Game equipment fault monitoring method

Abstract

The invention relates to the field of intelligent equipment monitoring and fault detection, in particular to a game equipment fault monitoring method. Comprising the steps of collecting operation data of game equipment, and constructing a multi-dimensional time sequence data set; performing inter-dimension cross-correlation recursive analysis on the multi-dimensional time sequence data set, and generating a nested time window matrix by using multi-layer nested time window analysis; an interaction matrix is introduced, the nested time window matrix is processed through an adaptive interdependence network model, and a comprehensive node state vector is generated; calculating a prediction fault score based on the comprehensive node state vector; and based on the predicted fault score, monitoring the fault of the game equipment. The problems that when complex game equipment operation data are processed, false alarm or missing report is likely to be generated, the state change of the game equipment cannot be dynamically adapted, potential correlation among all dimensions cannot be fully utilized to improve the fault detection precision, and model parameters cannot be adjusted in time to adapt to newly occurring data features are solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent device monitoring and fault detection, and in particular to a method for monitoring faults of gaming devices. Background Art

[0002] In the development of modern gaming devices, with the improvement of hardware performance and the increasing complexity of software functions, the structure and operating environment of gaming devices have become increasingly complex. This complexity is not only reflected in the collaborative work of multiple components and modules within the gaming device, but also involves the interaction between the gaming device and external networks, servers and other environments. In order to provide a more immersive and smooth gaming experience, the utilization of the hardware configuration of gaming devices, such as CPU, GPU and memory, continues to increase, and the software level also needs to run a large number of processes and real-time computing tasks. At the same time, the dependence of gaming devices on network connections has gradually increased. For example, functions such as online games, cloud games and real-time updates require gaming devices to have high-quality network communication capabilities.

[0003] As these requirements increase, gaming devices face greater challenges in stability and reliability in actual operation. High-load operation of hardware may lead to problems such as overheating and resource contention; complex task management and multi-task parallel processing at the software level may also easily cause memory leaks, process abnormalities, etc.; instability at the network level may lead to increased latency, data packet loss and other problems, affecting user experience. Therefore, real-time monitoring of the operating status of gaming devices and timely identification and handling of potential faults have become the key to ensuring the normal operation of gaming devices.

[0004] However, the existing game device fault monitoring methods have the following technical problems: they are unable to fully capture potential problems in the operation of game devices, are prone to false alarms or missed alarms when processing complex game device operation data, and are unable to dynamically adapt to changes in the status of game devices, resulting in a significant reduction in the timeliness and accuracy of fault warnings, and fail to fully utilize the potential correlations between various dimensions to improve the accuracy of fault detection; when dealing with dynamically changing game device operating environments, they lack adaptive capabilities and are usually unable to adjust model parameters in time to adapt to newly emerging data features, which makes them perform poorly when dealing with complex and dynamic game device failure scenarios and cannot meet the high precision and real-time requirements in practical applications. Summary of the invention

[0005] The present invention provides a game device fault monitoring method to solve the problems that the potential problems in the operation of the game device cannot be fully captured, false alarms or missed alarms are easily generated when processing complex game device operation data, and the game device status changes cannot be dynamically adapted, resulting in the timeliness and accuracy of fault warnings being greatly reduced, and the potential correlations between various dimensions cannot be fully utilized to improve the accuracy of fault detection; when processing the dynamically changing game device operation environment, there is a lack of adaptive capabilities, and the model parameters are usually unable to be adjusted in time to adapt to the newly emerging data features, which makes them perform poorly when dealing with complex and dynamic game device failure scenarios, and cannot meet the technical problems of high precision and real-time requirements in practical applications.

