Game virtual object resource dynamic loading method and system

Abstract

The invention relates to the technical field of data loading, in particular to a game virtual object resource dynamic loading method and system.The method comprises the steps that original loading delay data of a virtual object in a dynamic scene is collected, and correction data is obtained through preprocessing; analyzing system load data in real time, and quantifying resource allocation fluctuation quantity caused by load fluctuation; based on the fluctuation quantity and the correction data, controlling the loading delay within a preset performance range through a feedback control algorithm, and collecting resource occupation distribution data at the same time; testing resource operation states under different scheduling strategies by using a performance simulation method, and generating a simulation result; determining resource scheduling optimization parameters through an optimization algorithm in combination with the game frame rate fluctuation data and the simulation result; finally, a game engine loading mechanism is adaptively adjusted based on parameters, dynamic response to load change and delay fluctuation is achieved, resource scheduling rationality and loading efficiency are improved, and the problems that delay control lags behind and optimization capacity is insufficient in a traditional static scheduling strategy are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data loading, in particular to a game virtual object resource dynamic loading method and system. BACKGROUND

[0002] With the rapid development of the game industry, the complexity of game scenes and the diversification of virtual objects have put forward higher requirements on resource loading efficiency. In dynamic game scenes, real-time loading of virtual objects involves scheduling of a large number of three-dimensional models, textures, animations and other resources. System load fluctuations, hardware performance differences and scene switching frequency and other factors can easily lead to unstable resource loading delays, excessive frame rate fluctuations and other problems, which seriously affect user experience.

[0003] In the prior art, traditional resource loading methods mostly use static scheduling strategies, which lack dynamic response capabilities to real-time load changes and loading performance. On the one hand, they cannot effectively handle the problem of uneven resource allocation caused by sudden changes in system load in dynamic scenes, making it difficult to quickly correct when the loading delay exceeds the reasonable range. On the other hand, the simulation analysis of resource running state under different scheduling strategies is not accurate enough, and it is not possible to adaptively optimize the loading mechanism combined with real-time performance data such as frame rate fluctuations, resulting in unreasonable resource occupation and low loading efficiency and other technical bottlenecks.

[0004] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide a game virtual object resource dynamic loading method and system, which aims to solve the technical problem that traditional game virtual object resource loading methods mostly use static scheduling strategies, which lack dynamic response capabilities to real-time load changes and loading performance.

[0006] To achieve the above purpose, the present application provides a game virtual object resource dynamic loading method, which comprises: Collecting original loading delay data of virtual objects in dynamic scenes during game running, and preprocessing the original loading delay data to obtain loading delay correction data; According to the monitored system real-time load data, the resource scheduling pressure is analyzed to obtain the resource allocation fluctuation caused by load fluctuation; According to the resource allocation fluctuation and the loading delay correction data, the resource loading delay is controlled to return to the preset performance range by feedback control algorithm, and after the resource loading is completed, the resource occupation distribution data of each virtual object instance in the game is collected; According to the resource occupation distribution data, a performance simulation method is used to test the running state of the resource under different scheduling strategies, and a running state simulation result is obtained; Frame rate fluctuation data in a game running process are acquired, and based on the running state simulation result and the frame rate fluctuation data, a resource scheduling optimization parameter is determined; The resource loading mechanism of the game engine is adaptively adjusted based on the resource scheduling optimization parameter.

[0007] Optionally, the preprocessing of the original loading delay data to obtain loading delay correction data comprises: According to the mean value and the variance of the original loading delay data, a preset outlier threshold is used to identify abnormal delay data points from the original loading delay data; According to the Lagrange interpolation method, the abnormal delay data points in the original loading delay data are interpolated and repaired by adjacent effective data points to obtain a repaired delay sequence; The repaired delay sequence is subjected to wavelet transform to obtain a time-frequency distribution diagram, and a delay main frequency band is identified according to the time-frequency distribution diagram; According to the delay main frequency band, an elliptical filter is used to perform frequency band filtering on the repaired delay sequence to obtain a smoothed delay signal; The smoothed delay signal is subjected to three-layer decomposition using Haar wavelet to obtain multi-resolution wavelet coefficients, and the wavelet coefficients are subjected to soft threshold processing to reconstruct the loading delay correction data.

[0008] Optionally, the analysis of the resource scheduling pressure according to the monitored system real-time load data to obtain a resource allocation fluctuation amount caused by load fluctuation comprises: The exponential weighted moving average method is used to calibrate the baseline of the system real-time load data to obtain stable load data; The stable load data is used to calculate a load change gradient, and if the load change gradient exceeds a preset load change threshold, the load change is converted into an initial resource fluctuation amount according to a system resource allocation mechanism; The initial resource fluctuation amount is subjected to capacity compensation to obtain the resource allocation fluctuation amount caused by load fluctuation.

[0009] Optionally, the feedback control algorithm is used to control the resource loading delay to return to a preset performance range according to the resource allocation fluctuation amount and the loading delay correction data, comprising: According to the resource allocation fluctuation amount and the loading delay correction data, a memory allocation compensation amount required for a target resource loading state is calculated; The scheduling correction amount is calculated according to the memory allocation compensation amount, and a target allocation parameter that the resource scheduler needs to reach is determined according to the scheduling correction amount and a current resource loading delay; The best response parameter of the feedback control algorithm is calculated according to the target allocation parameter that the resource scheduler needs to reach by using a particle swarm algorithm; Real-time delay data in the resource loading process is collected, and a standard deviation of the real-time delay data is calculated in a sliding time window to obtain a delay fluctuation amplitude; The delay fluctuation amplitude is compared with a preset delay reference value to obtain a resource delay deviation amount; The resource delay deviation amount is subjected to proportional-integral operation by using the feedback control algorithm according to the best response parameter, to obtain an original control amount, and the resource loading delay is controlled to return to a preset performance range according to the original control amount.

[0010] Optionally, the running state simulation result is obtained by testing the running state of the resource under different scheduling strategies according to the resource occupation distribution data by using a performance simulation method, and includes: Memory and video memory occupation data of each virtual object instance in the game are collected, and a resource distribution topology graph is constructed according to the occupation data of each virtual object instance; The competition condition of the resource in the scheduling process is simulated according to the resource distribution topology graph by using a discrete event simulation method; The occupation trend of the resource under different load conditions is predicted according to the competition condition of the resource in the scheduling process by using a recurrent neural network; The waiting time in the resource scheduling process is calculated according to the occupation trend, and a resource bottleneck node is identified according to the waiting time; Performance counter signals of each resource node in the game running process are collected, and main fluctuation frequencies are extracted from the performance counter signals by short-time Fourier transform, and the main fluctuation frequencies are taken as dynamic load conditions; The running state simulation result is obtained by testing the running state of the resource under different scheduling strategies according to the dynamic load conditions, the resource occupation distribution data and the resource bottleneck node by using a performance simulation method.

[0011] Optionally, the resource scheduling optimization parameter is determined based on the running state simulation result and the frame rate fluctuation data, including: A frame rate fluctuation and loading performance relationship model is established according to the running state simulation result and the frame rate fluctuation data; A resource allocation weight is set according to a system response characteristic, and a load correction coefficient is calculated by using a load influence factor according to the resource delay deviation amount and system load data during game running; The load correction coefficient is combined with a resource allocation weight to minimize loading delay as an optimization target, and a simulated annealing algorithm is used to iteratively optimize resource scheduling parameters until a preset termination condition is met, and the optimized scheduling parameters are output. Control instructions of the resource scheduler are generated according to the optimized scheduling parameters. The resource loading queue of the game engine is adjusted according to the control instructions to obtain an allocation sequence of the resource scheduler. The allocation sequence is subjected to load testing, and a performance error caused by frame rate fluctuation is evaluated by using a frame rate fluctuation and loading performance relationship model, and resource scheduling optimization parameters are output.

