Adaptive model updating method for full-data AI control system

Abstract

The present invention relates to the field of computer system technology, and further to an adaptive model updating method for a full-data AI management and control system. It includes the following steps: Step 1: Acquire the original data of the system, perform tensor decomposition on the original data, extract spatiotemporal features, and obtain a characteristic tensor; through the interaction of the vorticity field and the nonlinear function, project the characteristic tensor into a nonlinear dynamic space to obtain a mapping result; Step 2: Perform dynamic evolution of the characteristic flow on the mapping result to obtain an evolution result; output the prediction model output; calculate the entropy increase rate of the system; Step 3: According to the entropy increase rate, calculate the feedback correction value of the system; according to the feedback correction value, perform an adaptive model update of the system. Through the adaptive feedback mechanism of the present invention, the system can achieve efficient and real-time dynamic adjustment in a complex and changeable environment, significantly improving the prediction accuracy and stability of the system.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer systems, and specifically relates to an adaptive model updating method for a full-data AI management and control system. Background Art

[0002] In current data-intensive application scenarios, with the deepening of informatization and intelligence, how to effectively monitor and manage complex and changeable data has become a core issue faced by various systems. Especially in a full-data environment, the system needs to process massive amounts of data in real time, while traditional static control and simple prediction models can no longer meet the application requirements in dynamic and complex environments. Many traditional systems often rely on fixed model parameters, lack adaptive adjustment capabilities, and are difficult to respond efficiently in rapidly changing environments. In order to meet this challenge, researchers have proposed a variety of intelligent control systems based on AI and big data analysis in recent years to improve the system's adaptability and prediction capabilities to dynamic data environments. However, although these technologies have achieved data analysis and feedback to a certain extent, there are still many shortcomings in complex environments, especially in dealing with dynamic environments with high complexity and high uncertainty.

[0003] At present, traditional control systems often use static modeling methods to establish parameterized models for specific application scenarios and use historical data to predict future states. These models are mainly based on methods such as linear regression, decision trees or simple time series analysis. These methods can provide reasonable predictions in scenarios where data changes are relatively stable, but static models are difficult to adapt to complex dynamic needs when data fluctuates greatly or the environment changes frequently. This is mainly because static models rely on historical data and fixed parameters. Once the environment changes, the effectiveness of model parameters will decrease significantly. Therefore, traditional static models often have problems such as inaccurate predictions and delayed responses in rapidly changing environments, making it difficult to meet actual needs. In order to solve this problem, researchers have begun to explore control systems based on adaptive technology. These systems can automatically adjust model parameters based on real-time data feedback to improve the dynamic adaptability of the system. For example, some systems use adaptive filters to track changes in input data in real time, adjust model parameters through feedback, and enhance the response speed of the system. However, these adaptive methods are often only effective in specific application scenarios. Especially in high-dimensional data spaces and complex environments, the adjustment range and adjustment rate of traditional adaptive methods are greatly limited. In addition, the feedback mechanism of many adaptive systems is limited to linear feedback, and the processing ability of nonlinear and complex data association is limited. Therefore, these traditional adaptive methods still find it difficult to provide accurate prediction and feedback in highly nonlinear dynamic environments. In complex data environments, data changes are often not simple linear relationships, but nonlinear dynamic processes affected by multiple factors. To this end, some studies have attempted to apply nonlinear models to control systems, such as using nonlinear algorithms such as neural networks and support vector machines to model, which can improve the system's ability to capture complex data to a certain extent. However, traditional nonlinear models usually require a large amount of data to train, and the training process is complex, consumes a lot of computing resources, and is difficult to adjust quickly in a real-time environment. In addition, although models such as neural networks have certain learning and adaptation capabilities, they often require high computing costs and storage resources, and cannot perform efficient feedback corrections to environmental changes. In a multi-dimensional, highly complex data environment, nonlinear models such as neural networks have strong update lags and are difficult to adapt to high-frequency real-time changes. Summary of the invention

[0004] The main purpose of the present invention is to provide an adaptive model updating method for a full-data AI control system, which enables the system to achieve efficient, accurate, and stable adaptive control in a complex and changeable data environment. Through this method, the present invention significantly improves the prediction accuracy, dynamic adaptability, and stability of the full-data AI control system, and can ensure the smooth operation of the system in a highly complex and uncertain environment, providing strong support for various intelligent and information-based applications.

[0005] In order to solve the above problems, the technical solution of the present invention is achieved as follows:

[0006] The adaptive model updating method of the full data AI control system includes the following steps:

[0007] Step 1: Obtain the original data of the system, perform tensor decomposition on the original data, extract the spatiotemporal features, and obtain the characteristic tensor; through the interaction between the vorticity field and the nonlinear function, project the characteristic tensor into a nonlinear dynamic space to obtain the mapping result;

[0008] Step 2: dynamically evolve the feature flow of the mapping result to obtain the evolution result; input the evolution result into the preset prediction model and output the prediction model output; calculate the entropy increase rate of the system according to the prediction model output;

[0009] Step 3: Calculate the feedback correction value of the system based on the entropy increase rate; update the adaptive model of the system based on the feedback correction value, and calculate the overall stability of the system. When the overall stability of the system is lower than the set threshold, issue an early warning.

