Mobile game dynamic control optimization method based on adaptive deep learning

Abstract

The invention discloses a mobile game dynamic control optimization method based on adaptive deep learning, and the method comprises the following steps: S1, collecting player operation data and scene background parameters, and generating feature data of a unified scale; s2, generating a self-adaptive spatial-temporal feature sequence through the spatial-temporal feature self-adaptive convolutional network; s3, constructing an adaptive memory network, recognizing a recent high-frequency operation mode through a short-term focusing mechanism, and generating operation intention features; s4, generating a self-supervised learning task through a self-supervised learning multi-layer perceptron, and adaptively adjusting the feature weight to generate a control parameter; s5, a feedback score is generated in real time through a feedback scoring mechanism, and the learning rate is adjusted; s6, establishing a satisfaction evaluation mechanism based on feedback scores and data analysis; and S7, recording operation feedback and scoring results in real time, and realizing adaptive optimization based on feedback circulation. According to the method, mobile game control is dynamically optimized through the adaptive deep learning method, and high adaptability and excellent user experience are achieved.

Description

Technical Field

[0001] The present invention relates to the field of adaptive deep learning technology, and in particular to a mobile phone game dynamic control optimization method based on adaptive deep learning. Background Art

[0002] With the rapid development and popularization of mobile games, players' requirements for game operation experience are constantly increasing, especially in game scenes with high complexity and high response speed requirements. Accurate, smooth and personalized control experience has become an important measure of game quality. However, the existing mobile game control optimization technology faces many bottlenecks and technical difficulties in practical applications, and it is difficult to meet the ever-increasing user needs. Traditional mobile game control is mainly based on preset fixed operation parameters and simple reaction speed adjustment strategies, lacking the ability to identify and adapt to individual player operation habits. This method not only lacks intelligence and dynamic adaptability, but also often cannot effectively respond to the diverse needs of different players and different scenarios, making players prone to stiffness, delays, or even mis-touch during operation, thus affecting the game experience.

[0003] Existing control optimization methods generally rely on simple rule-based optimization models or traditional machine learning methods. These methods have obvious deficiencies in adaptability and are difficult to quickly respond to players' dynamic operation needs in real-time game environments. Specifically, traditional rule-based control optimization methods attempt to improve control sensitivity or response speed by adjusting preset parameters, but due to the lack of a deep understanding of players' operating habits, such methods are usually unable to provide personalized operation experience in changing game scenarios. In addition, although traditional machine learning methods have improved the ability to recognize operation features to a certain extent, most of them are based on static data training, and cannot update and optimize models in real time during actual operations, making it difficult to adapt to the operation preferences of different players in real time. Therefore, in the actual game process, the player's operation experience is often unsatisfactory, especially in game scenarios with high-frequency operations, significant changes in touch force, and frequent scene switching. It is difficult for existing technologies to maintain a smooth control effect.

[0004] To solve the above problems, some studies in recent years have introduced deep learning technology to extract operation features and identify behavior patterns through convolutional neural networks, recurrent neural networks, etc. However, these technologies are usually used for offline analysis and training of complex operations, and still have problems such as large computational complexity and poor adaptability in real-time environments. Although convolutional neural networks (CNNs) have high expressiveness in feature extraction, existing methods are usually unable to adaptively adjust the size and shape of the convolution kernel to adapt to the diversity of touch operations. In addition, although traditional recurrent neural networks (RNNs) have certain advantages in extracting memory features and temporal features, they have a weak response to frequent changes in short-term operations, and their ability to screen and remember historical data is insufficient. They cannot effectively update and focus on the current and historical operations in real-time operations, resulting in poor performance in changing game situations.

[0005] Existing technologies also face many challenges in processing operation intention prediction and multi-modal feature fusion. Self-supervised learning has certain applications in unlabeled data processing, but traditional self-supervised learning methods usually rely on static features, which makes it difficult to effectively combine real-time scene features, operation intentions, and spatiotemporal features, and their performance in personalized control optimization is not ideal. Existing technologies often use simple linear combinations or static weight allocations when processing multi-modal feature fusion, lacking flexible adaptability to different scenarios, resulting in the inability to accurately judge feature priorities in specific scenarios. In addition, existing technologies are also relatively simple in the design of feedback scoring mechanisms, usually relying only on a single operation success rate or response speed, lacking a comprehensive scoring mechanism for control accuracy, operation fluency, and player satisfaction, and are difficult to provide comprehensive feedback for subsequent adjustments.

[0006] Based on the above background, the limitations of the existing technology are mainly reflected in the following aspects: First, there is a lack of adaptive adjustment capability for touch features, which makes it impossible to extract and respond to dynamic features in operation data in real time; second, there is a lack of intelligent memory update mechanism for operation history data, and it is difficult for the existing recurrent neural network to dynamically associate the features of current operation with those of historical operation, resulting in inaccurate recognition of operation intention; third, it is difficult for the existing self-supervised learning method to effectively generate adaptive multi-modal feature fusion, and it is impossible to comprehensively consider the importance of different features in different scenarios, resulting in inaccurate control optimization process; fourth, there is a lack of a comprehensive feedback scoring mechanism, and the existing technology fails to make full use of feedback scoring for dynamic adjustment and adaptive learning of operation parameters, making it difficult to achieve control optimization based on real-time feedback. The above defects make the existing mobile game control optimization technology obviously insufficient in dealing with high-frequency, complex, and diversified game operation requirements.

[0007] Therefore, how to provide a mobile game dynamic control optimization method based on adaptive deep learning is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0008] One purpose of the present invention is to propose a method for optimizing dynamic control of mobile games based on adaptive deep learning. The present invention introduces an adaptive deep learning method, adopts a spatiotemporal feature adaptive convolutional network, an adaptive memory network and a multi-dimensional feedback scoring mechanism, to optimize the control of mobile games in real time and dynamically. The method can adaptively adjust the control parameters according to the touch force, operation intention and scene requirements, and generate accurate and personalized operation feedback. In complex and highly dynamically changing game scenarios, the present invention significantly improves the fluency and response speed of control, has the advantages of high precision, high adaptability and good user experience, and provides players with a natural and smooth interactive operation experience.