[0012] Optionally, the frame rate fluctuation and loading performance relationship model is established according to the running state simulation result and the frame rate fluctuation data, comprising: Resource competition index data is extracted from the running state simulation result, and feature extraction is performed on the resource competition index data by using an independent component analysis method, and core components of the resource competition index data exceeding a preset competition threshold are extracted; Load transfer weights between different resource nodes are calculated according to the core components, and the frame rate fluctuation data is normalized by using the load transfer weights to obtain frame rate fluctuation correction data; System load data and real-time performance data during game running are collected, the frame rate fluctuation correction data, the system load data and the real-time performance data are fused, and a loading performance score matrix is constructed by using an entropy weight method; Frame rate fluctuation correction data and loading performance score matrix are mapped and analyzed by using a support vector machine algorithm to construct a frame rate fluctuation and loading performance relationship model.

[0013] In addition, to achieve the above-mentioned purpose, the application further provides a game virtual object resource dynamic loading system, which comprises: A data preprocessing module is used to collect original loading delay data of virtual objects in a dynamic scene during game running, and the original loading delay data is preprocessed to obtain loading delay correction data; A load quantification module is used to analyze resource scheduling pressure according to monitored system real-time load data to obtain resource allocation fluctuation amount caused by load fluctuation; A feedback control module is used to control resource loading delay to return to a preset performance range by a feedback control algorithm according to the resource allocation fluctuation amount and the loading delay correction data, and resource occupation distribution data of each virtual object instance in the game is collected after resource loading is completed. The performance simulation module is used to test the operating status of resources under different scheduling strategies based on the resource occupancy distribution data and to obtain the operating status simulation results. The parameter optimization module is used to acquire frame rate fluctuation data during game operation, and determine resource scheduling optimization parameters based on the simulation results of the running state and the frame rate fluctuation data. The mechanism control module is used to adaptively adjust the resource loading mechanism of the game engine based on resource scheduling optimization parameters.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a game virtual object resource dynamic loading device, the device comprising: a memory, a processor, and a game virtual object resource dynamic loading program stored on the memory and executable on the processor, the game virtual object resource dynamic loading program being configured to implement the steps of the game virtual object resource dynamic loading method as described above.

[0015] In addition, to achieve the above objectives, the present invention also provides a storage medium storing a game virtual object resource dynamic loading program, wherein when the game virtual object resource dynamic loading program is executed by a processor, it implements the steps of the game virtual object resource dynamic loading method as described above.

[0016] This invention provides a method for dynamically loading virtual object resources in games. By collecting system load data in real time and quantifying resource allocation fluctuations, the method can accurately capture changes in hardware performance. Compared to traditional static scheduling strategies, it can dynamically adjust the resource loading rhythm, avoiding uncontrolled loading delays or resource allocation imbalances caused by sudden load changes, thus improving the system's adaptability to complex scenarios. Based on preprocessed loading delay correction data and a feedback control algorithm, a closed-loop control mechanism is formed, which can quickly control the loading delay within a preset performance range. This effectively reduces screen stuttering caused by slow loading or resource redundancy caused by premature loading, ensuring the consistency of game screen rendering and resource loading timing, and improving user experience smoothness. By collecting resource usage distribution data and simulating the operating states of different scheduling strategies, combined with real-time frame rate fluctuation data, optimization parameters are dynamically determined, achieving a closed-loop resource scheduling strategy. Compared to the limitations of traditional methods that rely on experience-based parameter tuning, this solution can adaptively match the optimal loading strategy for different game scenarios, reducing resource waste and improving the utilization efficiency of hardware resources such as memory and bandwidth. By dynamically adjusting the resource loading mechanism, the game engine can perceive changes in the operating environment in real time, automatically balance loading speed and running performance, avoid drastic frame rate fluctuations caused by hardware compatibility or scene complexity, provide a universal optimization solution for cross-platform games, reduce manual tuning costs, and improve the engine's intelligent scheduling capabilities. Attached Figure Description

[0017] Figure 1 is a game virtual object resource dynamic loading device structure schematic diagram of the hardware running environment involved in the embodiment scheme of the present application. Figure 2 is a flowchart of the game virtual object resource dynamic loading method first embodiment of the present application. Figure 3 is a structural block diagram of the game virtual object resource dynamic loading system first embodiment of the present application.

[0018] The implementation of the object of the present application, functional features and advantages will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0019] It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0020] Referring to Figure 1 , Figure 1 is a game virtual object resource dynamic loading device structure schematic diagram of the hardware running environment involved in the embodiment scheme of the present application.

[0021] As Figure 1 shown, the game virtual object resource dynamic loading device can include: a processor 1001, for example, a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection communication between the components. The user interface 1003 can include a display screen (Display), and the optional user interface 1003 can also include a standard wired interface, a wireless interface, and the wired interface of the user interface 1003 can be a USB interface in the present application. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 can be a high-speed random access memory (RAM), and can also be a stable memory (Non-volatile Memory, NVM), such as a disk memory. The memory 1005 can also be an independent storage device from the aforementioned processor 1001.

[0022] Those skilled in the art can understand that Figure 1 the structure shown in the figure does not constitute a limitation on the game virtual object resource dynamic loading device, and can include more or fewer components than the figure, or combine certain components, or different component arrangements.

[0023] As Figure 1As shown, the memory 1005 as a computer storage medium can include an operating system, a network communication module, a user interface module, and a game virtual object resource dynamic loading program.

[0024] In Figure 1 In the game virtual object resource dynamic loading device shown, the network interface 1004 is mainly used for connecting a background server and communicating data with the background server, and the user interface 1003 is mainly used for connecting an external device. The game virtual object resource dynamic loading device calls the game virtual object resource dynamic loading program stored in the memory 1005 through the processor 1001 and executes the game virtual object resource dynamic loading method provided by the embodiment of the application.

[0025] Based on the above hardware structure, an embodiment of the game virtual object resource dynamic loading method of the application is proposed.

[0026] Referring to Figure 2 , Figure 2 The flowchart of the first embodiment of the game virtual object resource dynamic loading method of the application is provided.

[0027] In the first embodiment, the game virtual object resource dynamic loading method includes the following steps: In one embodiment, a game virtual object resource dynamic loading method is provided, and the method includes the following steps. Step S100, collecting original loading delay data of a virtual object in a dynamic scene during game running, and preprocessing the original loading delay data to obtain loading delay correction data.

[0028] The original loading delay data refers to the time interval originally recorded from the request for loading to the complete availability of resources when the game engine loads three-dimensional models, textures, or animations, etc. The resource loading timestamp can be recorded in real time and the difference can be calculated through the built-in performance monitoring interface of the game engine or third-party tools. The preprocessing refers to the technical process of cleaning, denoising, and standardizing the original data, for example, the abnormal values can be eliminated through sliding average filtering or the delay data of different magnitudes can be unified to the same value range through normalization. The loading delay correction data is the data after preprocessing, which is used to eliminate noise interference and enhance the analyzability of the data. The abnormal values caused by sudden hardware failure are eliminated through an algorithm, and the data is smoothed through a statistical method, so as to improve the accuracy and consistency of the data and avoid the deviation caused by the noise of the original data in subsequent analysis. The statistical method can be the method of mean or median.

[0029] Step S200, analyze the resource scheduling pressure according to the monitored system real-time load data, and obtain the resource allocation fluctuation caused by load fluctuation.

