan is implemented in a rolling deployment manner on the automated deployment platform. The rolling deployment method includes first implementing the update on at least 10% of the production nodes, and gradually expanding to all production nodes after confirming stable operation, and real-time monitoring and collecting player behavior data, system performance data, community sentiment data, and BUG feedback data to form a closed-loop feedback, and the closed-loop feedback triggers dynamic iterative updates at a set time interval.

[0112] In the present invention, the adjustment data and the verification scheme data are integrated in accordance with a predetermined data format using the JSON data format to generate a full update scheme. To ensure the standardization and compatibility of data integration, the system uses JSON as the data exchange format because the JSON format has characteristics such as clear structure, easy parsing, and strong cross-platform compatibility. During the specific operation process, the system encodes the adjustment data generated by the adaptive adjustment module and the verification scheme data generated through statistical analysis and parallel multi-objective optimization processing in the standard JSON format respectively. Each data item is expressed in the form of a key-value pair to ensure that the field names and data types are consistent among different modules. For example, the key parameters in the adjustment data such as "difficulty_adjustment", "currency_flow_adjustment", and "balance_parameter" are recorded in numerical form, while the monitoring metrics in the verification scheme data such as "cpu_usage", "memory_usage", "latency", "player_engagement", and "bug_rate" are stored in floating-point or integer form. When the system integrates, it combines the two JSON data objects into a unified JSON object through a predetermined data interface call. The integration process can be expressed by the following pseudocode:

[0113] JSON_Adjustment = {"difficulty_adjustment":X,"currency_flow_adjustment":Y,"balance_parameter":Z,...}

[0114] JSON_Verification = {"cpu_usage":A,"memory_usage":B,"latency":C,"player_engagement":D,"bug_rate":E,...}

[0115] Full_Update_Scheme = merge(JSON_Adjustment,JSON_Verification)

[0116] Among them, the merge function preserves or renames the same fields according to predefined rules to ensure the integrity and unified format of the JSON object data after integration. The full-scale update plan data after integration not only includes the real-time parameters after adaptive adjustment, but also includes various monitoring index data for verifying the update effect, ensuring the compatibility between the data input and the subsequent processing modules. To ensure the stability and security of data transmission, the system uses the standard HTTPS protocol and RESTful API for data exchange. At the same time, to ensure data consistency, the system performs data verification during the integration process, checks whether the required fields in each JSON object are complete, and validates the data format. After the above steps, the generated full-scale update plan data is stored in the data warehouse in JSON format and used as the input data for the subsequent automated deployment module to further achieve full-process automated updates. The advantage of this data integration method is that the standardized data format facilitates cross-module data calls and can efficiently transmit complex multi-dimensional data, making the full-scale update plan have high data consistency and traceability during implementation, providing accurate and real-time data support for game update decisions. In this way, developers can quickly locate the adjustment requirements of various parameters in the system and achieve refined control at the data level to ensure the efficient operation and scientific nature of the entire update process.

[0117] In the present invention, the deployment of the full-scale update plan adopts the rolling deployment method implemented by an automated deployment platform. The purpose is to gradually promote the update plan to all production nodes on the premise of ensuring the stable operation of the system, thereby reducing the update risk and achieving a smooth transition. The rolling deployment method first implements the update on at least 10% of the production nodes. In this stage, the system pushes the full-scale update plan data to some nodes through a standardized interface for a small-scale trial run. In specific implementation, the automated deployment platform divides the production environment into several node groups according to a preset strategy, and each node group independently executes the update task. After the update starts, the system monitors the running status of each node group, including key indicators such as server load, response time, error rate, and player feedback. The monitoring data is collected through a real-time monitoring module and fed back to the central data processing unit in JSON format. If no abnormal fluctuations occur in the running indicators of each node group during the trial run stage, the system gradually expands the update scope, for example, from the initial 10% to 30%, and then to 60%, and finally realizes the full-scale node update.

[0118] During the rolling deployment process, the automated deployment platform uses containerization technologies (such as Docker and Kubernetes) to achieve deployment flexibility and fault tolerance. Before each node is updated, a health check is performed to ensure that the node resources (such as memory, CPU, and network bandwidth) meet the update requirements. During the update process, the platform adopts the blue-green deployment or canary release strategy, that is, based on the parallel operation of the old and new versions, it judges whether the performance of the new version is better than that of the old version by real-time monitoring data. Once the new version shows the expected stability and performance improvement on the trial-run nodes, the platform will automatically extend the update to more nodes; otherwise, the system will quickly roll back to the old version to ensure that the entire production environment is not affected.