[0009] A method for optimizing dynamic control of mobile games based on adaptive deep learning according to an embodiment of the present invention comprises the following steps:

[0010] S1. Collect player operation data and scene background parameters, and generate feature data of uniform scale through data denoising, normalization processing and false touch data filtering;

[0011] S2. The size and shape of the convolution kernel are adjusted through the spatiotemporal feature adaptive convolution network, and the convolution kernel and time sampling step are dynamically adjusted according to the touch force and trajectory to generate an adaptive spatiotemporal feature sequence;

[0012] S3. Build an adaptive memory network to dynamically update the memory pool based on the temporal relationship between the current operation and the historical operation, identify recent high-frequency operation patterns through a short-term focus mechanism, and generate operation intention features;

[0013] S4. Generate self-supervised learning tasks through self-supervised learning multi-layer perceptron, and integrate the feature data of uniform scale, adaptive spatiotemporal feature sequence and operation intention features into multi-modal features, and adaptively adjust the feature weights to generate control parameters according to the scene requirements;

[0014] S5. Generate feedback scores in real time through the feedback scoring mechanism, and dynamically adjust the learning rate based on the feedback scores to achieve adaptive adjustment;

[0015] S6. Establish a satisfaction evaluation mechanism based on feedback scoring and data analysis, and adjust the control parameters in real time when unsatisfactory operation is detected through a relearning mechanism;

[0016] S7. Test the response speed and operation accuracy, record the operation feedback and scoring results in real time, and implement adaptive optimization based on the feedback loop.

[0017] Optionally, the S2 specifically includes:

[0018] S21, initializing the extraction module of the spatiotemporal feature adaptive convolutional network, inputting the collected touch data into the input layer of the adaptive convolutional network, and extracting the touch position, touch force and trajectory dynamic features;

[0019] S22, based on real-time touch force F strength (t) and trajectory change rate R trajectory (t), dynamically adjust the size parameters of the convolution kernel:

[0020]

[0021] Among them, K adaptive_size ( t ) Represents the dynamic adaptive parameter of the convolution kernel size, α conv_scale Represents the overall adjustment coefficient of the convolution kernel, λ path_mod represents the path adaptive adjustment coefficient, σ strength_bias represents the intensity deviation coefficient, η response_mod Represents the response adjustment coefficient of touch intensity, θ decay represents the exponential decay coefficient, β adapt_smooth represents the adaptive smoothing coefficient, δ strength_norm Indicates the touch intensity normalization coefficient;

[0022] S23, according to the continuity and direction change of the operation trajectory, the convolution kernel shape parameter K adaptive_shape (t) Dynamically adjust the shape of the convolution kernel to adapt to the changing direction of the touch path;

[0023] S24, when the touch acceleration A is detected feedback When (t) changes, the sampling step size is dynamically adjusted:

[0024]

[0025] Among them, T dynamic_interval ( t ) represents the dynamic adaptive parameter of the time sampling step, γ base_interval Represents the basic sampling step coefficient, δ accel_mod represents the acceleration adjustment coefficient, ∈ log_smooth represents the logarithmic smoothing coefficient, ν strength_weight represents the touch intensity weight coefficient, κ decay_rate represents the exponential decay coefficient;

[0026] S25, normalizing the spatiotemporal features obtained with different time sampling steps to generate a normalized spatiotemporal feature sequence;

[0027] S26, integrating the adjusted convolution kernel size, shape parameters and normalized spatiotemporal feature sequence to generate an adaptive spatiotemporal feature sequence S sta (t).

[0028] Optionally, the S3 specifically includes:

[0029] S31, initializing the adaptive memory network, inputting the current operation data and the historical operation sequence into the memory network, establishing the time association between the current operation and the historical operation, and forming a multi-dimensional time series feature;

[0030] S32. Build a dynamic memory pool to store historical operation features that are highly relevant to the current operation. The dynamic memory pool state is represented by M dynamic (t), the dynamic memory pool state M dynamic (t) Based on the current operation characteristics V current (t) and historical operating characteristics H history (t) to achieve dynamic update:

[0031]

[0032] Among them, α cur Represents the weight coefficient of the current operation, β hist represents the weight coefficient of historical operations, tanh represents the hyperbolic tangent function, and λ rate represents the time-dependent rate of change, σ decay represents the adaptive decay coefficient of the dynamic memory pool, γ memory represents the fusion coefficient of the dynamic memory pool, δ stabilizer represents the state stability coefficient, exp represents the exponential function;

[0033] S33, introduce a short-term focusing mechanism to aggregate the operation frequencies at the current moment and the most recent moments, and define an aggregation time window W focus , in the aggregation time window W focus Internal statistical operation frequency characteristics F freq (t), capturing high-frequency operating characteristics in the short term;

[0034] S34, based on dynamic memory pool state M dynamic (t) and operating frequency characteristics F freq (t) Generate high-frequency mode features V pattern (t):

[0035]

[0036] Among them, γ freq represents the operating frequency characteristic weight coefficient, δ mem Represents the dynamic memory pool state weight coefficient, ηinteraction represents the interaction adjustment coefficient between frequency characteristics and memory state, θ context represents the contextual correlation coefficient, κ stabilizer represents the mode stability coefficient;

[0037] S35, high frequency mode characteristics V pattern (t) and the current operating characteristics V current (t) Perform feature fusion to generate a preliminary operation intention vector V intent_raw (t);

[0038] S36, preliminary operation intention vector V intent_raw (t) Perform smoothing to eliminate noise and generate the final operation intention feature V intent (t).