[0010] Furthermore, in step 1, the following formula is used to decompose the original data into tensors:

[0011] ;

[0012] in, is the original data; and Respectively represent the original data in A pair of eigenvalues ​​corresponding to each other in dimension; Represents the decomposition rank The summation operation, Represents the rank of tensor decomposition, that is, the number of layers of tensor decomposition; Represents the characteristic dimension of the original data; and All are integer subscript indices; Indicates Dimensional characteristic curve The closed path integral of ; Indicates The characteristic curve on the dimension is used to capture the changes in the data on that dimension; The original data dimensional feature vector The second time derivative of ; The original data is along the dimensional feature vector; is the time scale factor, For the The time decay factor of the dimension, is the time difference; Represents the tensor product operation of vectors; is the curve integral differential, which means A tiny unit for integration on a dimensional characteristic curve.

[0013] Furthermore, in step 1, the following formula is used to project the characteristic tensor into a nonlinear dynamic space through the interaction between the vorticity field and the nonlinear function to obtain the mapping result:

[0014] ;

[0015] in is the mapping result; Represents the feature space The integral of Represents the feature tensor The feature space in which it is located; represents the curl operator; is the cross product operator; Used to calculate the vorticity field to capture the characteristic tensor In feature space The rotation behavior in generates a vorticity field; is the integral differential element of the feature space; represents the vorticity field, which is a vector field used to capture the rotational dynamic characteristics of the system; The calculation of combines the characteristic tensor with the vorticity field to generate a quantity that reflects the rotational behavior of the characteristic tensor; is the dimension of the feature tensor; express In the The weight of the dimension; Indicates dimensional spatial coordinates; is the number of nonlinear functions; and All are integer subscript indices; Indicates A non-linear function.

[0016] Furthermore, in step 2, the mapping result is dynamically evolved by the following formula to obtain the evolution result:

[0017] ;

[0018] in, As a result of evolution; is the current time; is the characteristic flow surface; is the integral differential element of the characteristic flow surface; is the feature tensor in Dimensional scaling factor; is the evolution time; is the evolution time integral differential.

[0019] Furthermore, in step 2, the evolution result is input into the preset prediction model through the following formula, and the prediction model output is output:

[0020] ;

[0021] in, Indicated in The prediction region in the dimension is used to limit the spatial range of the prediction, which is a local region set in the feature space; The original data is along the dimensional feature vector; represents the Laplace operator; The mapping result is Dimensional component; Indicated in The integral differential element of the prediction region in dimension; Indicates prediction functions; types of prediction functions include: polynomial function, exponential function and / or radial basis function based on Gaussian distribution; Output of the prediction model.

[0022] Furthermore, in step 2, the entropy increase rate of the system is calculated based on the output of the prediction model using the following formula:

[0023] ;

[0024] in, is the entropy increase rate; is the amplitude of the evolution result; The feature space The boundary normal vector of is the information flow density of the system; Output of the prediction model No. Dimensional component; The mapping result is Dimensional Component The square of the magnitude of the gradient.

[0025] Furthermore, in step 3, according to the entropy increase rate, the feedback correction value of the system is calculated using the following formula:

[0026] ;

[0027] in, is the time kernel function; The evolution result Dimensional component; is the feedback correction value.

[0028] Furthermore, in step 3, the adaptive model of the system is updated according to the feedback correction value through the following formula:

[0029] ;

[0030] in, Represents a system parameter set; is the learning rate.

[0031] Furthermore, in step 3, the overall stability of the system is calculated by the following formula:

[0032] ;

[0033] in, Represents the prediction area in all dimensions.

[0034] The adaptive model updating method of the full data AI control system of the present invention has the following beneficial effects:

[0035] First of all, the present invention effectively captures the spatiotemporal characteristics of multidimensional data by introducing the decomposition and dynamic evolution of feature tensors. Through tensor decomposition technology, the system can decompose high-dimensional raw data into several feature dimensions, and perform detailed analysis of these dimensions to obtain a more accurate feature representation. Traditional systems usually only focus on a single dimension or a small number of key dimensions, and it is difficult to fully capture the multidimensional correlation of data. The present invention not only performs tensor decomposition on the data in the initial stage, but also projects the feature tensor into the nonlinear dynamic space to form a vorticity field that contains rotation and flow characteristics, so as to better reveal the dynamic behavior and complex correlation of the data. This design enables the system to more accurately characterize the nonlinear relationship between multidimensional features, providing in-depth information support for subsequent dynamic evolution and adaptive adjustment.