[0119] To ensure the efficient and reliable data transmission and task scheduling during the rolling deployment process, the system adopts a distributed task scheduling mechanism. Each node receives update commands through the standard RESTful API and realizes the asynchronous processing of tasks through the message queue. The rolling deployment strategy not only greatly reduces the risk of system failures that may be caused by a single full-scale deployment, but also enables developers to monitor and adjust the deployment strategy in real time through phased implementation, further optimizing the update process. During this process, the update status, running metrics, and user feedback of each node are saved in a standardized log record for subsequent data analysis and decision support. Through this rolling deployment method, the system realizes a progressive update from a small-scale pilot to a full-scale promotion, effectively ensuring the smoothness of the update process and the continuity of the system. In the example, if the average response time T of a certain node group is less than the set threshold T th and the error rate E is lower than the preset standard E th , then this node group is considered to be running stably, and the system will then expand the update scope. During the rolling deployment process, the detailed data analysis and monitoring reports of each stage provide data support for the decision-making of the next stage, ensuring that the update plan is fully verified before full-scale promotion, so that the entire update process is both efficient and safe.

[0120] The present invention uses real-time monitoring data to form a closed-loop feedback mechanism to periodically trigger dynamic iterative updates to achieve continuous optimization of the full-scale update plan. After the full-scale deployment, the system uses an automated monitoring module to collect player behavior data, system performance data, and BUG feedback data in real time. These data not only reflect the running status of the current full-scale update plan in the actual production environment, but also provide a basis for the system to detect potential problems. The monitoring data is uploaded to the data center in JSON format through a standardized interface. The data includes server load, response time, error rate, player interaction frequency, and BUG generation rate, etc. The data center statistically analyzes these real-time data and compares them with the preset normal operation thresholds.

[0121] When the system detects that one or more indicators exceed the preset range, the closed-loop feedback mechanism will automatically trigger dynamic iterative updates. In specific operations, the feedback module calculates the deviation of each indicator by constructing an objective function, and the form of the objective function is:

[0122]

[0123] where x i represents the current value of the i-th monitoring indicator, x i,th is the preset threshold of this indicator, α i is the corresponding weight, and N is the total number of monitoring indicators. The value of the objective function L reflects the deviation between the system operation state and the ideal state. When L exceeds the set critical value, the adaptive adjustment module immediately intervenes and automatically adjusts the key parameters in the update plan, such as the level difficulty, resource allocation, and economic balance parameters, through parameter tuning methods based on gradient descent or evolutionary algorithms. The adjusted parameters will be integrated again through the data interface to generate a new update plan, which will be pushed to the production environment in a rolling manner through the automated deployment platform.

[0124] The design of the closed-loop feedback mechanism ensures that the system can achieve adaptive iterative updates driven by real-time monitoring data. The feedback data collection adopts high-frequency data transmission technology to ensure that the system status is automatically checked at preset time intervals (such as every 24 hours or shorter) and the parameters are dynamically adjusted. All data transmissions in the system use the standard HTTPS protocol and RESTful API to ensure data security and no delay. During the feedback process, the monitoring module not only records the current data status but also generates historical data reports for subsequent trend analysis. For example, when the real-time monitoring data shows that the average response time T of players exceeds the set upper limit T max , the system will automatically calculate the response time deviation ΔT = T - T max , and conduct a comprehensive evaluation in combination with other relevant indicators, and then adjust the load balancing strategy or modify the task scheduling parameters to reduce the system response time to within the preset standard. Similarly, when the BUG generation rate B is higher than the preset standard B th , the system will trigger a fault warning, automatically adjust the resource allocation or trigger a patch update to ensure the stable operation of the system.

[0125] Through this closed-loop feedback mechanism, the system can not only capture and respond to abnormal states in the production environment in a timely manner, but also continuously optimize and update the solution during long-term operation, forming an intelligent closed-loop with dynamic iterative updates. The entire feedback process consists of data collection, data transmission, statistical analysis, objective function calculation, and adaptive parameter adjustment. Standardized processes are implemented in each link to ensure data consistency and processing efficiency. This invention has been proven in multiple experiments that its adaptive iterative update can significantly improve the stability and user experience of the game system, while reducing the operation risks caused by system anomalies. The closed-loop feedback mechanism ultimately realizes the full-process closed-loop of the system from data monitoring to dynamic optimization, providing a scientific, accurate, and real-time decision support platform for game updates, thereby promoting the game system to maintain sustainable competitiveness in a changing operation environment.