[0006] A method for monitoring a gaming device fault of the present invention specifically includes the following technical solutions:

[0007] A method for monitoring a gaming device fault comprises the following steps:

[0008] S1. Collect different operation data of the game device to construct a multi-dimensional time series data set; perform inter-dimensional cross-correlation recursive analysis on the multi-dimensional time series data set, and use multi-layer nested time window analysis to generate a nested time window matrix;

[0009] S2. Introduce the interaction matrix, process the nested time window matrix through a multi-layer adaptive interdependent network model, and generate a comprehensive node state vector; calculate the predicted fault score based on the comprehensive node state vector; and monitor the game device fault based on the predicted fault score.

[0010] Preferably, the S1 specifically includes:

[0011] In the process of recursive analysis of the inter-dimensional correlation of multidimensional time series data sets, a cross-correlation function is defined to identify dimension pairs with strong coupling relationships; the strong coupling relationships between dimensions are refined through multi-level recursive processing to obtain recursive results at different levels.

[0012] Preferably, the S2 specifically includes:

[0013] The multi-layer adaptive interdependent network model introduces an adaptive interdependent mechanism into the multi-layer network structure. By constructing a multi-layer adaptive interdependent network structure and processing the nested time window matrix, the node state vector of each layer of the network is obtained. The specific formula is as follows:

[0014]

[0015] Among them, h (l) (t) represents the node state vector of the lth layer network at time t; h (l)(t-1) represents the node state vector of the lth layer network at time t-1; represents the self-updated weight matrix of layer l; represents the input weight matrix of the lth layer; W (l) (t) is the nested time window matrix of level l; represents the integration of interdependent influences from layers other than layer l; s represents different network layer indices, which are used to traverse all other layers different from layer l; γ ls represents the interdependence weight from layer s to layer l; h (s) (t) represents the node state vector of the s-th layer network at time t; Ω ls represents the interaction matrix from layer s to layer l; L is the number of layers in the multi-layer adaptive interdependence network;

[0016] The comprehensive node state vector is generated by weighted summing of the node state vectors of each layer.

[0017] Preferably, the S2 specifically includes:

[0018] The synthesized node state vector is converted into a scalar fault score through a linear transformation.

[0019] Preferably, the S2 specifically includes:

[0020] Design the loss function of the multi-layer adaptive interdependent network model, and calculate the interaction matrix Ω ls to update.

[0021] Preferably, the S2 specifically includes:

[0022] By introducing the mean square error, robustness term and sparsity regularization term and performing weighted summation on the mean square error, robustness term and sparsity regularization term, the loss function of the multi-layer adaptive interdependent network model is obtained.

[0023] Preferably, the S2 specifically includes:

[0024] Based on the interaction matrix, an adaptive adjustment mechanism is introduced to make the interaction matrix dynamically adapt to the changes in the states of nodes in each layer; through iterative optimization, the interaction matrix is ​​adjusted according to the loss function of the multi-layer adaptive interdependent network model.

[0025] Preferably, the S2 specifically includes:

[0026] After obtaining the predicted fault score, a threshold is set, and the predicted fault score is compared with the threshold to determine whether the current gaming device is in a fault state.

[0027] The beneficial effects of the technical solution of the present invention are:

[0028] 1. The present invention deploys a variety of sensors to collect hardware signals, software signals and network signals of gaming devices in real time, constructs a multi-dimensional time series data set, and can identify the potential correlation between multi-dimensional data through complex inter-dimensional cross-correlation recursive analysis and multi-level recursive processing, thereby achieving early warning and accurate detection of gaming device failures.

[0029] 2. The introduction of a multi-layer adaptive interdependent network model can dynamically adjust the interdependence between layers according to changes in the state of the game device. By adaptively adjusting the interaction matrix, the multi-layer adaptive interdependent network model can gradually enhance or weaken the interaction between different layers, thereby better adapting to the complex operating environment of the game device and improving the accuracy of fault detection.