[0030] The system real-time load data refers to the real-time monitoring data of hardware resources such as CPU, GPU utilization, memory occupancy, and bandwidth usage during game running, which can be collected in real time through system monitoring tools, such as performance counters and sensors. The resource scheduling pressure is the load pressure caused by resource requests exceeding the system processing capacity, which can be quantified by comparing the load data with the preset threshold. The resource allocation fluctuation is a quantitative indicator reflecting the change in resource allocation ratio caused by system load mutation, such as calculating the difference between the current allocation and the historical average to represent the fluctuation degree. This process analyzes the load change trend through a dynamic threshold algorithm, which can be a sliding window statistics, and compares the current load with the preset baseline to calculate the resource allocation fluctuation, such as using variance or standard deviation to measure the dispersion degree of the allocation amount, thereby accurately identifying the load mutation event and providing a basis for dynamically adjusting the resource loading rhythm.

[0031] Step S300, according to the resource allocation fluctuation and the loading delay correction data, control the resource loading delay to return to the preset performance range through a feedback control algorithm, and after the resource loading is completed, collect the resource occupancy distribution data of each virtual object instance in the game.

[0032] The preset performance range is the acceptable delay interval defined by the developer, such as controlling the loading delay within 50 to 150 milliseconds to ensure smoothness. The feedback control algorithm is a regulation method based on closed-loop control theory, such as PID control, fuzzy control, or adaptive control algorithm. The resource occupancy distribution data refers to the resource occupancy proportion of each virtual object instance in memory, video memory, or bandwidth, which can be collected through memory analysis tools or engine interfaces. This step inputs the loading delay correction data and system load fluctuation into the feedback control algorithm, such as a PID controller dynamically adjusts the resource loading bandwidth allocation according to the proportional, integral, and differential terms, and corrects the loading rhythm to constrain the delay within the preset range; after loading is completed, further record the distribution proportion of each object on different resource types (such as texture, model), so as to quickly respond to delay fluctuation and provide resource usage evidence for subsequent strategy optimization.

[0033] Step S400, according to the resource occupancy distribution data, test the running state of the resource under different scheduling strategies using performance simulation methods, and obtain the running state simulation results.

[0034] Performance simulation method is to simulate the impact of different resource scheduling strategies on system performance through offline or online simulation technology, where different resource scheduling strategies are, for example, priority queue, dynamic batch loading, such as constructing a load model through an offline simulation tool or predicting the effect of the strategy in real time through online sampling. The running state simulation result refers to the indicators recorded during the simulation, such as average delay, peak memory occupancy, frame rate stability, etc. This step is based on the resource occupancy distribution data to build a mathematical model of the resource scheduling strategy, such as the functional relationship between load and delay, parameterize different scheduling strategies, such as adjusting the loading order, dynamically adjusting the loading granularity, and using the simulation engine to simulate its performance in typical scenarios (such as scene switching, multi-player screen sharing), for example, to simulate the impact of the "priority loading of visible area resources" strategy on delay and memory occupancy, so as to quickly evaluate the feasibility of multiple strategies and reduce the trial and error cost.

[0035] Step S500, obtaining frame rate fluctuation data during game running, and determining resource scheduling optimization parameters based on running state simulation results and frame rate fluctuation data.

[0036] Frame rate fluctuation data refers to the real-time change data of frames per second (FPS) during game rendering, and the fluctuation amplitude reflects the degree of conflict between rendering load and resource loading timing, which can be quantified by statistical methods, such as calculating the standard deviation or peak fluctuation rate. Resource scheduling optimization parameters are configurable variables used to adjust scheduling strategies, such as loading priority weight, resource prefetch threshold, or dynamic bandwidth allocation ratio. This step correlates the frame rate fluctuation with the running state simulation results, such as delay distribution under different strategies, for example, using machine learning models or regression algorithms to establish the mapping relationship between parameters and performance indicators, and finally searching for the optimal parameter combination through optimization algorithms such as gradient descent, genetic algorithm, so that the loading delay and frame rate fluctuation are minimized at the same time, thereby dynamically correlating the scheduling strategy with real-time rendering performance and avoiding conflicts between resource loading and rendering tasks.

[0037] Step S600, adaptively adjusting the resource loading mechanism of the game engine based on the resource scheduling optimization parameters.

[0038] Adaptive adjustment refers to dynamically modifying the resource loading logic inside the engine according to the optimization parameters, such as modifying the loading priority rules, adjusting the resource block size, or enabling / disabling specific optimization modules. The resource loading mechanism is a collection of modules in the engine responsible for resource requests, decompression, decoding, caching, etc. This step feeds back the optimization parameters to the engine loading module in real time through parameter injection or API interface, such as adjusting the weight distribution of the loading queue or enabling block loading to reduce memory peaks, while considering hardware heterogeneity (such as the difference between mobile devices and PCs) and verifying the effectiveness of the strategy through A / B testing, thereby giving the engine environmental perception ability, dynamically adapting to different hardware and scene requirements, and reducing the dependence on manual tuning.

[0039] The game virtual object resource dynamic loading method provided by the embodiment can achieve the following technical effects: through closed-loop control and dynamic modeling, the resource allocation imbalance problem of the traditional static scheduling strategy is solved, delay control precision is realized, resource utilization optimization is realized, cross-platform compatibility is enhanced, and performance stability is improved, and finally an efficient and low-delay virtual object management scheme is provided for a dynamic game scene.

[0040] In one of the embodiments, the original loading delay data is preprocessed to obtain loading delay correction data, including: According to the mean and variance of the original loading delay data, the abnormal delay data points are identified from the original loading delay data by using a preset outlier threshold; The mean is a statistical quantity representing the central tendency of the delay data, which can be obtained by calculating the arithmetic mean of the original data. The variance is a statistical quantity measuring the degree of dispersion of the data, which can be obtained by calculating the average of the squares of the differences between each data point and the mean. The preset outlier threshold is an abnormal value determination criterion set according to statistical principles, for example, using the 3σ principle to define the region deviating from the mean by more than 3 times the standard deviation as the abnormal range. By comparing whether the absolute deviation of each data point from the mean exceeds the preset threshold, abnormal delay data points caused by external disturbances such as hardware failure or network fluctuation can be identified. This process can accurately isolate sudden outliers by combining statistical benchmarking and threshold determination, avoiding interference with subsequent analysis.

[0041] According to the Lagrange interpolation method, the abnormal delay data points in the original loading delay data are interpolated and repaired by using adjacent valid data points to obtain the repaired delay sequence; The Lagrange interpolation method is an interpolation technique based on a polynomial function, which constructs an interpolation polynomial by selecting valid data points adjacent to the abnormal point (for example, 3 points before and after). This method generates a repair value consistent with the trend of the time series by calculating the weight coefficients of the adjacent points, replacing the original abnormal data points. For example, if the abnormal point is located at the kth position of the time series, its repair value can be calculated by linear combination of the delay data and the corresponding time stamp of the adjacent points. This operation preserves the time continuity of the data while eliminating the sharp fluctuations caused by outliers, providing a smooth basis sequence for subsequent frequency domain analysis.

[0042] Wavelet transform is performed on the repaired delay sequence to obtain a time-frequency distribution diagram, and the main frequency band of the delay is identified according to the time-frequency distribution diagram; Wavelet transform is a time-frequency analysis tool that decomposes signals into different frequency components through multi-scale decomposition. The time-frequency distribution diagram is the visualization result of wavelet transform, with the horizontal axis representing time and the vertical axis representing frequency. The color depth reflects the energy distribution of different frequency bands. By analyzing the energy concentration area in the diagram, such as the high-frequency band corresponding to sudden fluctuations and the low-frequency band corresponding to trend changes, the key frequency band of the delay data can be determined, for example, the periodic fluctuation of 0.11 Hz. This process accurately locates the core frequency component of the delay signal through time-frequency joint analysis, providing a frequency band basis for subsequent filtering operations.