[0126] As Figure 4 shown, a game update system based on game development data adjustment is used to implement the game update method based on game development data adjustment. The system includes:

[0127] A collection preprocessing module is used to collect player behavior data, system performance data, community sentiment data, and BUG feedback data in the game environment, and perform noise filtering, time correction, format standardization, feature extraction, and fusion processing on the data to form fusion data. This module is usually composed of dedicated sensors, embedded processors, and data acquisition cards in terms of hardware. The sensors are used to capture various data in the game environment in real time, such as player input signals (e.g., key presses, mouse clicks), system status monitoring (CPU, memory, network load sensors), and network interface to collect community feedback and BUG reports. The data acquisition card converts these analog or digital signals into digital data, and then the embedded processor performs preliminary noise filtering, time correction, and format standardization operations. This processor can use high-performance ARM or DSP chips, support real-time data preprocessing, and transmit the data after feature extraction to the subsequent processing unit through a bus (such as PCIe or high-speed Ethernet) to form fusion data.

[0128] The core generation module is used to form a state vector based on the fusion data and preset key indicators, generate candidate solutions using a deep reinforcement learning model, and use a generative adversarial network to train and generate innovative content from historical scenario data and current game style data. Then, the candidate solutions and the innovative content are fused through a non-linear weighted fusion method to form an integrated update solution. In terms of hardware implementation, the core generation module relies on a high-performance computing platform, such as a server equipped with a GPU or a dedicated accelerator (such as a TPU or an FPGA). This platform receives the fusion data transmitted by the preprocessing module and forms a state vector using preset key indicators (stored in preset hardware registers or dedicated memories). Both the deep reinforcement learning model and the generative adversarial network can run on a GPU acceleration card, and parallel computing is used to implement the generation of candidate solutions and the training of innovative content. In hardware implementation, a dedicated deep learning server is often used, which is internally equipped with high-bandwidth memory and high-speed interconnection. Overall, through a non-linear weighted fusion circuit or parallel computing at the software level, an integrated update solution is finally formed.

[0129] The simulation adjustment module is used to deploy the integrated update solution on a digital twin simulation platform, perform simulation tests on the integrated update solution using a virtual player model, collect operation data, generate a verification solution using statistical analysis and multi-objective optimization, and automatically trigger adaptive parameter adjustment based on real-time feedback to form adjustment data. At the hardware level, this module usually relies on a dedicated simulation system of the digital twin platform, which can be a simulation server integrated with a multi-core CPU and a GPU, and is equipped with virtualization technology or a containerization platform to simulate the actual game environment. In this system, a virtual player model is pre-loaded (real-time simulation can be achieved through a dedicated FPGA or GPU) to simulate player behavior and system operation. During the simulation process, hardware sensors and monitoring chips continuously collect operation data (such as system load sensors, temperature sensors, etc.). After the data passes through internal storage and fast caching, a multi-objective optimization algorithm is executed by a built-in statistical analysis chip (or using the computing resources in the server) to generate a verification solution. The real-time feedback is transmitted through a high-speed data interface, and an adaptive adjustment controller (embedded microprocessor) automatically triggers parameter adjustment based on a set algorithm to form the final adjustment data.

[0130] A deployment feedback module is used to integrate the adjustment data and the verification scheme data into a full-scale update scheme according to a predetermined data format, implement rolling deployment of the full-scale update scheme through an automated deployment platform, and form a closed-loop feedback using real-time monitoring data to periodically trigger dynamic iterative updates. The hardware implementation of the deployment feedback module usually depends on the automated deployment platform, which realizes rolling deployment based on distributed server and cluster management technologies. Hardware-wise, this can consist of a standard server cluster, a load balancer, and network monitoring devices. After the adjustment data and the verification scheme data are uniformly formatted (such as JSON), they are transmitted through a high-speed network to the deployment manager, which is responsible for rolling out the update scheme by node. Initially, it is tested on some nodes (for example, 10%), and then gradually expanded. During the deployment process, network monitoring devices (such as SNMP monitors) and dedicated hardware monitoring cards collect data on player behavior, system performance, and BUG feedback in real time. This data is processed in a closed loop by the data center server and fed back to the deployment manager at a preset time interval to trigger the next dynamic iterative update. The entire system ensures low latency and high reliability of data transmission through standardized hardware interfaces and high-speed buses, thus realizing full-process automated updates and dynamic optimization.

[0131] A storage medium stores a computer program, which, when executed by a processor, implements the steps of the game update method based on game development data adjustment.