[0030] 3. The design of the loss function of the multi-layer adaptive interdependent network model of the present invention takes into account the accuracy of fault detection, the robustness of the prediction results and the sparsity of the interaction matrix. Through the mean square error, robustness term and sparsity regularization term, while improving the accuracy of fault detection, it effectively reduces the situation of false alarms and missed alarms, ensuring the reliability of fault warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The present invention is a flowchart of a method for monitoring faults of gaming devices. DETAILED DESCRIPTION

[0032] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0034] A specific scheme of a gaming device fault monitoring method provided by the present invention is described in detail below with reference to the accompanying drawings.

[0035] Refer to the attached Figure 1 , which shows a flow chart of a method for monitoring a gaming device failure provided by an embodiment of the present invention, the method comprising the following steps:

[0036] S1. Collect different operation data of the game device to construct a multi-dimensional time series data set; perform inter-dimensional cross-correlation recursive analysis on the multi-dimensional time series data set, and use multi-layer nested time window analysis to generate a nested time window matrix;

[0037] Multiple sensors are deployed to collect different operating data of the gaming device, including hardware signals, software signals, and network signals. The hardware signals mainly collect key hardware parameters of the gaming device, including CPU temperature, GPU usage, memory status, etc. The hardware signal data is acquired in real time through sensors, forming a basic data set for the hardware health status of the gaming device, which is used to monitor the basic operating status of the gaming device to ensure that early signs of hardware failure can be identified. The software signal acquisition focuses on operating system processes, log information, and memory usage. Through log analysis tools and process monitoring tools, data streams are acquired in real time to identify potential software failures, such as memory leaks, abnormal process terminations, etc., to provide a basis for subsequent software layer fault detection. The network signal acquisition mainly collects data such as network delay, packet loss rate, and bandwidth occupancy. The network signal is used to monitor the network communication quality of the gaming device to ensure that the gaming device can communicate stably with external servers or other devices to avoid failures caused by network problems.

[0038] The different operating data of the game devices collected form a multi-dimensional time series data set: X(t) = {x 1 (t), x 2 (t), ..., x m (t)}, where X(t) represents a multidimensional time series dataset at time t, x i (t) represents the data value of the i-th dimension of the multidimensional time series dataset at time t, and m is the total number of dimensions. The multidimensional time series dataset not only contains the current state of each dimension, but also can capture the correlation between the dimensions through subsequent processing.

[0039] In order to reveal the complex correlation between data in a multidimensional time series data set, a recursive cross-correlation analysis is performed on the multidimensional time series data set. The recursive cross-correlation analysis aims to identify the potential relationships in the multidimensional time series data. The formula is as follows:

[0040]

[0041] Among them, ρ i j(τ) is the cross-correlation function between the i-th dimension and the j-th dimension of the multidimensional time series data set at time t and t+τ; x j (t+τ) represents the data value of the th dimension of the multidimensional time series data set at time t+τ, where τ is the time delay; μ i and μ jare the means of the i-th dimension and the j-th dimension of the multidimensional time series data set respectively; σ i and σ j are the standard deviations of the i-th dimension and the ,th dimension of the multidimensional time series data set, respectively. By calculating the cross-correlation between different dimensions of the multidimensional time series data set, ρi is analyzed. j (τ) changes with the time delay τ, and finds the peak of the cross-correlation function. The position of the peak reflects the moment of maximum correlation between the i-th dimension and the i-th dimension, indicating that the two dimensions have a strong coupling relationship under the time delay τ, thereby identifying the dimension pairs with a strong coupling relationship on the time axis.

[0042] After identifying the strong coupling relationship, the strong coupling relationship between dimensions is further refined through multi-level recursive processing to make the detection result more accurate. The multi-level recursive formula is:

[0043]

[0044] Among them, R (k) (t) is the result of the k-th layer recursion; f is a nonlinear activation function used to enhance the nonlinear relationship that occurs during the recursive process; is the cross-correlation function in the k-1th recursive layer. Each recursive layer in the multi-level recursive processing depends on the output of the previous recursive layer, forming a recursive chain, which ensures the in-depth analysis of the relationship between the data of each dimension layer by layer.