[0043] According to the main frequency band of the delay, an elliptical filter is used to filter the repaired delay sequence to obtain a smoothed delay signal. The elliptical filter is an IIR digital filter with steep passband and stopband transition characteristics. By setting the passband boundary (such as including the main frequency band of 0.05 to 1.5 Hz) and the stopband cutoff frequency (such as above 2 Hz), the elliptical filter can retain the effective frequency band related to system load changes while suppressing high-frequency noise. This operation significantly improves the signal-to-noise ratio of the signal by accurately retaining and suppressing the frequency band, eliminating interference components unrelated to the delay characteristics.

[0044] The smoothed delay signal is decomposed into multi-resolution wavelet coefficients using Haar wavelet, and the wavelet coefficients are subjected to soft thresholding to reconstruct the load delay correction data.

[0045] Haar wavelet is a binary decomposition wavelet basis function that decomposes signals into approximation coefficients (low-frequency trend) and detail coefficients (high-frequency noise) through three-level decomposition. Soft thresholding is a denoising method that compresses wavelet coefficients by calculating a threshold (such as based on the square root criterion): coefficients below the threshold are set to zero, and coefficients above the threshold are reduced in proportion. This process retains the trend characteristics of the signal while eliminating high-frequency noise residues through multi-scale analysis, ultimately reconstructing high-precision load delay correction data.

[0046] This embodiment accurately isolates abnormal values by combining statistical thresholding and Lagrange interpolation, locates the core frequency band using wavelet transform and time-frequency analysis, and realizes noise suppression and feature retention using an elliptical filter and multi-resolution wavelet decomposition. Finally, the signal is optimized through soft thresholding reconstruction. This method cooperates through multiple stages of technology, avoiding the impact of data breaks on continuous analysis, and improving the signal-to-noise ratio through frequency domain and time-frequency joint processing. The final output of the correction data provides an accurate reference for subsequent feedback control, significantly enhancing the system's response to load mutations.

[0047] In one of the embodiments, the resource scheduling pressure is analyzed according to the monitored system real-time load data, and the resource allocation fluctuation caused by load fluctuation is obtained, including: The baseline calibration is performed on the system real-time load data by using the exponential weighted moving average method to obtain stable load data. The exponential weighted moving average method is a time series data smoothing technique, which assigns exponentially decreasing weights to historical data and gives higher priority to recent data. Its purpose is to eliminate high-frequency noise in system load data and generate a more stable baseline reference value. For example, the formula The dynamic baseline calibration is realized, wherein is a smoothing coefficient, which is set according to the load data fluctuation characteristics, such as 0.1-0.3, is the original system load data at the current time t, such as CPU utilization, memory occupancy, GPU bandwidth, and other real-time indicators, is the exponential weighted moving average result at the previous time t-1, i.e. the historical baseline value. This method generates smoothed baseline data by weighted summation, which can effectively suppress sensor noise or transient interference, such as temporary CPU occupation by background processes, making the baseline closer to the actual system load trend.

[0048] According to the stable load data, the load change gradient is calculated, and if the load change gradient exceeds the preset load change threshold, the load change is converted into an initial resource fluctuation according to the system resource allocation mechanism; The load change gradient refers to the rate of system load change per unit time, which can be obtained by calculating the ratio of the difference between adjacent time windows of stable load data and the time interval. For example, if the current window load is and the previous window is , the gradient is , wherein is the time interval of adjacent time windows, such as 100ms or 500ms. Its purpose is to quantify the severity of load change. The preset load change threshold is a load change rate threshold defined by the developer, which is used to judge whether the current load fluctuation exceeds the system tolerable range. For example, it can be set to 3 times the standard deviation of historical gradient to distinguish normal fluctuation from sudden events. When the load change gradient exceeds the threshold, the system maps the gradient to the initial resource fluctuation according to the resource allocation mechanism. For example, the initial adjustment amount is calculated by a linear relationship , wherein is a proportionality coefficient, which is pre-set according to the resource type and system capacity. This mapping needs to be combined with the allocation rules of specific resource types (such as memory, bandwidth) to ensure that the adjustment direction is consistent with the load change direction, such as reducing non-critical resource loading when the load rises.

[0049] The initial resource fluctuation is capacity compensated to obtain a resource allocation fluctuation caused by load fluctuation.

[0050] Capacity compensation refers to a process of correcting the initial resource fluctuation according to the physical capacity limit of the system hardware resource. For example, if the initial fluctuation exceeds the remaining available memory, adjustment is made through proportional scaling or priority clipping, where proportional scaling is, for example, compressing the fluctuation to 90% of the remaining capacity, and priority clipping is, for example, suspending low-priority resource loading to ensure the feasibility of resource allocation. The compensation process can introduce constraint conditions, such as a bandwidth allocation not being lower than a minimum guarantee value, to avoid making resource allocation infeasible due to excessive adjustment, for example, a crash caused by insufficient memory.

[0051] This embodiment can achieve the following technical effects: filtering noise through exponential weighted moving average, making load change judgment more reliable and reducing false triggering; gradient threshold mechanism quickly locates load mutation events, ensuring the timeliness of resource adjustment; capacity compensation avoids conflicts between theoretical calculation values and actual hardware capabilities, preventing system crashes caused by excessive adjustment; and through configurable proportional coefficients and compensation rules, the resource characteristics of different hardware platforms, such as the memory limit of mobile devices and the high bandwidth scenario of PCs, are adapted. In summary, this embodiment decomposes the quantification process of resource allocation fluctuation into operable sub-steps by combining mathematical modeling with hardware constraints, improves the response accuracy of the system to load mutations, and ensures the engineering feasibility of resource adjustment, providing reliable input data for subsequent feedback control and strategy optimization.

[0052] In one of the embodiments, according to the resource allocation fluctuation and the loading delay correction data, the resource loading delay is controlled to return to a preset performance range through a feedback control algorithm, including: According to the resource allocation fluctuation and the loading delay correction data, the memory allocation compensation amount required for the target resource loading state is calculated; The resource allocation fluctuation amount can be a quantitative index representing the degree of change in system resource allocation, and can be obtained by collecting the difference values of real-time data such as system memory occupancy and task queue length. For example, the resource allocation fluctuation amount can include a texture memory occupancy mutation value or an instantaneous change amount of thread concurrency during scene switching. The loading delay correction data can be loading delay time series data after filtering or normalization processing, and can be obtained by pre-processing the original delay data using a sliding average algorithm or a Kalman filter. For example, the loading delay correction data can include a GPU shader loading delay sequence after removing abnormal values. The memory allocation compensation amount can be an amount of memory resource adjustment required to offset the impact of resource fluctuations on delay, and is used to balance the difference between burst load and resource demand. For example, the memory allocation compensation amount can be manifested as increasing the texture cache reserved space or releasing the unused vertex buffer space. The technical operation can correlate the resource allocation fluctuation amount and the memory compensation demand by establishing a mathematical model, for example, using a linear regression model or a support vector regression model, inputting historical fluctuation data and delay correction data, and outputting the theoretical value of the compensation amount, so as to achieve the technical effect of quantifying the demand for dynamic adjustment of memory resources, avoiding loading blockage due to insufficient allocation or resource waste caused by over-allocation.

[0053] The scheduling correction amount is calculated according to the memory allocation compensation amount, and the target allocation parameter that the resource scheduler needs to reach is determined according to the scheduling correction amount and the current resource loading delay; The scheduling correction amount can be a parameter increment for correcting the resource loading rhythm, such as an increase or decrease ratio of the loading bandwidth or an adjustment amplitude of the loading priority weight, which can be realized by mapping the memory compensation amount to the scheduling parameter space. For example, the scheduling correction amount can include increasing the priority weight of a texture loading task or reducing the compression rate of a background map. The target allocation parameter can be the final control target that the resource scheduler needs to reach, such as the memory bandwidth allocation ratio or the concurrency number threshold of the loading task, which can be obtained by combining hardware constraints and real-time load state for parameter boundary checking. For example, the target allocation parameter can include the percentage of video memory bandwidth allocated to the rendering thread pool or the maximum concurrency number of model loading tasks. The technical operation can convert the memory compensation amount into an executable scheduling parameter adjustment amount, such as converting the compensation amount into a specific bandwidth percentage by proportional scaling, while considering hardware limitations such as memory capacity upper limit and CPU core number, so as to achieve the technical effect of converting abstract compensation amount into specific scheduling action, improving the operability of the control system.