[0036] Secondly, the present invention significantly improves the adaptive ability of the system through multi-level feedback correction and adaptive model update. The adaptive model in the traditional control system is usually limited by a fixed feedback mechanism, and it is difficult to achieve true adaptive adjustment in a complex environment. The present invention incorporates information complexity and uncertainty into feedback regulation through the calculation of the entropy increase rate, so that the system can perceive the uncertainty changes of the data in real time. Specifically, the entropy increase rate quantifies the complexity and volatility of the current data state. When the system detects that the entropy increase rate is high, it can quickly judge that the complexity of the current environment has increased, and then increase the adaptive adjustment efforts. This dynamic adjustment mechanism ensures that the system can respond quickly when the data fluctuates violently, improves the robustness and stability of the prediction model, and enables the system to operate flexibly in various environments.

[0037] In addition, the calculation of feedback correction values ​​introduces the time kernel function and the weight analysis of multi-dimensional features, so that the system can make detailed adjustments according to the changing trend of each feature dimension. The design of the time kernel function allows the system to increase feedback adjustments for feature dimensions that have changed significantly recently, thereby improving the system's adaptability to short-term changes; while for stable feature dimensions, the system's stability is maintained by reducing the intensity of feedback correction. The adjustment of multi-dimensional feedback weights enables the system to distinguish the dynamic changes of different features in the multi-dimensional feature space, avoiding the problem of over-balanced adjustments of each feature in traditional methods that lead to overall fluctuations in the system. This innovative design greatly improves the system's adaptability to high-dimensional data environments and feedback accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of an adaptive model updating method for a full-data AI management and control system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0040] Example 1, reference Figure 1 : The adaptive model updating method of the full-data AI control system includes the following steps:

[0041] Step 1: Obtain the original data of the system, perform tensor decomposition on the original data, extract the spatiotemporal features, and obtain the characteristic tensor; through the interaction between the vorticity field and the nonlinear function, project the characteristic tensor into a nonlinear dynamic space to obtain the mapping result;

[0042] Once the feature tensor is generated, the system projects it into the nonlinear dynamic space to better express the complex behavior of the data in the space-time dimension. The projection process utilizes the interaction between the vorticity field and the nonlinear function, allowing the system to more accurately capture the dynamic characteristics of the data. The vorticity field is a mathematical model used to describe the rotation and flow behavior of the data features in the system. By introducing the vorticity field into the feature tensor, the system can identify the rotation pattern and local change trend of the data in the space-time dimension, so as to better understand how the data evolves over time in different spatial dimensions. At the same time, the introduction of nonlinear functions provides the system with the ability to simulate complex nonlinear relationships. In practical applications, the change of data is usually not a linear relationship, but a nonlinear process affected by multiple factors. Therefore, by applying nonlinear functions to the feature tensor, the system can establish a more realistic dynamic model in the nonlinear dynamic space. The construction of this dynamic space allows the system to make more accurate predictions about the future state of the data, while enhancing the adaptability and robustness of the system in complex environments. Through this combination of tensor decomposition and nonlinear dynamic projection, the design of step 1 can not only effectively extract the multidimensional features in the original data, but also capture the behavior pattern of the data in a dynamic environment. This is different from traditional linear modeling methods, which usually assume that data changes are linear and have difficulty dealing with nonlinear relationships in practical applications. The advantage of this mechanism is that it provides a more expressive basis for the prediction and adaptive adjustment of the system, so that it can cope with a variety of complex and changing scenarios. At the same time, the projection of the feature tensor in the dynamic space is not a static process, but is continuously adjusted according to the changes in data at different time points. This means that the system can continuously update the feature tensor in a dynamic data environment to maintain adaptability to real-time data. After the projection is completed, the data features retained in the feature tensor can directly support the subsequent dynamic evolution of the feature flow, the construction of the prediction model, and feedback correction, thereby realizing the adaptive update of the system.

[0043] Step 2: dynamically evolve the feature flow of the mapping result to obtain the evolution result; input the evolution result into the preset prediction model and output the prediction model output; calculate the entropy increase rate of the system according to the prediction model output;

[0044] The dynamic evolution of feature flow mainly extracts the flow pattern and distribution characteristics of data changing over time by continuously analyzing the evolution trend of mapping results in time and space. Because data in a full data environment is usually highly diverse and complex, the system needs to capture the multi-scale features in the data in a fine manner through a dynamic evolution process. This multi-scale feature includes various rates of change of data at different time scales, such as rapid fluctuations, slow trends, and periodic changes. Through the dynamic evolution of feature flow, the system can effectively decompose the short-term and long-term changes in the feature flow, so that the information on these different time scales can be independently captured by the system and analyzed in depth. In the dynamic evolution process of feature flow, the system continuously iterates the mapping results to generate feature evolution results that are more adapted to the environment. This process fully considers the spatiotemporal dependence and dynamic characteristics of data. The mapping result is essentially a multi-dimensional dynamic representation that reflects the evolution of data over time in a complex environment. Dynamic evolution introduces a time scale factor, which enables the system to perform weighted analysis on feature flow according to the rate of change of data at different times. This weighted analysis provides the system with feature weights of data in multi-scale time dimensions, allowing the system to find the right balance between fast-changing and slow-changing trends. At the same time, dynamic evolution also takes into account the distribution characteristics of data in spatial dimensions, such as differences in the distribution of data in different locations and regions. To achieve this, the system maps different dimensions of data to various sub-regions in the feature space through a distributed computing framework. The differences between these sub-regions can be identified and captured by the dynamic evolution process, thereby further improving the system's adaptability to dynamic changes in data.