[0039] Optionally, the S4 specifically includes:

[0040] S41. Create a self-supervised learning task in the self-supervised learning multi-layer perceptron, construct a pseudo-label set of unlabeled data, and convert the feature data matrix X of the uniform scale into unified Input to the self-supervised learning multi-layer perceptron and construct a pseudo label set Y based on the game scene pseudo Simulate player operation behaviors to obtain preliminary learning objectives;

[0041] S42, fusion adaptive spatiotemporal feature sequence S sta (t) and the operation intention feature V intent (t), generate a preliminary multi-modal feature vector F multi ( t);

[0042] S43, multi-mode feature vector F multi 9t) and the uniform scale feature data matrix X unified Further integration generates a multi-mode comprehensive feature vector F combined 9t), by integrating data from different sources, comprehensively reflecting the player's operation characteristics and scenario requirements;

[0043] S44, dynamically adjust feature weights based on the current game scene, and construct a scene adaptive weight factor matrix W scene 9t), adjust the weights of the multi-modal comprehensive features to ensure that each feature has the best fusion ratio in different scenarios;

[0044] S45, multi-mode comprehensive feature vector F combined 9t) Perform nonlinear activation processing and generate the control feature vector F through the activation function control 9 t);

[0045] S46, manipulate the feature vector F control (t) is input to the feature reconstruction layer, and a nonlinear transformation is performed through the combination of the feature reconstruction weight matrix and the context adjustment matrix to generate the operation parameters:

[0046]

[0047] Among them, P control (t0 represents the operating parameters at time t, W control represents the feature reconstruction weight matrix, C context represents the context adjustment matrix, α represents the coefficient of the nonlinear activation term, β represents the smoothing factor, γ represents the exponential decay rate coefficient, δ represents the historical stability coefficient, and P control (t-1) represents the operating parameters at time t-1, P control 9 t-2) represents the operating parameters at time t-2.

[0048] Optionally, the S5 specifically includes:

[0049] S51. Construct a multi-dimensional feedback scoring mechanism, calculate the accuracy, response speed and success rate of player operations, and perform weighted combination to generate a feedback score value F in real time. score (t), used to quantify the current control effect

[0050] S52. Define the feedback rating change rate ΔF score (t) is used to monitor the changing trend of the feedback score and capture the rate of increase or decrease of the score. The feedback score change rate ΔF score (t) identifying a changing trend of the operational performance, thereby initiating a learning rate adjustment based on the changing trend;

[0051] S53. Design an adaptive learning rate adjustment function to nonlinearly adjust the learning rate according to the feedback score change trend:

[0052]

[0053] Among them, η ( t ) represents the dynamic learning rate at time t, η base represents the basic learning rate, α adjust represents the learning rate adjustment coefficient, tanh represents the hyperbolic tangent function, β rate represents the rating change rate adjustment coefficient, γ smooth represents the smoothing coefficient, δ stabilize represents the stability factor, F target Indicates the preset target score;

[0054] S54, introduce a quick adjustment mechanism, when the feedback score value Fscore (t) is lower than the set minimum score threshold F min When the score is stabilized, the learning rate adjustment is activated to adapt to the changing needs of the current operation performance. After the score is stabilized, it automatically returns to the normal adaptive adjustment mode.

[0055] S55. Construct a score smoothing module to perform mean smoothing on the score within a certain time window to reduce the direct impact of score fluctuations on the learning rate;

[0056] S56, in the whole process of feedback score generation and learning rate adjustment, based on the smoothed dynamic learning rate η smooth ( t) Achieve continuous adaptive optimization.

[0057] Optionally, the S6 specifically includes:

[0058] S61. Establish a comprehensive satisfaction evaluation model, and use the player's feedback score, operation flow and response delay key parameters to generate a comprehensive satisfaction score M satisfaction (t), quantify the current operating experience, comprehensive satisfaction score M satisfaction (t) Based on the calculation of multi-dimensional scoring function, the weight of each scoring dimension is set according to the operation requirements;

[0059] S62. Design a mechanism for analyzing the trend of satisfaction changes and calculate the comprehensive satisfaction score M at consecutive moments. satisfaction (t) Conduct dynamic monitoring to capture the changing trend and calculate the increment of the comprehensive satisfaction score ΔM at the current moment. satisfaction (t), if the score continues to decrease, it is marked as a downward trend, otherwise it is marked as an upward trend. When the downward trend is detected to exceed the set number of times, it is judged that the player satisfaction is decreasing, and the dissatisfied operation detection is triggered;

[0060] S63. Establish a dissatisfaction operation detection mechanism and set a minimum satisfaction threshold M min , used to identify whether the player's satisfaction with the current control meets expectations. If the comprehensive satisfaction score M satisfaction (t) is lower than the minimum satisfaction threshold M min , mark the current operation as an unsatisfactory operation and enter a low satisfaction state;

[0061] S64: When a low satisfaction state of unsatisfactory operation is detected, a relearning mechanism is activated, using an increased relearning rate η relearn Adjust the control parameters, the relearning rate η relearn Set according to satisfaction gap and change trend:

[0062] η relearn (t)=

[0063]

[0064] Among them, η relearn ( t ) represents the relearning rate at time t, η base represents the basic learning rate, α adapt Represents the learning rate increase coefficient, β trend represents the trend response coefficient, λ stability represents the stability adjustment coefficient, γ smooth represents the smoothing factor, δ balance Indicates the balance coefficient, M target Indicates goal satisfaction;

[0065] S65. During the relearning process, the control parameters are continuously optimized based on real-time feedback, feedback scores and satisfaction trends;

[0066] S66. When the satisfaction score returns to the set target satisfaction score M target Or when a stable state is met after several consecutive operations, the relearning rate is gradually reduced to return to the normal learning rate mode.

[0067] The beneficial effects of the present invention are:

[0068] First, the present invention uses a spatiotemporal feature adaptive convolutional network, which can dynamically adjust the size and shape of the convolution kernel, and adaptively capture the operation details in real time according to the touch force and trajectory changes, effectively avoiding the problem that the existing fixed convolution kernel cannot adapt to the diversity of touch. This innovation enables the system to more accurately extract the spatiotemporal features of the player's operation, capture the changes in different touch forces and positions, and maintain flexible adaptability to touch features in complex game scenes, thereby improving the smoothness and accuracy of the control and providing players with more natural operation feedback.