[0054] The best response parameter of the feedback control algorithm is calculated using a particle swarm algorithm according to the target allocation parameter that the resource scheduler needs to reach; The particle swarm algorithm (PSO) can be a swarm intelligence optimization algorithm for searching for optimal solutions of feedback control parameters. The particle swarm algorithm can be used to quickly explore the combination space of proportional, integral, and differential parameters. For example, the particle swarm algorithm can be used to optimize the combination of Kp, Ki, and Kd parameters in a PID controller. The technical operation can be performed by initializing a set of parameter particles, iteratively evaluating the performance indicators of each parameter combination in the simulation environment, such as the convergence speed and overshoot, updating the position of the particles according to the individual historical optimal solution and the group optimal solution, and gradually converging to the global optimal parameter combination. Thus, the technical effect of replacing manual trial and error or fixed parameter setting can be achieved, and the response accuracy and adaptability of feedback control can be significantly improved.

[0055] Real-time delay data is collected during the resource loading process, and the standard deviation of the real-time delay data is calculated in a sliding time window to obtain the delay fluctuation amplitude. The sliding time window can be a fixed time length sampling interval for rolling collection of real-time delay data, such as a window period of 1 second or 500 milliseconds. The delay fluctuation amplitude can be a statistical indicator that quantifies the degree of dispersion of the delay data. The delay fluctuation amplitude can be used to reflect the stability of the loading delay. For example, the delay fluctuation amplitude can be expressed as a fluctuation level with a standard deviation less than 5 milliseconds or more than 20 milliseconds. The technical operation can be performed by collecting a sequence of delay data in real time, calculating the mean value of the data in the window, and then taking the square root of the average of the squared differences between each data point and the mean value. Thus, the technical effect of accurately quantifying the severity of delay fluctuation can be achieved, providing a quantitative basis for subsequent control.

[0056] The delay fluctuation amplitude is compared with a preset delay reference value to obtain a resource delay deviation amount. The preset delay reference value can be a system-defined upper limit of the target delay standard deviation, such as a standard deviation of no more than 10 milliseconds set according to the user experience threshold. The resource delay deviation amount can be the difference between the delay fluctuation amplitude and the reference value. The resource delay deviation amount can be used to determine whether the current delay exceeds the tolerance range. For example, the resource delay deviation amount can include a positive deviation (excessive fluctuation) or a negative deviation (insufficient fluctuation). The technical operation can be performed by simple subtraction operation to calculate the deviation amount, such as deviation amount = current standard deviation - reference value. Thus, the technical effect of quickly identifying abnormal fluctuation can be achieved, ensuring the timeliness of the control action.

[0057] According to the best response parameters, the feedback control algorithm is used to perform proportional integral operation on the resource delay deviation amount to obtain the original control amount, and the original control amount is used to control the resource loading delay to return to the preset performance range.

[0058] The proportional integral operation can be a control equation calculation process based on the optimal response parameters, for example, through the formula: original control amount = Kp x deviation amount + Ki x ∫deviation amount dt, wherein Kp and Ki are the proportional and integral coefficients optimized by the particle swarm algorithm. The original control amount can be an adjustment signal to be applied, which is used to generate a smooth and fast-converging control instruction. For example, the original control amount can be manifested as adjusting the texture compression level or dynamically adjusting the size of the thread pool. The technical operation can achieve the technical effect of avoiding the problems of simple proportional control oscillation or integral saturation by substituting the optimized PID parameters into the control equation and combining the historical error and the current error for weighted summation, thereby ensuring that the delay quickly converges within the preset range.

[0059] The embodiment can achieve the technical effect of significantly improving the resource scheduling accuracy and system stability in complex scenarios by calculating the memory allocation compensation amount to quantify the influence of resource fluctuations, optimizing the feedback control parameters using the particle swarm algorithm, monitoring real-time delay fluctuations using a sliding window, and generating proportional integral control amounts. This scheme upgrades the feedback control from passive response to active prediction through multi-level optimization, avoiding both overallocation and underallocation of resources and achieving millisecond-level real-time suppression of delay deviation, thereby effectively reducing the probability of perceived lag and improving resource utilization.

[0060] In one embodiment, according to the resource occupancy distribution data, the performance simulation method is used to test the running state of the resource under different scheduling strategies, and the running state simulation results are obtained, including: The memory and video memory occupancy data of each virtual object instance in the game are collected, and a resource distribution topology graph is constructed according to the occupancy data of each virtual object instance.

[0061] The memory and video memory occupancy data can be quantitative indicators describing the real-time resource consumption of virtual object instances during running, which can be obtained through memory analysis tools (such as Valgrind, GPUProfiler) or video memory detection interfaces. For example, it includes video memory occupancy rate, memory allocation size, texture resource size, etc. The resource distribution topology graph can be a resource relationship network constructed by mathematical models or visual methods, with nodes representing specific resource instances and edges representing the dependency or competition relationship between resources, such as object A depending on the decompression operation of texture B. The construction of this topology graph can be achieved through resource IDs or object hierarchical relationships, for example, when an object instance needs to access multiple texture resources, edges pointing to these texture nodes are established in the topology graph.

[0062] According to the resource distribution topology graph, the discrete event simulation method is used to simulate the competition status of the resource in the scheduling process.

[0063] As the input basis, the resource distribution topology map, the discrete event simulation method can be an event-triggered simulation technology, which simulates the resource competition process by tracking the time sequence of discrete events such as resource loading and release. This method can be implemented through an event-driven architecture, for example, recording the arrival time, service time and completion time of resource access requests, and tracking resource conflict events through an event queue, such as two threads simultaneously requesting the decompression operation of the same texture resource. For example, the simulation process can quantify the number of resource conflicts, the length of the waiting queue and other competition condition indicators.

[0064] According to the competition condition of the resource in the scheduling process, the occupation trend of the resource under different load conditions is predicted by using a recurrent neural network.

[0065] The competition condition of the resource in the scheduling process can be a sequence of resource conflict events or a sequence of resource occupation time recorded in the simulation process. The recurrent neural network (such as long short-term memory network LSTM) can be a prediction model with time sequence memory capability, which predicts the rising or falling trend of resource occupation in a future time window by inputting historical resource occupation data (such as a sequence of memory occupation rate changes in the past 10 seconds). For example, the LSTM network can capture long-term dependencies in resource occupation data, thereby identifying potential resource overload periods.

[0066] According to the occupation trend, the waiting time in the resource scheduling process is calculated, and the resource bottleneck node is identified according to the waiting time.

[0067] The waiting time in the resource scheduling process can be the length of time that a resource request stays in the competition or queue, which can be derived from the resource occupation trend and the request queue length in the simulation data. The resource bottleneck node can be a resource instance or node whose waiting time exceeds a preset threshold, for example, a texture loading delay node caused by GPU bandwidth contention. The identification process can be achieved by comparing the waiting time distribution of each node. For example, if the average waiting time of a node is significantly higher than that of other nodes, it is determined to be a bottleneck.

[0068] The performance counter signals of each resource node during game running are collected, and the main fluctuation frequency is extracted from the performance counter signals through short-time Fourier transform, and the main fluctuation frequency is taken as the dynamic load condition.