[0045] After the cross-correlation recursive analysis is completed, the data pattern within the time range is further captured by introducing multi-layer nested time window analysis. The specific time range value will be determined according to the setting of the time window offset and the time point. It is usually set according to the application scenario and the characteristics of the game device, depending on the possible occurrence period of the failure in the specific application scenario. The nested time window matrix formula is:

[0046]

[0047] Among them, w l，n is the l-th nested time window offset, which determines the position of the n-th time point in the nested time window matrix relative to the current time t; W (l) (t) is the nested time window matrix of the lth layer, which represents the recursive result matrix at time t after time shift processing; R (k) (tw l，n ) is the k-th layer recursive result after the time window offset is w l，n The output after time offset processing is represented by l，n The nested time window matrix reveals dynamic patterns at different time scales by capturing the recursive results at multiple time points.

[0048] S2. Introduce the interaction matrix, process the nested time window matrix through a multi-layer adaptive interdependent network model, and generate a comprehensive node state vector; calculate the predicted fault score based on the comprehensive node state vector; and monitor the game device fault based on the predicted fault score.

[0049] The data in the nested time window matrix is ​​not isolated, but is processed by a multi-layer adaptive interdependent network model. The multi-layer adaptive interdependent network model introduces an adaptive interdependent mechanism in the multi-layer network structure, so that the output of each layer depends not only on its input, but also on the state of other layers, thereby capturing more complex temporal and dimensional associations. Specifically, when processing the nested time window matrix, the multi-layer adaptive interdependent network model constructs a multi-layer adaptive interdependent network structure. The node state of each layer of the network consists of two parts: its own network node state evolution and the influence of interdependence from other layers. The node state vector h of each layer l (l) The update formula of (t) is:

[0050]

[0051] Among them, h (l) (t) represents the node state vector of the lth layer network at time t; h (l) (t-1) represents the node state vector of the lth layer network at time t-1; represents the self-updated weight matrix of layer l; represents the input weight matrix of the lth layer; represents the integration of interdependent influences from other layers; s represents different network layer indices, which are used to traverse all other layers different from layer l; γ ls represents the interdependence weight from layer s to layer l; h (s) (t) represents the node state vector of the s-th layer network at time t; Ω ls represents the interaction matrix from layer s to layer l; L is the number of layers in the multi-layer adaptive interdependence network.

[0052] Combining the output of each layer, a comprehensive node state vector H(t) is generated by weighted summing the node state vectors of all layer networks:

[0053]

[0054] Among them, H(t) is a comprehensive node state vector, representing the comprehensive node state evaluation of the multi-layer adaptive interdependent network model at time t; α lIt is the weight coefficient of each layer of the network, which is optimized during the training process of the multi-layer adaptive interdependent network model to ensure that the comprehensive node state vector can fully reflect the information contribution of each layer of the network.

[0055] The output layer generates a predicted fault score based on the comprehensive node state vector H(t) The comprehensive node state vector is converted into a scalar fault score through linear transformation. The specific formula is as follows:

[0056]

[0057] Among them, ω o is the weight matrix of the output layer, responsible for mapping the high-dimensional comprehensive node state vector H(t) to a scalar output; b o is the bias term, used to adjust the baseline level of the scalar output; It is an activation function. A linear function or a nonlinear function, such as a sigmoid function, can be selected according to the specific application.

[0058] In the multi-layer adaptive interdependent network model, in order to adjust the interaction matrix Ω ls , a loss function needs to be designed to guide the optimization process. The design of the loss function is directly related to the optimization goal and performance of the multi-layer adaptive interdependent network model. The design of the loss function is based on the following key factors: the accuracy of fault detection, the robustness of the prediction results, and the sparsity of the interaction matrix.