[0069] Secondly, by constructing an adaptive memory network, the present invention achieves accurate prediction of operation intentions and efficient memory updates. The adaptive memory network can dynamically update the memory pool at different times, intelligently associate the time association of the current operation and the historical operation, and effectively identify recent high-frequency operation patterns through a short-term focusing mechanism to generate operation intention features. This dynamic update mechanism of the memory network makes up for the defects of inflexible memory association and inaccurate operation prediction in the prior art, enabling the system to always accurately identify the player's intentions during complex and changeable game operations, providing a more reliable basis for the adjustment of operation parameters.

[0070] In addition, the present invention also generates an adaptive multi-mode feature fusion mechanism through self-supervised learning multi-layer perceptron, which effectively integrates the feature data of uniform scale, adaptive spatiotemporal feature sequence and operation intention feature. Based on the requirements of the game scene, the system can dynamically adjust the weights of multi-mode features so that the contribution of different features in different scenes can be flexibly adjusted. This design enables the system to provide more targeted control optimization in highly dynamically changing game situations, ensures the reasonable allocation of the weights of each feature in complex scenes, effectively improves the generation accuracy of control parameters, and thus enhances the system's adaptability in diversified scenes.

[0071] In terms of feedback mechanism, the present invention constructs a multi-dimensional feedback scoring mechanism, generates feedback scores in real time, and dynamically adjusts the learning rate through feedback scores to ensure that the system can make timely adaptive adjustments when the operation performance fluctuates. The real-time acquisition of multi-dimensional feedback scores such as comprehensive operation accuracy, response speed and player satisfaction enables the system to comprehensively evaluate the current control effect, and adaptively adjust the learning rate according to the changing trend of the score, thereby avoiding the problem that a single score cannot reflect the overall control quality. Through this mechanism, the present invention can achieve highly sensitive adaptive optimization in the feedback link. The system can promptly accelerate the learning rate when the player's operation deviates, and adjust smoothly in a stable state, ensuring continuous improvement in control optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0073] Figure 1 This is a flow chart of a method for optimizing dynamic control of mobile games based on adaptive deep learning proposed by the present invention;

[0074] Figure 2 This is a structural schematic diagram of adjusting the learning rate and control parameters of a feedback scoring mechanism of a mobile game dynamic control optimization method based on adaptive deep learning proposed by the present invention. DETAILED DESCRIPTION

[0075] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0076] refer to Figure 1 and Figure 2 , a mobile game dynamic control optimization method based on adaptive deep learning, comprising the following steps:

[0077] S1. Collect player operation data and scene background parameters, and generate feature data of uniform scale through data denoising, normalization processing and false touch data filtering;

[0078] S2. The size and shape of the convolution kernel are adjusted through the spatiotemporal feature adaptive convolution network, and the convolution kernel and time sampling step are dynamically adjusted according to the touch force and trajectory to generate an adaptive spatiotemporal feature sequence;

[0079] S3. Build an adaptive memory network to dynamically update the memory pool based on the temporal relationship between the current operation and the historical operation, identify recent high-frequency operation patterns through a short-term focus mechanism, and generate operation intention features;

[0080] S4. Generate self-supervised learning tasks through self-supervised learning multi-layer perceptron, and integrate the feature data of uniform scale, adaptive spatiotemporal feature sequence and operation intention features into multi-modal features, and adaptively adjust the feature weights to generate control parameters according to the scene requirements;

[0081] S5. Generate feedback scores in real time through the feedback scoring mechanism, and dynamically adjust the learning rate based on the feedback scores to achieve adaptive adjustment;

[0082] S6. Establish a satisfaction evaluation mechanism based on feedback scoring and data analysis, and adjust the control parameters in real time when unsatisfactory operation is detected through a relearning mechanism;

[0083] S7. Test the response speed and operation accuracy, record the operation feedback and scoring results in real time, and implement adaptive optimization based on the feedback loop.

[0084] In this implementation, S2 specifically includes:

[0085] S21, initializing the extraction module of the spatiotemporal feature adaptive convolutional network, inputting the collected touch data into the input layer of the adaptive convolutional network, and extracting the touch position, touch force and trajectory dynamic features;

[0086] S22, based on real-time touch force F strength (t) and trajectory change rate R trajectory (t), dynamically adjust the size parameters of the convolution kernel:

[0087]

[0088] Among them, K adaptive_size ( t ) Represents the dynamic adaptive parameter of the convolution kernel size, α conv_scale Represents the overall adjustment coefficient of the convolution kernel, λ path_mod represents the path adaptive adjustment coefficient, σ strength_bias represents the intensity deviation coefficient, ηresponse_mod Represents the response adjustment coefficient of touch intensity, θ decay represents the exponential decay coefficient, β adapt_smooth represents the adaptive smoothing coefficient, δ strength_norm Indicates the touch intensity normalization coefficient;

[0089] S23, according to the continuity and direction change of the operation trajectory, the convolution kernel shape parameter K adaptive_shape (t) Dynamically adjust the shape of the convolution kernel to adapt to the changing direction of the touch path;

[0090] S24, when the touch acceleration A is detected feedback When (t) changes, the sampling step size is dynamically adjusted:

[0091]

[0092] Among them, T dynamic_interval ( t ) represents the dynamic adaptive parameter of the time sampling step, γ base_interval Represents the basic sampling step coefficient, δ accel_mod represents the acceleration adjustment coefficient, ∈ log_smooth represents the logarithmic smoothing coefficient, ν strength_weight represents the touch intensity weight coefficient, κ decay_rate represents the exponential decay coefficient;

[0093] S25, normalizing the spatiotemporal features obtained with different time sampling steps to generate a normalized spatiotemporal feature sequence;

[0094] S26, integrating the adjusted convolution kernel size, shape parameters and normalized spatiotemporal feature sequence to generate an adaptive spatiotemporal feature sequence S sta (t).