[0069] The performance counter signal can be a real-time collected sequence of hardware or software performance indicators, such as CPU utilization, GPU bandwidth occupancy, frame rate fluctuation, etc. The short-time Fourier transform (STFT) can be an analysis method for converting time-domain signals into time-frequency spectra, which identifies the frequency components of the dominant load fluctuations by windowing the signal and calculating the Fourier transform of each window. For example, after windowing the signal within 1 second, a scene switching frequency of 1 time per 2 seconds (corresponding to 0.5 Hz) can be detected. The main fluctuation frequency as a dynamic load condition can be to map the detected dominant frequency to the load injection period parameter in the simulation, such as generating resource request peaks with the same period.

[0070] According to the dynamic load condition, the resource occupancy distribution data, and the resource bottleneck node, the performance simulation method is used to test the running state of the resource under different scheduling strategies, and the running state simulation result is obtained.

[0071] The dynamic load condition can be a periodic load pattern generated based on the fluctuation frequency extracted by STFT, such as a GPU load peak triggered once every 2 seconds. The resource occupancy distribution data can be the resource consumption statistics of each node in the topology graph. The resource bottleneck node can be the high latency node identified in the previous step. The performance simulation method can be implemented through a parameterized simulation model (such as adjusting the load priority weight, block size), for example, the delay and occupancy changes after enabling the texture compression algorithm for the GPU bottleneck node. By comparing the average latency, resource utilization, and other indicators of different scheduling strategies under periodic load, the optimal strategy can be selected.

[0072] The embodiment quantifies the resource distribution characteristics by collecting resource occupancy data and constructing a topology graph, traces resource competition events using discrete event simulation, predicts resource occupancy trends using a recurrent neural network, locates bottleneck nodes through latency analysis, extracts load fluctuation frequencies using STFT to generate dynamic load conditions, and finally tests the running state of different scheduling strategies based on multi-dimensional data fusion. The following technical effects can be achieved: Through the combination of topology graph and competition simulation, deep bottlenecks such as GPU bandwidth contention or memory fragmentation can be accurately identified, improving the accuracy of bottleneck location; a dynamic load model is constructed based on the periodic fluctuation frequency extracted by STFT, making the simulation environment more similar to the load change pattern of the real scene; the prediction of resource occupancy trends using a recurrent neural network provides a basis for forward-looking adjustments of scheduling strategies, such as preloading resource types that will be overloaded; the fusion of multi-source data significantly improves the matching degree of simulation results and actual scenarios, reducing the deviation of strategy evaluation caused by model simplification. In summary, this method significantly improves the comprehensiveness and accuracy of resource scheduling strategy evaluation by introducing topology modeling, time series prediction, and dynamic load simulation technology, providing a reliable basis for the selection of optimization parameters.

[0073] In one of the embodiments, based on the running state simulation result and the frame rate fluctuation data, the resource scheduling optimization parameter is determined, including: A frame rate fluctuation and loading performance relationship model is established according to the running state simulation result and the frame rate fluctuation data; According to the system response characteristics, the resource allocation weight is set, and according to the resource delay deviation and the system load data during game running, the load correction coefficient is calculated by using the load influence factor; The load correction coefficient is combined with the resource allocation weight, and the simulated annealing algorithm is used to iteratively optimize the resource scheduling parameter with the optimization target of minimizing the loading delay until the preset termination condition is met, and the optimized scheduling parameter is output; According to the optimized scheduling parameter, the regulation and control instruction of the resource scheduler is generated; According to the regulation and control instruction, the resource loading queue of the game engine is adjusted to obtain the allocation sequence of the resource scheduler; The allocation sequence is load tested, the performance error caused by the frame rate fluctuation is evaluated by using the frame rate fluctuation and loading performance relationship model, and the resource scheduling optimization parameter is output.

[0074] The frame rate fluctuation and loading performance relationship model can be a mathematical model established by statistical or machine learning method, which is used to quantify the correlation between the frame rate fluctuation amplitude and the performance indicators such as resource loading delay and bandwidth occupation. The model can be trained by collecting delay distribution and real-time frame rate fluctuation data under different scheduling strategies, extracting key parameters through feature engineering, and using linear regression, support vector machine or deep neural network method. For example, the model can include a multiple regression analysis model or a neural network model for predicting the influence of scheduling strategy on frame rate. The resource allocation weight can be the priority coefficient of different resource types during scheduling, and its value determines the resource loading sequence and bandwidth allocation ratio. The weight can be set to an initial value through system response characteristics, for example, the texture loading weight is set to 0.6 and the model loading weight is set to 0.4. The load correction coefficient can be an adjustment coefficient calculated by the load influence factor and the current system load data, which is used to dynamically correct the resource allocation weight. Its calculation method can be the product of the load influence factor and the ratio of the current load to the preset load upper limit. The load influence factor can be determined by the correlation analysis of load data and historical performance, which reflects the quantitative influence of system load on scheduling effect.

[0075] The simulated annealing algorithm can be a global optimization algorithm that avoids getting stuck in local optima by probabilistically accepting inferior solutions in the solution space, simulating the thermodynamic annealing process. The algorithm can take the load correction factor and resource allocation weight as initial parameters and define a fitness function that aims to minimize the loading delay. In a specific embodiment, after initializing parameters such as temperature and cooling rate, the algorithm generates neighbor solutions through random perturbation and decides whether to accept new solutions based on the Metropolis criterion. The temperature is gradually reduced during the iteration process until the termination condition is met, such as reaching the maximum number of iterations or the fitness change being less than a threshold, thereby searching for the globally optimal combination of resource scheduling parameters, such as the loading bandwidth allocation ratio or task block size. The control instructions of the resource scheduler can be a set of instructions that map the optimized scheduling parameters to specific operations. For example, when the optimized texture loading weight is increased to 0.7, the priority rules of the loading queue can be modified through the engine API to prioritize high-weight resources at the head of the queue, and the task block size or concurrency can be adjusted. The instructions need to be compatible with the interface of the game engine to ensure the actual execution effect of the scheduling logic. The allocation sequence of the resource scheduler can be an optimized resource loading order and parameter configuration list, such as a loading task list arranged in priority and the corresponding bandwidth allocation ratio. The sequence is generated by adjusting the loading queue through control instructions, and its technical effect is to convert abstract parameters into scheduling logic executable by the engine, ensuring that the optimization results actually take effect. The performance error can be the difference between the actual runtime frame rate fluctuation and the model prediction value, which is used to evaluate the accuracy of the scheduling strategy. The error can be calculated by collecting actual frame rate data and loading delay through load testing, for example, using indicators such as mean square error. By feeding the error back to the model parameters or weight adjustment of the optimization algorithm, the final output of the resource scheduling optimization parameters can be corrected to ensure their effectiveness in actual scenarios.

[0076] The embodiment quantifies the impact of the scheduling strategy on the frame rate by establishing a model of the relationship between frame rate fluctuation and loading performance, forms adaptive parameter configuration by combining dynamically adjusted resource allocation weight and load correction factor, searches for globally optimal scheduling parameters using the simulated annealing algorithm, converts the parameters into loading queue control instructions executable by the engine, and forms a closed-loop verification through load testing and error evaluation. The following technical effects can be achieved: First, the optimization target is expanded to frame rate stability through the quantitative model, avoiding rendering congestion caused by too fast loading; second, the dynamic weight adjustment mechanism balances resource demand in real time, reducing competition conflicts; third, the global optimization algorithm breaks through the limitation of local optimum, improving the quality of multi-parameter combined scheduling strategy; finally, the closed-loop verification mechanism reduces the risk of model bias, ensuring the effectiveness of the optimization parameters in complex scenarios. This solution upgrades resource scheduling from experience adjustment to data-driven intelligent decision-making, taking into account both loading delay control and frame rate stability, providing a more robust resource management solution.