[0045] The dynamic evolution of feature flow is not just a passive data processing process. It also combines nonlinear dynamic models to make the evolution of feature flow more predictive and adaptive. The introduction of nonlinear dynamic models enables the system to identify complex nonlinear relationships in the data, such as the mutual influence and feedback mechanism between feature flows in different regions. Through this nonlinear modeling, the system can capture complex nonlinear interactions in the dynamic evolution of feature flow and provide more accurate data input for subsequent prediction models. For example, rapid changes in data in one area may cause changes in data in another area, and this interaction is difficult to capture with traditional linear models. Through this enhancement of nonlinear dynamics, the dynamic evolution of feature flow not only provides a tool to describe the spatiotemporal changes of data, but also a key support for the adaptive adjustment of the system in complex environments. After the dynamic evolution results of feature flow are calculated, the system will input these results into the preset prediction model. Based on these dynamically evolving data, the prediction model can generate more accurate future state predictions, providing important data support for the feedback correction of the system. The dynamic evolution results contain the trend prediction information of the feature flow in the future, so the system can capture possible trends in advance before the future state changes, so that the system can quickly adjust when the actual state changes. In addition, the dynamic evolution of the feature flow also helps the system understand the local details of data changes, such as local peaks and valleys of data at different time points. This information is extremely critical for the prediction model because it determines whether the model can identify the significant change characteristics of the data and make corresponding predictions and responses in time. This dynamic evolution process of the feature flow also provides important support for the adaptability of the system. Due to the strong mobility and randomness of data in the full data environment, the prediction and feedback models of the system need to be constantly adjusted to adapt to these changes. In the dynamic evolution process of the feature flow, the system can adjust the dynamic parameters of the model in time according to the current state and change speed of the data, so as to ensure that the prediction model can run stably in a changing environment. More importantly, this dynamic evolution mechanism can not only identify the periodic changes and trend changes of the data, but also issue warnings when there are sudden fluctuations in the data, so that the system can respond in advance. For example, when unusually high fluctuations occur in the data stream, the system can make additional adjustments in the predictive model to account for these sudden changes, thereby ensuring the overall stability of the system.

[0046] Step 3: Calculate the feedback correction value of the system based on the entropy increase rate; update the adaptive model of the system based on the feedback correction value, and calculate the overall stability of the system. When the overall stability of the system is lower than the set threshold, issue an early warning.

[0047] Specifically, after the system obtains the output of the prediction model, by calculating the entropy increase rate of the current state, the system can further determine the degree of uncertainty of the data and decide whether adjustments are needed. The entropy increase rate here acts as an indicator of the system state, used to measure the uncertainty or complexity of the system information in the current state. The higher the entropy increase rate, the greater the uncertainty and volatility of the system, indicating that the current environment is changing rapidly and the credibility of the system prediction results may be low. Therefore, the system can adjust the intensity of feedback correction according to the entropy increase rate, thereby providing decision support for subsequent adaptive model updates.

[0048] After obtaining the entropy increase rate, the system starts to calculate the feedback correction value. The feedback correction value is the result of quantifying the deviation between the system's predicted output and the actual state, reflecting the gap between the current prediction model and the actual state. Through this feedback correction value, the system can evaluate the error size of the current model and the directionality of the deviation. This feedback correction mechanism is the core of the entire system to achieve adaptability. By making real-time corrections to the actual deviation of the prediction results, the system can more effectively respond to dynamic changes in data and reduce prediction errors caused by environmental changes. In the process of feedback correction, the system will also consider the characteristics of different data dimensions, such as the frequency of data changes and fluctuations, so as to apply different feedback weights to different dimensions, so that the correction value is more in line with the actual situation of the current data. This multi-dimensional feedback correction design enables the system to not only identify global deviations, but also perform local feedback adjustments based on different features, thereby improving prediction accuracy. Once the feedback correction value is calculated, the system will perform adaptive model updates based on this correction value. Adaptive model update means that after receiving feedback information, the system automatically adjusts the model parameters or structure so that the model can better adapt to changes in the current environment. Traditional models usually require manual adjustment or rely on fixed parameters, but in the adaptive model update method of the full-data AI control system, the system can automatically optimize itself so that the model's predictive performance always remains at its best in a changing environment. Through adaptive model updates, the system can automatically adjust the model's learning rate, weights and other parameters when faced with rapid changes or emergencies, further improving its adaptability to complex environments. Especially in scenarios with high data mobility and large changes, the adaptive update mechanism can help the system continuously optimize parameters based on real-time data changes, thereby providing higher prediction accuracy and system stability in different scenarios.