[0095] In this implementation, S3 specifically includes:

[0096] S31, initializing the adaptive memory network, inputting the current operation data and the historical operation sequence into the memory network, establishing the time association between the current operation and the historical operation, and forming a multi-dimensional time series feature;

[0097] S32. Build a dynamic memory pool to store historical operation features that are highly relevant to the current operation. The dynamic memory pool state is represented by M dynamic (t), the dynamic memory pool state M dynamic (t) Based on the current operation characteristics V current (t) and historical operating characteristics H history (t) to achieve dynamic update:

[0098]

[0099] Among them, α cur Represents the weight coefficient of the current operation, β hist represents the weight coefficient of historical operations, tanh represents the hyperbolic tangent function, and λ rate represents the time-dependent rate of change, σ decay represents the adaptive decay coefficient of the dynamic memory pool, γ memory represents the fusion coefficient of the dynamic memory pool, δ stabilizer represents the state stability coefficient, exp represents the exponential function;

[0100] S33, introduce a short-term focusing mechanism to aggregate the operation frequencies at the current moment and the most recent moments, and define an aggregation time window W focus , in the aggregation time window W focus Internal statistical operation frequency characteristics F freq (t), capturing high-frequency operating characteristics in the short term;

[0101] S34, based on the dynamic memory pool state M dynamic (t) and operating frequency characteristics F freq (t) Generate high-frequency mode features V pattern (t):

[0102]

[0103] Among them, γ freq represents the operating frequency characteristic weight coefficient, δ mem Represents the dynamic memory pool state weight coefficient, η interaction represents the interaction adjustment coefficient between frequency characteristics and memory state, θ context represents the contextual correlation coefficient, κ stabilizer represents the mode stability coefficient;

[0104] S35, high frequency mode characteristics V pattern (t) and the current operating characteristics V current (t) Perform feature fusion to generate a preliminary operation intention vector V intent_raw (t);

[0105] S36, preliminary operation intention vector V intent_raw (t) Perform smoothing to eliminate noise and generate the final operation intention feature V intent (t).

[0106] In this implementation, S4 specifically includes:

[0107] S41. Create a self-supervised learning task in the self-supervised learning multi-layer perceptron, construct a pseudo-label set of unlabeled data, and convert the feature data matrix X of the uniform scale into unified Input to the self-supervised learning multi-layer perceptron and construct a pseudo label set Y based on the game scene pseudo Simulate player operation behaviors to obtain preliminary learning objectives;

[0108] S42, fusion adaptive spatiotemporal feature sequence S sta (t) and the operation intention feature V intent (t), generate a preliminary multi-modal feature vector F multi ( t);

[0109] S43, multi-mode feature vector F multi (t) and the uniformly scaled feature data matrix X unified Further integration generates a multi-mode comprehensive feature vector F combined (t) By integrating data from different sources, it fully reflects the player's operating characteristics and scenario requirements;

[0110] S44, dynamically adjust feature weights based on the current game scene, and construct a scene adaptive weight factor matrix W scene (t), adjust the weights of multi-modal comprehensive features to ensure that each feature has the best fusion ratio in different scenarios;

[0111] S45, multi-mode comprehensive feature vector F combined (t) Perform nonlinear activation processing and generate the manipulated feature vector F through the activation function control (t);

[0112] S46, manipulate the feature vector F control (t) is input to the feature reconstruction layer, and a nonlinear transformation is performed through the combination of the feature reconstruction weight matrix and the context adjustment matrix to generate the operation parameters:

[0113]

[0114] Among them, P control (t) represents the operating parameters at time t, W control represents the feature reconstruction weight matrix, C context represents the context adjustment matrix, α represents the coefficient of the nonlinear activation term, β represents the smoothing factor, γ represents the exponential decay rate coefficient, δ represents the historical stability coefficient, and P control (t-1) represents the operating parameters at time t-1, P control ( t-2) represents the operating parameters at time t-2.

[0115] In this implementation manner, S5 specifically includes:

[0116] S51. Construct a multi-dimensional feedback scoring mechanism, calculate the accuracy, response speed and success rate of player operations, and perform weighted combination to generate a feedback score value F in real time. score (t), used to quantify the current control effect

[0117] S52. Define the feedback rating change rate ΔF score (t) is used to monitor the changing trend of the feedback score and capture the rate of increase or decrease of the score. The feedback score change rate ΔF score (t) identifying a changing trend of the operational performance, thereby initiating a learning rate adjustment based on the changing trend;

[0118] S53. Design an adaptive learning rate adjustment function to nonlinearly adjust the learning rate according to the feedback score change trend:

[0119]

[0120] Among them, η ( t ) represents the dynamic learning rate at time t, η base represents the basic learning rate, α adjust represents the learning rate adjustment coefficient, tanh represents the hyperbolic tangent function, β rate represents the rating change rate adjustment coefficient, γ smooth represents the smoothing coefficient, δ stabilize represents the stability factor, F target Indicates the preset target score;

[0121] S54, introduce a quick adjustment mechanism, when the feedback score value F score (t) is lower than the set minimum score threshold F min When the score is stabilized, the learning rate adjustment is activated to adapt to the changing needs of the current operation performance. After the score is stabilized, it automatically returns to the normal adaptive adjustment mode.

[0122] S55. Construct a score smoothing module to perform mean smoothing on the score within a certain time window to reduce the direct impact of score fluctuations on the learning rate;

[0123] S56, in the whole process of feedback score generation and learning rate adjustment, based on the smoothed dynamic learning rate η smooth ( t) Achieve continuous adaptive optimization.