[0077] In one of the embodiments, a frame rate fluctuation and loading performance relationship model is established according to the running state simulation result and the frame rate fluctuation data, including: The resource competition index data is extracted from the running state simulation result, and the resource competition index data is characterized by using an independent component analysis method, and the core component of the resource competition index data exceeding a preset competition threshold is extracted; The resource competition index data refers to a numerical parameter quantifying the resource contention conflict between different resource nodes, which can be obtained by recording the resource occupancy rate peak, the number of conflicts and other data in the running state simulation process, and exemplary includes the number of memory bandwidth contention, the peak occupancy rate of GPU shader unit, etc. The independent component analysis method is a blind source separation technology, which is used to decompose the mixed signal into statistically independent components, and exemplary can be implemented by using the FastICA algorithm. The core component refers to the independent component that significantly affects the frame rate fluctuation after being screened by the independent component analysis, and its characteristic is that the resource competition index exceeds the preset threshold (such as the resource occupancy rate exceeding 80%), which may cause a performance bottleneck. By using the independent component analysis method to characterize the resource competition index data, the core component can be screened by combining statistical tests, for example, identifying memory bandwidth contention and GPU shader conflict as the main factors of frame rate drop, so as to eliminate redundant features and improve the relevance of subsequent modeling.

[0078] The load transfer weight between different resource nodes is calculated according to the core component, and the frame rate fluctuation data is normalized by using the load transfer weight to obtain frame rate fluctuation correction data; The load transfer weight is a coefficient quantifying the mutual influence degree of different resource nodes, which can be calculated by historical data statistics or causal inference method, and exemplary includes the influence coefficient of memory bandwidth contention on GPU delay. The normalization process refers to the process of converting the frame rate fluctuation data into dimensionless numerical values, and exemplary can be implemented by linear transformation or weight weighted standardization. By calculating the load transfer weight and combining the normalization process, the original frame rate fluctuation data can be divided by the total load value weighted by the weight, for example, if the memory competition causes the GPU delay weight to be 0.7, the corrected frame rate fluctuation data can eliminate the coupling effect between resources and more accurately reflect the influence of single resource scheduling.

[0079] The system load data and real-time performance data during game running are collected, the frame rate fluctuation correction data, the system load data and the real-time performance data are fused, and an entropy weight method is used to build a loading performance score matrix; The system load data refers to the hardware resource usage state parameters recorded during the running process, for example, CPU utilization, GPU temperature, memory remaining space, etc.; the real-time performance data refers to performance indicators directly related to the rendering or loading process, for example, texture loading time consumption, rendering time consumption, etc. The loading performance score matrix is a matrix structure containing weighted scores of multiple dimensions of performance indicators, which can be generated by weighted summation, for example, frame rate fluctuation weight 0.4, memory bandwidth weight 0.3, etc. The entropy weight method is a method for calculating objective weights of indicators based on information entropy, which can determine the weight distribution by calculating the information entropy values of each indicator. By fusing multiple source data and using the entropy weight method to construct the score matrix, indicators of different dimensions can be converted into a unified scoring system to provide comparable input data for subsequent modeling.

[0080] The frame rate fluctuation correction data and the loading performance score matrix are mapped and analyzed using a support vector machine algorithm to construct a frame rate fluctuation and loading performance relationship model.

[0081] The support vector machine algorithm is a supervised learning method that realizes classification or regression by finding an optimal hyperplane, which can use a radial basis function (RBF) as a kernel function to handle nonlinear relationships. The frame rate fluctuation correction data is used as input features, and the loading performance score matrix is used as label data, which can be trained by minimizing the structural risk, for example, by solving a convex quadratic programming problem to determine the classification hyperplane or regression function. By constructing a mapping relationship using a support vector machine, the nonlinear relationship between the frame rate fluctuation standard deviation and the loading delay peak value can be captured, for example, a regression equation is generated to quantify the delay change corresponding to an increase of 1% in frame rate fluctuation, thereby improving the prediction ability of the model for complex performance relationships.

[0082] The embodiment extracts resource competition indicator data and uses independent component analysis to filter core components to eliminate redundant features, calculates load transfer weights and normalizes frame rate data to eliminate resource coupling effects, fuses multiple dimensions of data and constructs a score matrix using an entropy weight method to unify the input system, and finally constructs a nonlinear mapping model using a support vector machine to capture complex performance correlations, which can significantly improve the accuracy and practicality of the model. The specific performance is: accurately identifying the core resource competition factors that dominate frame rate fluctuation, dynamically correcting data to be close to the real scene, objectively weighting to realize comprehensive evaluation of multiple source data, and enhancing the prediction ability through nonlinear modeling, thereby providing a reliable basis for resource scheduling optimization, and finally realizing the coordinated improvement of loading efficiency and rendering fluency.

[0083] In addition, the embodiment of the present application also proposes a storage medium, which stores a game virtual object resource dynamic loading program. When the game virtual object resource dynamic loading program is executed by a processor, the steps of the game virtual object resource dynamic loading method described above are realized.

[0084] Further, with reference to Figure 3 , the embodiment of the present application also proposes a game virtual object resource dynamic loading system, the game virtual object resource dynamic loading system comprises: a data preprocessing module 10, used for collecting original loading delay data of virtual objects in a dynamic scene during game running, and preprocessing the original loading delay data to obtain loading delay correction data; a load quantification module 20, used for analyzing resource scheduling pressure according to monitored system real-time load data to obtain resource allocation fluctuation caused by load fluctuation; a feedback control module 30, used for controlling resource loading delay to return to a preset performance range through a feedback control algorithm according to the resource allocation fluctuation and the loading delay correction data, and collecting resource occupation distribution data of each virtual object instance in the game after resource loading is completed; a performance simulation module 40, used for testing running states of resources under different scheduling strategies according to the resource occupation distribution data by using a performance simulation method to obtain running state simulation results; a parameter optimization module 50, used for obtaining frame rate fluctuation data during game running, and determining resource scheduling optimization parameters based on the running state simulation results and the frame rate fluctuation data; a mechanism regulation module 60, used for adaptively adjusting a resource loading mechanism of a game engine based on the resource scheduling optimization parameters.

[0085] Other embodiments or specific implementations of the game virtual object resource dynamic loading system described in the present application can refer to the above-mentioned method embodiments, which will not be described here.

[0086] It should be noted that in this paper, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or system. Without more limitations, the element defined by the statement "includes a" does not exclude the presence of another identical element in the process, method, article or system including the element.

[0087] The above-mentioned embodiment number of the present application is only for description, not representing the advantages and disadvantages of the embodiments. In the system unit claims of several systems, several of these systems can be embodied by the same hardware item. The use of the words first, second, and third does not represent any order, and these words can be interpreted as names.

[0088] Those skilled in the art can clearly understand the above-mentioned example method can be realized by means of software and the necessary general hardware platform, of course, also can be through hardware, but in many cases the former is a better implementation. Based on such understanding, the technical solutions of the present application essentially or say the part of the prior art contribution can be embodied in the form of software products, the computer software product is stored in a storage medium (such as read only memory image (Read Only Memory image, ROM) / random access memory (Random Access Memory, RAM), disk, optical disk), including a number of instructions to make a terminal user equipment (may be a mobile phone, computer, server, air conditioner, or network user equipment, etc.) executes the method described in various embodiments of the present application.