[0059] Specifically, the main part of the loss function comes from the accuracy evaluation of the multi-layer adaptive interdependence network model on fault detection. Assuming that the actual fault state at time t is y(t), the predicted fault score of the multi-layer adaptive interdependence network model is The mean square error is used to measure the difference between the predicted fault score and the actual fault status:

[0060]

[0061] Among them, L mse is the mean square error; T is the total length of the time series; the above mean square error loss function ensures that the multi-layer adaptive interdependent network model can predict the fault state as accurately as possible.

[0062] In order to avoid the drastic fluctuation of the fault score predicted by the multi-layer adaptive interdependence network model in the time series, a robustness term L is introduced into the loss function. smooth , which is used to penalize the excessive change of the fault score predicted by the multi-layer adaptive interdependent network model between adjacent time points. The specific expression is:

[0063]

[0064] The introduced robustness term helps the multi-layer adaptive interdependent network model to output smoother predicted fault scores and avoid overfitting short-term noisy data.

[0065] In order to make the interaction matrix Ω ls More concise and explanatory, a sparsity regularization term L is introduced into the loss function sparse , the sparsity regularization term L sparse Based on the interaction matrix Ω ls L 1 Norm:

[0066]

[0067] By introducing sparsity constraints, the loss function encourages most elements in the interaction matrix to approach zero, so that only the most significant inter-layer dependencies are retained. This helps to improve the generalization ability of the multi-layer adaptive interdependence network model and simplify the structure of the interaction matrix, making the interaction matrix and fault prediction score easier to interpret.

[0068] The final comprehensive loss function is a weighted sum of the mean squared error, the robustness term, and the sparsity regularization term:

[0069]

[0070] in, is the loss function of the multi-layer adaptive interdependent network model; β 1 , β 2 and β 3 are the weight coefficients of the mean square error, robustness term, and sparsity regularization term, respectively, which can be adjusted experimentally to optimize the performance of the multi-layer adaptive interdependent network model.

[0071] During the optimization process, the loss function is minimized using the gradient descent method or its variants, such as the Adam optimizer, and the interaction matrix Ω is adaptively adjusted. ls .

[0072] In the multi-layer adaptive interdependent network model, the interaction matrix Ω ls Determines the interdependence between different layers, the interaction matrix Ω ls The generation and adaptive adjustment process of is as follows:

[0073] For each pair of layers l and s (where s ≠ l), construct the initial interaction matrix The initial construction is based on the correlation calculation of the node states between the two layers. The correlation coefficient matrix of the node states of the sth layer and the lth layer at time point t is calculated, and its elements are:

[0074]

[0075] in, represents the rth node of the lth layer and the sth node of the sth layer The correlation coefficient matrix between nodes; Represents the node state vector of the rth node in the lth layer at time t; represents the sth layer The node state vector of the node at time t; μl and μs are the means of the node state vectors of the lth layer and the sth layer respectively; T is the total length of the time series. The correlation coefficient matrix describes the linear dependency of the node states between the two layers.

[0076] Normalize the correlation coefficient matrix to obtain the initial interaction matrix The initial interaction matrix is ​​only a basic estimate. In order to enable the interaction matrix to dynamically adapt to the changes in the states of nodes in each layer, an adaptive adjustment mechanism is introduced. Through iterative optimization, the interaction matrix is ​​adjusted according to the loss function in the training process of the multi-layer adaptive interdependent network model.

[0077] Specifically, in each round of iteration, according to the loss function of the multi-layer adaptive interdependent network model The interaction matrix Ω ls To prevent the interaction matrix from excessively fluctuating or falling into local optimality during the training of the multi-layer adaptive interdependent network model, a stability constraint is introduced. A regularization term is added to the update process of the interaction matrix to balance the adjustment strength of the interaction matrix. The specific formula is:

[0078]

[0079] in, and are the interaction matrices before and after the update, respectively; η is the learning rate; is the gradient of the loss function with respect to the interaction matrix; is the regularization coefficient. The update rule enables the interaction matrix to gradually adapt to the complex dependencies between layers during the training of the multi-layer adaptive interdependent network model, and gradually strengthen or weaken the interactions between different layers. After multiple rounds of iterations and adaptive adjustments, the final interaction matrix not only reflects the static correlation between layers, but also captures the complex dependencies between the states of nodes in each layer during the dynamic training process.