[0124] In this implementation manner, S6 specifically includes:

[0125] S61. Establish a comprehensive satisfaction evaluation model, and use the player's feedback score, operation flow and response delay key parameters to generate a comprehensive satisfaction score M satisfaction (t), quantify the current operating experience, comprehensive satisfaction score M satisfaction (t) Based on the calculation of multi-dimensional scoring function, the weight of each scoring dimension is set according to the operation requirements;

[0126] S62. Design a mechanism for analyzing the trend of satisfaction changes and calculate the comprehensive satisfaction score M at consecutive moments. satisfaction (t) Conduct dynamic monitoring to capture the changing trend and calculate the increment of the comprehensive satisfaction score ΔM at the current moment. satisfaction (t), if the score continues to decrease, it is marked as a downward trend, otherwise it is marked as an upward trend. When the downward trend is detected to exceed the set number of times, it is judged that the player satisfaction is decreasing, and the dissatisfied operation detection is triggered;

[0127] S63. Establish a dissatisfaction operation detection mechanism and set a minimum satisfaction threshold M min , used to identify whether the player's satisfaction with the current control meets expectations. If the comprehensive satisfaction score M satisfaction (t) is lower than the minimum satisfaction threshold M min , mark the current operation as an unsatisfactory operation and enter a low satisfaction state;

[0128] S64: When a low satisfaction state of unsatisfactory operation is detected, a relearning mechanism is activated, using an increased relearning rate η relearn Adjust the control parameters, the relearning rate η relearn Set according to satisfaction gap and change trend:

[0129]

[0130] Among them, η relearn ( t ) represents the relearning rate at time t, η base represents the basic learning rate, α adapt Represents the learning rate increase coefficient, β trend represents the trend response coefficient, λ stability represents the stability adjustment coefficient, γ smooth represents the smoothing factor, δ balance Indicates the balance coefficient, M target Indicates goal satisfaction;

[0131] S65. During the relearning process, the control parameters are continuously optimized based on real-time feedback, feedback scores and satisfaction trends;

[0132] S66. When the satisfaction score returns to the set target satisfaction score Mtarget Or when a stable state is met after several consecutive operations, the relearning rate is gradually reduced to return to the normal learning rate mode.

[0133] Embodiment 1:

[0134] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a real-time strategy mobile game. The game requires players to quickly perform multi-finger operations in complex battle scenes, such as commanding character movement, triggering skills, avoiding enemy attacks, etc. The game operation environment is complex and changeable, and has extremely high requirements for operation accuracy, response speed, and control fluency. It is difficult for existing technologies to adapt to such diverse operation requirements in real time. Players often encounter problems such as slow touch response, misjudgment of operation intentions, and inability to accurately recognize gestures during operation, resulting in a poor gaming experience. Especially in emergency battle scenes with high-frequency operations, players' satisfaction and operation fluency are low.

[0135] The adaptive deep learning dynamic control optimization method of the present invention is applied in the game. By collecting the player's touch operation data (including touch position, strength, sliding trajectory, etc.) and the background parameters of the current game scene (such as game difficulty, combat status, enemy position, etc.), the touch data is denoised and normalized and then input into the adaptive convolution network and adaptive memory network of the system, so as to generate control parameters in real time and dynamically adjust the character movement and skill release in the game. The adaptive convolution network adjusts the size and shape of the convolution kernel according to the changes in touch strength and touch position, extracts the touch spatiotemporal features, and enables the system to identify the operation intentions of different touch positions and strengths. The adaptive memory network dynamically updates the memory pool according to the player's current operation and historical operation data, extracts recent high-frequency operation modes through a short-term focusing mechanism, and accurately identifies the player's intentions. Based on multi-mode feature fusion, the self-supervised learning mechanism generates control parameters, and scores real-time operations through a feedback scoring mechanism, and dynamically adjusts the learning rate based on the scoring trend, so that the system responds to dissatisfied operations in real time.

[0136] In the actual test, we selected two groups of players to conduct comparative tests in the game scenes under the same conditions. The first group of players used the conventional control mode, and the second group of players used the dynamic control optimization method of the present invention. The test scene of the game is set to a difficult battle level, with a large number of enemies and frequent scene changes, and players need to perform touch operations continuously. The entire test process was carried out in a game testing center in Beijing from May 10, 2024 to May 17, 2024. The test involved a total of 50 players, each of whom completed 20 operation tasks.

[0137] Table 1 Comparison of touch response speed data of players in the conventional group and the invention group

[0138]

[0139]

[0140] Table 2 Comparison of touch accuracy data of players in the conventional group and the invention group

[0141]

[0142]

[0143] Table 3 Comparison of satisfaction scores of players in the conventional group and the invention group

[0144] Test Date Player Group Satisfaction rating (out of 5 points) May 10, 2024 Conventional group 3.9 May 10, 2024 The present invention group 4.8 May 11, 2024 Conventional group 3.7 May 11, 2024 The present invention group 4.6 May 12, 2024 Conventional group 3.8 May 12, 2024 The present invention group 4.7 May 13, 2024 Conventional group 3.6 May 13, 2024 The present invention group 4.7 May 14, 2024 Conventional group 3.8 May 14, 2024 The present invention group 4.6 May 15, 2024 Conventional group 3.7 May 15, 2024 The present invention group 4.8 May 16, 2024 Conventional group 3.8 May 16, 2024 The present invention group 4.7 May 17, 2024 Conventional group 3.8 May 17, 2024 The present invention group 4.7 average Conventional group 3.8 average The present invention group 4.7

[0145] In the dynamic control optimization test of the present invention, by analyzing the data of different player groups in terms of touch response time, operation accuracy and user satisfaction, the significant improvement effect of the present invention in the mobile game control experience is fully demonstrated. First, in terms of touch response time (see Table 1), the average response time of the player group using the present invention is 84 milliseconds, compared with 112 milliseconds in the conventional group, an average reduction of 24.7%, and the standard deviation of the response time is also significantly reduced, indicating that the application of the present invention in dynamic game scenarios significantly improves the stability and speed of the control response. The fast response not only optimizes the player's operating experience, but also improves the player's control accuracy in high-frequency operation scenarios.