[0049] In addition, the system after the adaptive model is updated will further calculate the overall stability. Overall stability is an important evaluation indicator, which reflects whether the current model has reached an ideal equilibrium state after adjustment. When the system undergoes multiple adaptive adjustments, the system may become unstable due to frequent model updates, and the calculation of overall stability is to ensure that the system can remain stable at the new equilibrium point after adjustment. The calculation of overall stability is based on factors such as the system entropy increase rate, feedback correction value, and parameter adjustment amount of the model, which comprehensively reflects the reliability of the system in the current state. When the system detects that the overall stability is lower than the preset threshold, it will automatically trigger the early warning mechanism to remind the system of potential instability. At this time, the system can choose to further increase the intensity of adaptive adjustment, or take other measures (such as increasing data filtering, reducing model sensitivity) to ensure that the system returns to a stable state. Through the comprehensive calculation of feedback correction, adaptive model update and overall stability, step 3 implements a closed-loop control mechanism of adaptive regulation and early warning in the system. This mechanism is significantly different from the traditional static control system. Traditional systems usually rely on fixed feedback control parameters and cannot respond to rapid fluctuations in real-time data. The adaptive model update mechanism of the full-data AI control system can dynamically respond to data changes by adjusting feedback correction values ​​and model parameters in real time, maintaining the robustness and stability of the system in a highly complex environment. This design not only enhances the system's self-regulation ability in a changing environment, but also significantly improves the system's sensitivity to prediction errors, ensuring that the system can quickly adjust to the optimal state in abnormal fluctuations or emergencies.

[0050] Example 2: In step 1, the following formula is used to perform tensor decomposition on the original data:

[0051] ;

[0052] in, is the original data; and Respectively represent the original data in A pair of eigenvalues ​​corresponding to each other in dimension; Represents the decomposition rank The summation operation, Represents the rank of tensor decomposition, that is, the number of layers of tensor decomposition; Represents the characteristic dimension of the original data; and All are integer subscript indices; Expressing the Dimensional characteristic curve The closed path integral of ; Indicates The characteristic curve on the dimension is used to capture the changes in the data on that dimension; The original data dimensional feature vector The second time derivative of ; The original data is along the dimensional feature vector; is the time scale factor, For the The time decay factor of the dimension, is the time difference; Represents the tensor product operation of vectors; is the curve integral differential, which means A tiny unit for integration on a dimensional characteristic curve.

[0053] Specifically, in this formula, the original data is decomposed into a multidimensional tensor through decomposition and summation operations ,in Represents the original data, which contains multi-dimensional information. The essence of tensor decomposition is to disassemble high-dimensional data layer by layer, so that the system can gradually extract the features of a specific dimension from the overall structure of the data. This hierarchical decomposition method uses progressive summation and product operations layer by layer, so that the system can identify the main features of the data in a specific dimension in each layer while maintaining the integrity of the data. Therefore, It is not only the result of data decomposition, but also the basic structure for systematic understanding and analysis of complex data. Represents the decomposition rank The summation operation, Determines the depth of the decomposition level, and each level represents a part of the characteristic structure in the original data. Through such hierarchical decomposition, the system can extract the features of the data at different levels and integrate these features to build an overall dynamic model. At the same time, Represents the item-by-item product of the data on the feature dimension, where It represents the characteristic dimension of the data, that is, the number of independent dimensions involved in the original data. Each characteristic dimension corresponds to the performance of the data in a specific direction, such as the dimension of physical quantities such as temperature and pressure. This item-by-item product operation ensures that the system fully accumulates the characteristic information of each characteristic dimension during the decomposition process, so that the system can not only identify the data features of a single dimension, but also capture the interaction between data in different dimensions through the interaction of multi-dimensional features.

[0054] In the formula Expressing the The closed path integral of the dimensional characteristic curve is used to capture the changing trend of the data in that dimension. The closed path integral enables the system to identify the overall characteristics of a specific dimension, rather than just the instantaneous value of a specific point. It is a characteristic curve of the data in the entire dimension. Integrating over this ensures that the system can extract valuable feature information from the overall changes in the data. This operation is equivalent to a global scan in the feature space, integrating the data fluctuations and change patterns along the entire path, making the system's understanding of each feature dimension more comprehensive. During the tensor decomposition process, the second-order time derivative is also introduced. which is used to describe the acceleration characteristics of the eigenvector . The use of the second derivative not only focuses on the instantaneous values of the data but also on the trend of its rate of change. For a system with dynamic changes, acceleration reflects the change trend and volatility of the data over time, revealing the inherent dynamics of the data in this dimension. This dynamic information is crucial for the subsequent adaptive update mechanism because it enables the system to perceive the direction and rate of data changes, and thus make corresponding predictions and adjustments based on the current data trend. In the formula, is the time scale factor, where is the time decay factor, is the time interval. The introduction of this time scale factor endows the data with weighted characteristics in time. By performing exponential decay processing on the data during the calculation, the system can assign higher weights to recent data changes and gradually reduce the influence of earlier data. This design enables the system to focus more on the current state rather than historical data when processing spatio-temporal data in a dynamic environment, thereby improving the model's sensitivity to current changes. This time-weighting mechanism also ensures that the system can respond more promptly to data fluctuations in a rapidly changing environment, and can smoothly analyze the long-term trends of the data when the data changes are relatively stable. In the formula, and represent the corresponding eigenbasis vectors of the data in the dimension. Through the tensor product operation of the vectors, the feature structure of this dimension can be obtained. The role of the tensor product operation here is to combine the eigenbases of the data in different dimensions with each other, thereby constructing a multi-dimensional tensor, enabling the system to understand the features of the data in a multi-dimensional space. The existence of the vector basis provides a reference direction for the system in each dimension, enabling it to capture data changes in a single dimension and, through the combination of these basis vectors, obtain a more three-dimensional feature structure. Finally, represents the feature curve, while is the integral and differential unit, meaning that every tiny region on the curve is incorporated into the calculation. This meticulous integration operation ensures that every detail in the feature dimension is noticed by the system. In particular, some subtle fluctuations and changes can be captured through such curve integration.