[0080] Getting the predicted failure score After that, the predicted fault score is compared with the threshold value pre-set according to the expert experience method to determine whether the current gaming device is in a fault state. Once the gaming device is determined to be in a fault state, the early warning mechanism will be triggered to notify the maintenance personnel and start the detailed fault diagnosis procedure.

[0081] In summary, a method for monitoring faults of gaming devices is completed.

[0082] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0083] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0084] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for monitoring a gaming device failure, characterized in that: The following steps are involved: S1. Collect different operation data of game devices and construct a multi-dimensional time series data set; Perform inter-dimensional recursive analysis of the cross-correlation between dimensions on multi-dimensional time series data sets, and use multi-layer nested time window analysis to generate a nested time window matrix; S2. Introduce the interaction matrix, process the nested time window matrix through a multi-layer adaptive interdependent network model, and generate a comprehensive node state vector; calculate the predicted fault score based on the comprehensive node state vector; and monitor the game device fault based on the predicted fault score.

2. The method for monitoring a gaming device failure according to claim 1, wherein: The S1 specifically includes: In the process of recursive analysis of the inter-dimensional correlation of multidimensional time series data sets, a cross-correlation function is defined to identify dimension pairs with strong coupling relationships; the strong coupling relationships between dimensions are refined through multi-level recursive processing to obtain recursive results at different levels.

3. The method for monitoring a gaming device failure according to claim 1, wherein: The S2 specifically includes: The multi-layer adaptive interdependent network model introduces an adaptive interdependent mechanism into the multi-layer network structure. By constructing a multi-layer adaptive interdependent network structure and processing the nested time window matrix, the node state vector of each layer of the network is obtained. The specific formula is as follows: Among them, h (l) (t) represents the node state vector of the lth layer network at time t; h (l) (t-1) represents the node state vector of the lth layer network at time t-1; represents the self-updated weight matrix of layer l; represents the input weight matrix of the lth layer; W (l) (t) is the nested time window matrix of level l; Ω ls represents the integration of interdependent influences from layers other than layer l; s represents different network layer indices, which are used to traverse all other layers different from layer l; γ ls represents the interdependence weight from layer s to layer l; h (s) (t) represents the node state vector of the s-th layer network at time t; Ω ls represents the interaction matrix from layer s to layer l; L is the number of layers in the multi-layer adaptive interdependence network; The comprehensive node state vector is generated by weighted summing of the node state vectors of each layer.

4. The method for monitoring a gaming device failure according to claim 3, wherein: The S2 specifically includes: The synthesized node state vector is converted into a scalar fault score through a linear transformation.

5. The method for monitoring gaming device failure according to claim 1, wherein: The S2 specifically includes: Design the loss function of the multi-layer adaptive interdependent network model, and calculate the interaction matrix Ω ls to update.

6. The method for monitoring a gaming device failure according to claim 5, wherein: The S2 specifically includes: By introducing the mean square error, robustness term and sparsity regularization term and performing weighted summation on the mean square error, robustness term and sparsity regularization term, the loss function of the multi-layer adaptive interdependent network model is obtained.

7. The method for monitoring a gaming device failure according to claim 6, wherein: The S2 specifically includes: Based on the interaction matrix, an adaptive adjustment mechanism is introduced to make the interaction matrix dynamically adapt to the changes in the states of nodes in each layer; through iterative optimization, the interaction matrix is ​​adjusted according to the loss function of the multi-layer adaptive interdependent network model.

8. The method for monitoring a gaming device failure according to claim 1, wherein: The S2 specifically includes: After obtaining the predicted fault score, a threshold is set, and the predicted fault score is compared with the threshold to determine whether the current gaming device is in a fault state.

Images