[0146] In terms of operation accuracy (see Table 2), the average touch accuracy of the player group using the present invention reached 98.9%, which is significantly higher than the 85.3% of the conventional group. This shows the significant improvement effect of the present invention in complex touch operations, especially in multi-finger, high-speed sliding and frequent switching operations, the false touch rate is only 1.1%, which is significantly lower than the 14.7% of the conventional group. These data show that the adaptive feature extraction and memory mechanism of the present invention can accurately capture the operation intention, effectively reduce the false touch situation, and improve the accuracy of the operation.

[0147] In terms of player satisfaction (see Table 3), the average score of the player group using the present invention was 4.7 points (out of 5 points), which is significantly higher than the 3.8 points of the conventional group. This difference reflects the players' high recognition of the control experience provided by the present invention. The improvement in satisfaction stems from the overall improvement of the control optimization method in terms of response speed, touch accuracy, and operation fluency, especially in frequently changing game scenes, players have a more natural and smooth interactive experience.

[0148] It can be seen that the dynamic control optimization method of the present invention has achieved significant advantages in touch response speed, operation accuracy and player satisfaction through multi-level deep learning technology, which verifies the effectiveness and superiority of the present invention in complex game environments.

[0149] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A mobile game dynamic control optimization method based on adaptive deep learning, characterized in that: The steps include: S1. Collect player operation data and scene background parameters, and generate feature data of uniform scale through data denoising, normalization processing and false touch data filtering; S2. The size and shape of the convolution kernel are adjusted through the spatiotemporal feature adaptive convolution network, and the convolution kernel and time sampling step are dynamically adjusted according to the touch force and trajectory to generate an adaptive spatiotemporal feature sequence; S3. Build an adaptive memory network to dynamically update the memory pool based on the temporal relationship between the current operation and the historical operation, identify recent high-frequency operation patterns through a short-term focus mechanism, and generate operation intention features; S4. Generate self-supervised learning tasks through self-supervised learning multi-layer perceptron, integrate feature data of uniform scale, adaptive spatiotemporal feature sequence and operation intention features into multi-modal features, and adaptively adjust feature weights to generate control parameters according to scenario requirements; S5. Generate feedback scores in real time through the feedback scoring mechanism, and dynamically adjust the learning rate based on the feedback scores to achieve adaptive adjustment; S6. Establish a satisfaction evaluation mechanism based on feedback scoring and data analysis, and adjust the control parameters in real time when unsatisfactory operation is detected through a relearning mechanism; S7. Test the response speed and operation accuracy, record the operation feedback and scoring results in real time, and implement adaptive optimization based on the feedback loop.

2. A mobile game dynamic control optimization method based on adaptive deep learning according to claim 1, characterized in that: The S2 specifically includes: S21, initializing the extraction module of the spatiotemporal feature adaptive convolutional network, inputting the collected touch data into the input layer of the adaptive convolutional network, and extracting the touch position, touch force and trajectory dynamic features; S22, based on real-time touch force F strength (t) and trajectory change rate R trajectory (t), dynamically adjust the size parameters of the convolution kernel: Among them, K adaptive_size (t) represents the dynamic adaptive parameter of the convolution kernel size, α conv_scale Represents the overall adjustment coefficient of the convolution kernel, λ path_mod represents the path adaptive adjustment coefficient, σ strength_bias represents the intensity deviation coefficient, η response_mod Represents the response adjustment coefficient of touch intensity, θ decay represents the exponential decay coefficient, β adapt_smooth represents the adaptive smoothing coefficient, δ strength_norm Indicates the touch intensity normalization coefficient; S23, according to the continuity and direction change of the operation trajectory, the convolution kernel shape parameter K adaptive_shape (t) Dynamically adjust the shape of the convolution kernel to adapt to the changing direction of the touch path; S24, when the touch acceleration A is detected feedback When (t) changes, the sampling step size is dynamically adjusted: Among them, T dynamic_interval (t) represents the dynamic adaptive parameter of the time sampling step, γ base_interval Represents the basic sampling step coefficient, δ accel_mod represents the acceleration adjustment coefficient, ∈ log_smooth represents the logarithmic smoothing coefficient, ν strength_weight represents the touch intensity weight coefficient, κ decay_rate represents the exponential decay coefficient; S25, normalizing the spatiotemporal features obtained with different time sampling steps to generate a normalized spatiotemporal feature sequence; S26, integrating the adjusted convolution kernel size, shape parameters and normalized spatiotemporal feature sequence to generate an adaptive spatiotemporal feature sequence S sta (t).

3. The method for optimizing mobile game dynamic control based on adaptive deep learning according to claim 1, characterized in that: The S3 specifically includes: S31, initializing the adaptive memory network, inputting the current operation data and the historical operation sequence into the memory network, establishing the time association between the current operation and the historical operation, and forming a multi-dimensional time series feature; S32. Build a dynamic memory pool to store historical operation features that are highly relevant to the current operation. The dynamic memory pool state is represented by M dynamic (t), the dynamic memory pool state M dynamic (t) Based on the current operation characteristics V current (t) and historical operating characteristics H history (t) to achieve dynamic update: Among them, α cur Represents the weight coefficient of the current operation, β hist represents the weight coefficient of historical operations, tanh represents the hyperbolic tangent function, and λ rate represents the time-dependent rate of change, σ decay represents the adaptive decay coefficient of the dynamic memory pool, γ memory represents the fusion coefficient of the dynamic memory pool, δ stabilizer represents the state stability coefficient, exp represents the exponential function; S33, introduce a short-term focusing mechanism to aggregate the operation frequencies at the current moment and the most recent moments, and define an aggregation time window W focus , in the aggregation time window W focus Internal statistical operation frequency characteristics F freq (t), capturing high-frequency operating characteristics in the short term; S34, based on the dynamic memory pool state M dynamic (t) and operating frequency characteristics F freq (t) Generate high-frequency mode features V pattern (t): Among them, γ freq represents the operating frequency characteristic weight coefficient, δ mem Represents the dynamic memory pool state weight coefficient, η interaction represents the interaction adjustment coefficient between frequency characteristics and memory state, θ context represents the contextual correlation coefficient, κ stabilizer represents the mode stability coefficient; S35, high frequency mode characteristics V pattern (t) and the current operating characteristics V current (t) Perform feature fusion to generate a preliminary operation intention vector V intent_raw (t); S36, preliminary operation intention vector V intent_raw (t) Perform smoothing to eliminate noise and generate the final operation intention feature V intent (t).