[0055] Example 3: In step 1, the following formula is used to project the characteristic tensor into a nonlinear dynamic space through the interaction between the vorticity field and the nonlinear function to obtain a mapping result:

[0056] ;

[0057] in is the mapping result; Represents the feature space The integral of Represents the feature tensor The feature space in which it is located; represents the curl operator; is the cross product operator; Used to calculate the vorticity field to capture the characteristic tensor In feature space The rotation behavior in generates a vorticity field; is the integral differential element of the feature space; represents the vorticity field, which is a vector field used to capture the rotational dynamic characteristics of the system; The calculation of combines the characteristic tensor with the vorticity field to generate a quantity that reflects the rotational behavior of the characteristic tensor; is the dimension of the feature tensor; express In the The weight of the dimension; Indicates dimensional spatial coordinates; is the number of nonlinear functions; and All are integer subscript indices; Indicates A non-linear function.

[0058] Specifically, in this formula, the feature tensor is first introduced , which contains the multi-dimensional feature information extracted by the previous tensor decomposition. In the process of projecting the feature tensor into the nonlinear dynamic space, the system uses the concept of vorticity field. The vorticity field is a mathematical field used to describe the rotation and flow characteristics of data, through which the rotation behavior of the feature tensor in the feature space can be captured. The curl operator and the cross product operation in the formula form the vorticity operator , acts on This operation combines the eigenvalue tensor with the vorticity field vector Combined with the above, a dynamic expression of rotation behavior is generated. This rotation behavior reflects how the feature data flows and changes in space, thereby revealing the dynamic characteristics of the data in multi-dimensional space. The overall integration operation accumulates all rotational features into an overall expression of the vorticity field, allowing the system to understand the flow and rotation patterns of the data at a higher level. This feature is particularly important because rotation and flow behavior are often manifestations of complex dynamic relationships in the system and can provide key insights into data change trends. In addition to the introduction of the vorticity field, this mapping also uses nonlinear functions to supplement the analysis of the dynamic changes of the characteristic tensor. In practical applications, data changes are often affected by multiple nonlinear factors, and simple linear expressions are difficult to capture the complex interactions in the data. Therefore, the second part of the formula contains a combination of the spatial derivatives of the characteristic components and nonlinear functions. Each characteristic component In its space coordinates The second derivative on Used to capture the acceleration or deceleration trend of the feature in space, this spatial derivative provides information about the diffusion, concentration or smooth change of the data. Performing second-order spatial derivative analysis on the data enables the system to identify subtle changes in the distribution of data in space, and through this analysis of the spatial change rate, the system can more clearly see the trend characteristics of the data at different locations, especially when the data fluctuates violently in space, this part of the calculation is particularly important.

[0059] Furthermore, the family of nonlinear functions The introduction of gives the formula the ability to handle time dynamics. The effect of each nonlinear function on the time derivative reflects a specific nonlinear change relationship. For example, some nonlinear functions can simulate periodic characteristics, such as sine or cosine functions; while other functions can simulate rapid increases and decreases, such as exponential functions or logarithmic functions. By applying these nonlinear function families to the time derivatives one by one, the system obtains a more diverse dynamic performance in the mapping results, and can capture various change patterns of data over time. Since different nonlinear functions have different response characteristics, the system can select appropriate nonlinear functions according to the data characteristics and build a mapping model covering a variety of dynamic behaviors. This approach provides the system with the flexibility to deal with nonlinear changes in data. In a complex data environment, the system can dynamically respond to various changing trends in the data stream. In addition, this mapping operation combines the rotational behavior of the vorticity field and the time dynamics of the nonlinear function, allowing the system to achieve joint analysis of dynamic features in time and space. By combining the feature tensor The rotation characteristics and nonlinear changes are combined to map the results It is not just a result of data projection, but a comprehensive model that includes dynamic rotation and nonlinear changes. This model provides a basis for the subsequent dynamic evolution of feature flow, allowing the system to further dynamically track and predict the mapping results. This mapping operation enables the system to accurately identify the relationship between features in multidimensional space, so that the core dynamic characteristics of the data can still be captured when the data undergoes complex changes.