[0089] The above is only the preferred embodiment of the present application, not therefore limit the patent scope of the present application, any equivalent structure or equivalent flow transformation using the content of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for dynamically loading virtual object resources in a game, characterized in that, The method includes: The original loading delay data of virtual objects in dynamic scenes during game operation is collected, and the original loading delay data is preprocessed to obtain loading delay correction data. Based on the monitored real-time system load data, the resource scheduling pressure is analyzed to obtain the amount of resource allocation fluctuation caused by load fluctuation. Based on the resource allocation fluctuation and the loading delay correction data, the resource loading delay is controlled to return to the preset performance range through a feedback control algorithm, and the resource occupancy distribution data of each virtual object instance in the game is collected after the resource loading is completed; Based on the resource occupancy distribution data, the performance simulation method is used to test the operating status of resources under different scheduling strategies, and the operating status simulation results are obtained. Obtain frame rate fluctuation data during game operation, and determine resource scheduling optimization parameters based on the simulation results of the running state and the frame rate fluctuation data; The resource loading mechanism of the game engine is adaptively adjusted based on resource scheduling optimization parameters.

2. The method for dynamically loading game virtual object resources as described in claim 1, characterized in that, The preprocessing of the original loading delay data to obtain loading delay correction data includes: Based on the average value and variance of the original loading delay data, abnormal delay data points are identified from the original loading delay data using a preset outlier threshold. Based on the Lagrange interpolation method, abnormal delay data points in the original loading delay data are interpolated and repaired by using nearby valid data points to obtain the repaired delay sequence; The repaired delay sequence is subjected to wavelet transform to obtain a time-frequency distribution map, and the main delay frequency bands are identified based on the time-frequency distribution map; Based on the main frequency band of the delay, an elliptic filter is used to perform band filtering on the repaired delay sequence to obtain a smoothed delay signal. The smoothed delayed signal is decomposed into three levels using Haar wavelets to obtain multi-resolution wavelet coefficients. The wavelet coefficients are then subjected to soft thresholding to reconstruct the loaded delay correction data.

3. The method for dynamically loading game virtual object resources as described in claim 1, characterized in that, The analysis of resource scheduling pressure based on monitored real-time system load data yields the resource allocation fluctuation caused by load fluctuations, including: The real-time load data of the system is baseline-calibrated using an exponentially weighted moving average method to obtain stable load data. The load change gradient is calculated based on the stable load data. If the load change gradient exceeds the preset load change threshold, the load change is converted into an initial resource fluctuation based on the system resource allocation mechanism. The initial resource fluctuation is compensated for by capacity to obtain the resource allocation fluctuation caused by load fluctuation.

4. The method for dynamically loading game virtual object resources as described in claim 1, characterized in that, The step of controlling the resource loading delay to return to a preset performance range through a feedback control algorithm based on the resource allocation fluctuation and the loading delay correction data includes: Based on the resource allocation fluctuation and the loading delay correction data, the memory allocation compensation required for the target resource loading state is calculated. The scheduling correction amount is calculated based on the memory allocation compensation amount, and the target allocation parameters that the resource scheduler needs to achieve are determined based on the scheduling correction amount and the current resource loading delay. Based on the target allocation parameters that the resource scheduler needs to achieve, the optimal response parameters of the feedback control algorithm are calculated using the particle swarm optimization algorithm. Collect real-time latency data during resource loading and calculate the standard deviation of the real-time latency data within a sliding time window to obtain the latency fluctuation amplitude; The delay fluctuation amplitude is compared with a preset delay benchmark value to obtain the resource delay deviation. Based on the optimal response parameters, the resource delay deviation is calculated using a feedback control algorithm to obtain the original control value, and the resource loading delay is controlled to return to the preset performance range based on the original control value.

5. The method for dynamically loading game virtual object resources as described in claim 1, characterized in that, The step of testing the operational status of resources under different scheduling strategies using performance simulation methods based on the resource occupancy distribution data, and obtaining the operational status simulation results, includes: Collect memory and video memory usage data for each virtual object instance in the game, and construct a resource distribution topology map based on the usage data of each virtual object instance; Based on the resource distribution topology, the competition for resources during the scheduling process is simulated using the discrete event simulation method. Based on the competition for resources during the scheduling process, a recurrent neural network is used to predict the resource occupancy trend under different load conditions. The waiting time during resource scheduling is calculated based on the occupancy trend, and resource bottleneck nodes are identified based on the waiting time. The performance counter signals of each resource node during game operation are collected, and the main fluctuation frequency is extracted from the performance counter signals through short-time Fourier transform. The main fluctuation frequency is used as the dynamic load condition. Based on the dynamic load conditions, the resource occupancy distribution data, and the resource bottleneck nodes, the performance simulation method is used to test the operating status of resources under different scheduling strategies, and the operating status simulation results are obtained.

6. The method for dynamically loading game virtual object resources as described in claim 4, characterized in that, The determination of resource scheduling optimization parameters based on the simulation results of the operating state and the frame rate fluctuation data includes: A model of the relationship between frame rate fluctuation and loading performance is established based on the simulation results of the running state and the frame rate fluctuation data. Resource allocation weights are set based on system response characteristics, and load correction coefficients are calculated using load impact factors based on the resource latency deviation and system load data during game operation. By combining the load correction factor with the resource allocation weight, with the goal of minimizing loading latency, the simulated annealing algorithm is used to iteratively optimize the resource scheduling parameters until the preset termination condition is met, and the optimized scheduling parameters are output. The resource scheduler's control instructions are generated based on the optimized scheduling parameters; Adjust the game engine's resource loading queue according to the control instructions to obtain the resource scheduler's allocation sequence; The allocation sequence is subjected to load testing. The performance error caused by frame rate fluctuation is evaluated using a model that shows the relationship between frame rate fluctuation and loading performance. Resource scheduling optimization parameters are then output.

7. The method for dynamically loading game virtual object resources as described in claim 6, characterized in that, The step of establishing a model relating frame rate fluctuations to loading performance based on the simulation results of the running state and the frame rate fluctuation data includes: Resource competition index data are extracted from the simulation results of the operating state, and the independent component analysis method is used to extract features from the resource competition index data to extract the core components of the resource competition index data that exceed the preset competition threshold. The load transfer weight between different resource nodes is calculated based on the core components, and the frame rate fluctuation data is normalized using the load transfer weight to obtain frame rate fluctuation correction data. Collect system load data and real-time performance data during game runtime, fuse the frame rate fluctuation correction data, system load data and real-time performance data, and construct a loading performance scoring matrix using the entropy weight method; The support vector machine algorithm is used to map the frame rate fluctuation correction data with the loading performance score matrix to construct a model of the relationship between frame rate fluctuation and loading performance.

8. A dynamic loading system for game virtual object resources, characterized in that, The game virtual object resource dynamic loading system includes: The data preprocessing module is used to collect the original loading delay data of virtual objects in dynamic scenes during game operation, and to preprocess the original loading delay data to obtain loading delay correction data. The load quantization module is used to analyze the resource scheduling pressure based on the monitored real-time system load data and obtain the amount of resource allocation fluctuation caused by load fluctuation. The feedback control module is used to control the resource loading delay to return to a preset performance range based on the resource allocation fluctuation and the loading delay correction data, and to collect the resource occupancy distribution data of each virtual object instance in the game after the resource loading is completed. The performance simulation module is used to test the operating status of resources under different scheduling strategies based on the resource occupancy distribution data and to obtain the operating status simulation results. The parameter optimization module is used to acquire frame rate fluctuation data during game operation, and determine resource scheduling optimization parameters based on the simulation results of the running state and the frame rate fluctuation data. The mechanism control module is used to adaptively adjust the resource loading mechanism of the game engine based on resource scheduling optimization parameters.

9. A device for dynamically loading virtual object resources in a game, characterized in that, The device includes: a memory, a processor, and a game virtual object resource dynamic loading program stored in the memory and executable on the processor, the game virtual object resource dynamic loading program being configured to implement the steps of the game virtual object resource dynamic loading method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a game virtual object resource dynamic loading program, which, when executed by the processor, implements the steps of the game virtual object resource dynamic loading method as described in any one of claims 1 to 7.

Images