4. The method for optimizing mobile game dynamic control based on adaptive deep learning according to claim 1, characterized in that: The S4 specifically includes: S41. Create a self-supervised learning task in the self-supervised learning multi-layer perceptron, construct a pseudo-label set of unlabeled data, and convert the feature data matrix X of the uniform scale into unified Input to the self-supervised learning multi-layer perceptron and construct a pseudo label set Y based on the game scene pseudo Simulate player operation behaviors to obtain preliminary learning objectives; S42, fusion adaptive spatiotemporal feature sequence S sta (t) and the operation intention feature V intent (t), generate a preliminary multi-modal feature vector F multi (t); S43, multi-mode feature vector F multi (t) and the uniform-scale feature data matrix X unified Further integration generates a multi-mode comprehensive feature vector F combined (t) By integrating data from different sources, it fully reflects the player's operating characteristics and scenario requirements; S44, dynamically adjust feature weights based on the current game scene, and construct a scene adaptive weight factor matrix W scene (t), adjust the weights of multi-modal comprehensive features to ensure that each feature has the best fusion ratio in different scenarios; S45, multi-mode comprehensive feature vector F combined (t) Perform nonlinear activation processing and generate the manipulated feature vector F through the activation function control (t); S46, manipulate the feature vector F control (t) is input to the feature reconstruction layer, and a nonlinear transformation is performed through the combination of the feature reconstruction weight matrix and the context adjustment matrix to generate the operation parameters: Among them, P control (t) represents the operating parameters at time t, W control represents the feature reconstruction weight matrix, C context represents the context adjustment matrix, α represents the coefficient of the nonlinear activation term, β represents the smoothing factor, γ represents the exponential decay rate coefficient, δ represents the historical stability coefficient, and P control (t-1) represents the operating parameters at time t-1, P control (t-2) represents the operating parameters at time t-2.

5. The method for optimizing mobile game dynamic control based on adaptive deep learning according to claim 1, characterized in that: The S5 specifically includes: S51. Construct a multi-dimensional feedback scoring mechanism, calculate the accuracy, response speed and success rate of player operations, and perform weighted combination to generate a feedback score value F in real time. score (t), used to quantify the current control effect S52. Define the feedback rating change rate ΔF score (t) is used to monitor the changing trend of the feedback score and capture the rate of increase or decrease of the score. The feedback score change rate ΔF score (t) identifying a changing trend of the operational performance, thereby initiating a learning rate adjustment based on the changing trend; S53. Design an adaptive learning rate adjustment function to nonlinearly adjust the learning rate according to the feedback score change trend: Among them, η(t) represents the dynamic learning rate at time t, η base represents the basic learning rate, α adjust represents the learning rate adjustment coefficient, tanh represents the hyperbolic tangent function, β rate represents the score change rate adjustment coefficient, γ smooth represents the smoothing coefficient, δ stabilize represents the stability factor, F target Indicates the preset target score; S54, introduce a quick adjustment mechanism, when the feedback score value F score (t) is lower than the set minimum score threshold F min When the score is stabilized, the learning rate adjustment is activated to adapt to the changing needs of the current operation performance. After the score is stabilized, it automatically returns to the normal adaptive adjustment mode. S55. Construct a score smoothing module to perform mean smoothing on the score within a certain time window to reduce the direct impact of score fluctuations on the learning rate; S56, in the whole process of feedback score generation and learning rate adjustment, based on the smoothed dynamic learning rate η smooth (t) Achieve continuous adaptive optimization.

6. The method for optimizing mobile game dynamic control based on adaptive deep learning according to claim 1, characterized in that: The S6 specifically includes: S61. Establish a comprehensive satisfaction evaluation model, and use the player's feedback score, operation flow and response delay key parameters to generate a comprehensive satisfaction score M satisfaction (t), quantify the current operating experience, comprehensive satisfaction score M satisfaction (t) Based on the calculation of multi-dimensional scoring function, the weight of each scoring dimension is set according to the operation requirements; S62. Design a mechanism for analyzing the trend of satisfaction changes and calculate the comprehensive satisfaction score M at consecutive moments. satisfaction (t) Conduct dynamic monitoring to capture the changing trend and calculate the increment of the comprehensive satisfaction score ΔM at the current moment. satisfaction (t), if the score continues to decrease, it is marked as a downward trend, otherwise it is marked as an upward trend. When the downward trend is detected to exceed the set number of times, it is judged that the player satisfaction is decreasing, and the dissatisfied operation detection is triggered; S63. Establish a dissatisfaction operation detection mechanism and set a minimum satisfaction threshold M min , used to identify whether the player's satisfaction with the current control meets expectations. If the comprehensive satisfaction score M satisfaction (t) is lower than the minimum satisfaction threshold M min , mark the current operation as an unsatisfactory operation and enter a low satisfaction state; S64: When a low satisfaction state of unsatisfactory operation is detected, a relearning mechanism is activated, using an increased relearning rate η relearn Adjust the control parameters, the relearning rate η relearn Set according to satisfaction gap and change trend: Among them, η relearn (t) represents the relearning rate at time t, η base represents the basic learning rate, α adapt Represents the learning rate increase coefficient, β trend represents the trend response coefficient, λ stability represents the stability adjustment coefficient, γ smooth represents the smoothing factor, δ balance Indicates the balance coefficient, M target Indicates goal satisfaction; S65. During the relearning process, the control parameters are continuously optimized based on real-time feedback, feedback scores and satisfaction trends; S66. When the satisfaction score returns to the set target satisfaction score M target Or when a stable state is met after several consecutive operations, the relearning rate is gradually reduced to return to the normal learning rate mode.

Images