[0060] Embodiment 4: In step 2, the mapping result is dynamically evolved by the feature flow through the following formula to obtain the evolution result:

[0061] ;

[0062] in, As a result of evolution; is the current time; is the characteristic flow surface; is the integral differential element of the characteristic flow surface; is the feature tensor in Dimensional scaling factor; is the evolution time; is the evolution time integral differential.

[0063] Specifically, the mapping results It is obtained by projecting the feature tensor through the nonlinear dynamic space, which contains the dynamic information of the data in the multi-dimensional space. When performing feature flow evolution, the system further processes this mapping result to understand how the data changes in time and space. The first part of the formula contains a surface integral operation , which is used to describe the characteristic flow on the characteristic surface Distribution and variation on the characteristic flow surface is a multidimensional space surface containing the mapping results, and the integral differential This allows the system to accumulate all feature changes on the surface. With its gradient The system captures the rate and direction of change of the feature flow on the surface, revealing the diffusion or concentration trend of the features in different spatial regions. In this way, the system can identify the dynamic change pattern of the features in the spatial dimension. This pattern provides spatial trend information for the prediction model, enabling the system to better understand how the features flow in space. The second part of the formula describes the dynamic evolution of the mapping result in the time dimension. This part is done through the time integration operation. Here, represents the evolution time, and the current time is the upper limit of the integral, representing the evolution process from the initial moment to the current time point. It is the first-order derivative of the mapping result in time, representing the rate at which the feature flow changes over time. During the evolution process, the system uses the time derivative to capture the instantaneous changes of the feature flow, allowing the system to identify how the feature flow evolves at different times. This temporal evolution information is critical to the real-time prediction of the system because it provides the system with an estimate of the future state.

[0064] In addition, the scale factor is introduced into the formula , which is used to adjust the weight of each feature dimension. Feature Tensor Indicates that the original data is The weight of the dimension, The weighting is performed according to the importance of each dimension. This weighting method allows the system to flexibly distinguish features of different dimensions during the dynamic evolution of the feature flow. For example, if a certain dimension has a greater impact on the system state, then the scale coefficient of this dimension will be higher, so that the feature changes in this dimension will receive more attention in the dynamic evolution. In this way, the system can effectively avoid the drawback of treating all features equally, so that more critical features occupy a greater weight in the system evolution, thereby improving the overall prediction accuracy. Used to accumulate changes in feature flows over time, ensuring that the system can gradually accumulate and track the dynamic evolution of features. This process not only focuses on the instantaneous changes in feature flows, but also analyzes its cumulative effects over a longer time scale. Over time, the system can obtain global evolution trends in feature flows by continuously accumulating feature change information. This cumulative dynamic evolution analysis allows the system to adapt to rapidly changing environments, while also identifying stable trends in data over longer time scales. This is especially important for full-data AI management and control systems, because the system needs to find a balance between short-term fluctuations and long-term trends in data changes in order to make stable and accurate predictions. Through this feature flow dynamic evolution formula, the system generates an evolution result that includes temporal and spatial change characteristics. . This evolution result is not only a reflection of the current data state, but also the basis for the system to predict the future trend of data changes. It provides a reasonable estimate of the future state for the full-data AI management and control system, and thus provides reliable data support for adaptive model updates and feedback mechanisms. The generation of evolution results enables the system to have stronger dynamic prediction capabilities and can more effectively deal with uncertain factors in the environment. This design also makes the system more stable in complex scenarios, and can quickly adjust its own state when the data changes drastically, thereby maintaining the efficient operation of the system.

[0065] Example 5: Input the evolution results into a preset prediction model and output the prediction model output:

[0066] ;

[0067] in, Indicated in The prediction region in the dimension is used to limit the spatial range of the prediction, which is a local region set in the feature space; The original data is along the dimensional feature vector; represents the Laplace operator; The mapping result is Dimensional component; Indicated in The integral differential element of the prediction region in dimension; Indicates prediction functions; types of prediction functions include: polynomial function, exponential function and / or radial basis function based on Gaussian distribution; Output of the prediction model.

[0068] Specifically, the integral region in the formula Indicated in The dimensional prediction region is used to set the spatial range of the prediction. In the feature space, Each dimension is divided into local regions to facilitate detailed spatial prediction. Each region corresponds to a feature dimension, which enables the system to capture the changing trend of features in a local range and avoid information ambiguity or insufficient generalization caused by global prediction. Through this regional restriction, the system can better accurately predict the features of different spatial regions and improve the spatial resolution of the prediction model in complex environments. In the formula, Indicates the evolution result in dimensional feature vector The gradient on reflects the rate of change of the feature flow in the local space. Through gradient analysis, the system can identify the trend and direction of the feature flow and use this as a basis to predict future changes. The trend information captured by this gradient part is particularly critical because it helps the system identify the potential direction of change of the data in different dimensions. For example, when the gradient value of a certain feature dimension is large, the system can predict that changes in this dimension may occur quickly, resulting in larger fluctuations in future states. This provides the system with an "early warning" capability for specific dimensions, which helps the system to respond in advance in areas of drastic changes. The Laplace operator contained in the formula The spatial characteristics of the system are further enhanced. The Laplace operator is a second-order differential operator used to measure the diffusion of characteristic flows in a local area. Dimensional Component By performing Laplace operations, the system can identify the diffusion or concentration characteristics of features. This means that the system can identify whether the data has a concentrated or diffuse trend in a specific area, thus providing an important basis for the spatial changes of future states. For example, when the data shows high diffusion in a certain area, the system can predict that the area may show a large distribution change in the future; on the contrary, if the Laplace value is small, it may mean that the distribution change in the area in the future tends to be stable.