[0132] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0133] The above are only the embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A game update method based on game development data adjustment, characterized in that, It includes the following steps: Collect player behavior data, system performance data, community sentiment data, and BUG feedback data from the game environment. After preprocessing all the data, perform feature extraction and fusion processing to form fusion data; Construct a state vector based on the fusion data and preset key indicators. Use a deep reinforcement learning model to generate candidate solutions. At the same time, use a generative adversarial network to train historical scenario data and current game style data to generate innovative content, and fuse the candidate solutions and innovative content through a non-linear weighted fusion method to form an integrated update solution; Deploy the integrated update solution on the digital twin simulation platform. Use a virtual player model to conduct simulation tests on the integrated update solution, collect operation data, and generate a verification solution through statistical analysis and multi-objective optimization. At the same time, automatically trigger adaptive parameter adjustment based on real-time feedback to form adjustment data; Integrate the adjustment data with the verification solution data to generate a full-scale update solution. Implement rolling deployment of the full-scale update solution through an automated deployment platform, and use real-time monitoring data to form a closed-loop feedback to periodically trigger dynamic iterative updates.

2. The method according to claim 1, wherein The preset key indicators are player retention rate, system response indicators, and economic balance indicators, and the fusion data and the key indicators are combined according to a predetermined weight to form a state vector.

3. The method according to claim 1, characterized in that The deep reinforcement learning model adopts a multi-task transfer learning strategy, uses the state vector as input to generate candidate solutions, and the candidate solutions clearly specify the level difficulty, currency circulation volume, and character balance parameters.

4. The method according to claim 1, wherein The generative adversarial network adopts a network structure composed of several convolutional layers, trains historical scenario data and current game style data to generate an innovative content output, and uses the feature vector extracted by the variational autoencoder to enhance the innovative content output. Fuse the candidate solutions and the enhanced innovative content through a non-linear weighted fusion method to form integrated update solution data.

5. The method according to claim 1, wherein The process of fusing the candidate solutions and the innovative content through the non-linear weighted fusion method to form an integrated update solution specifically includes: Fuse the features of the candidate solutions and the innovative content by constructing an interaction matrix and applying a non-linear activation function to form an integrated update solution.

6. The method according to claim 1, wherein The process of deploying the integrated update solution on the digital twin simulation platform and using a virtual player model to conduct simulation tests on the integrated update solution and collect operation data specifically includes: Use containerization technology to deploy the integrated update solution on the digital twin simulation platform, and conduct simulation tests on the integrated update solution based on the Agent-Based Modeling virtual player model to collect system load, player behavior, economic balance, and BUG generation rate data as operation data.

7. The method according to claim 1, wherein The process of generating a verification solution through statistical analysis and multi-objective optimization specifically includes: The operation data undergoes standard statistical analysis and parallel multi-objective optimization processing based on a population search strategy to generate a verification solution.

8. The method according to claim 1, wherein The adjusted data and the verification scheme data are integrated in accordance with a predetermined data format to form a full-scale update scheme, which is implemented in a rolling deployment manner on an automated deployment platform. The rolling deployment manner includes first implementing the update on at least 10% of the production nodes, and gradually expanding to all production nodes after confirming stable operation, and real-time monitoring and collecting player behavior data, system performance data, community sentiment data, and BUG feedback data to form a closed-loop feedback, and the closed-loop feedback triggers dynamic iterative updates at a set time interval.

9. A game update system based on game development data adjustment, for implementing the game update method based on game development data adjustment according to any one of claims 1-8, characterized in that, The system includes: A collection and preprocessing module, which is used to collect player behavior data, system performance data, community sentiment data, and BUG feedback data in the game environment, and after preprocessing all the data, perform feature extraction and fusion processing to form fusion data; A core generation module, which is used to construct a state vector based on the fusion data and preset key indicators, generate a candidate scheme using a deep reinforcement learning model, and use a generative adversarial network to train and generate innovative content from historical scenario data and current game style data, and then fuse the candidate scheme with the innovative content through a non-linear weighted fusion method to form an integrated update scheme; A simulation and adjustment module, which is used to deploy the integrated update scheme on a digital twin simulation platform, and use a virtual player model to conduct simulation tests on the integrated update scheme, collect operation data, generate a verification scheme using statistical analysis and multi-objective optimization, and automatically trigger adaptive parameter adjustment based on real-time feedback to form adjusted data; A deployment and feedback module, which is used to integrate the adjusted data and the verification scheme data in accordance with a predetermined data format to generate a full-scale update scheme, implement rolling deployment of the full-scale update scheme through an automated deployment platform, and use real-time monitoring data to form a closed-loop feedback to periodically trigger dynamic iterative updates.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps of the game update method based on game development data adjustment described in any one of claims 1 to 8.

Images