[0069] The formula also includes the prediction function family , which for each component of the feature tensor Nonlinear processing is performed to further enhance the expressive power of the model. The types of prediction functions include polynomial functions, exponential functions, and radial basis functions based on Gaussian distribution. Polynomial functions can capture the nonlinear change trend of data. Through the combination of polynomials of different orders, the system can simulate the complex curve changes of data; exponential functions are suitable for simulating rapid increases and decreases in data, such as sudden growth or decay of data within a certain period of time; radial basis functions (RBF) are suitable for processing localized change patterns, such as sudden concentration or diffusion of features in a certain area. This diverse family of prediction functions ensures that the system can select the most appropriate prediction method based on the different characteristics and change patterns of the data, making the prediction results closer to the actual situation. In addition, integral differential elements This ensures that the system can accumulate feature changes in local areas in detail, so that feature changes in each area can be captured by the system. During the prediction process, the system performs cumulative calculations on each microelement to generate a global prediction output. , thus achieving a comprehensive prediction of future states in time and space. The integration process is particularly important here, which ensures that subtle changes in each feature dimension in multi-dimensional space can be recognized and integrated by the system. Finally, the prediction output It includes possible change patterns of features in different spatial and temporal dimensions, providing the system with a high-precision estimate of future states. This prediction result is not only an extension of the current evolution results, but also a preview of the future behavior of the system. In this way, the full-data AI control system can effectively predict data changes and make pre-adjustments before future environmental changes occur, thereby maintaining the stability and efficiency of the system in a dynamic environment.

[0070] Example 6: In step 2, the entropy increase rate of the system is calculated according to the output of the prediction model using the following formula:

[0071] ;

[0072] in, is the entropy increase rate; is the amplitude of the evolution result; The feature space The boundary normal vector of is the information flow density of the system; Output of the prediction model No. Dimensional component; The mapping result is Dimensional Component The square of the magnitude of the gradient.

[0073] Specifically, the first part of the formula uses the surface integral operation , which describes the changes in information flow on the system boundary. Here Represents the information flow density of the system, which reflects the rate and direction of information flow within the system. The greater the information flow density, the stronger the system's characteristic fluidity and the more significant the data changes. is the normal vector of the feature space boundary, which is used to indicate the flow direction of information flow on the boundary. and The system can identify the outflow or inflow of information flow on the boundary, thereby revealing the diffusion or concentration trend of data in space. The information flow density is normalized to This can balance the strength of information flow and avoid the impact of the difference in the strength of feature flow being too large or too small. This boundary information flow analysis allows the system to identify possible sudden changes at the boundary of the feature space. In particular, when strong information flow appears at the boundary, the system can determine whether the current state has entered an unstable area and provide an early warning signal. The second part of the entropy increase rate formula is the feature space Volume fraction within , which is used to analyze the predicted rate of change within the system. is the first-order derivative of the prediction model output in the time dimension, which indicates the rate of change of the prediction output in time. By calculating the square of the time change rate, the system can quantify the volatility of the prediction results and reveal the intensity of the data change in the feature space. This change intensity can reflect the rapid or slow change of the data in time. A large time change rate indicates that the data has a large uncertainty and may show a significant change trend in the future. This information is particularly critical for the dynamic regulation of the system, because in a state of large volatility, the system needs a higher adaptive ability to cope with rapid changes. The formula also contains This item indicates the first Dimensional Component The square of the gradient modulus. The square of the gradient modulus is used to measure the intensity of spatial variation of the feature flow in that dimension. The larger the gradient modulus, the more significant the spatial variation of the feature flow in that dimension, and the data may diffuse or concentrate rapidly in space. By calculating the square of the gradient modulus, the system can identify sensitive areas of data in space, that is, areas with drastic or relatively gentle changes. Combined with the square of the time rate of change, the system can identify which feature dimensions have both rapid time changes and significant spatial fluctuations, thereby locating the source of uncertainty within the system. This analysis of spatial variation is extremely important for judging the stability of the current state of the system, because it directly affects the distribution characteristics of the data in space. A larger spatial rate of change usually means that the data has higher complexity and unpredictability, and the system may need to make more frequent feedback corrections to cope with these fluctuations. Overall, the entropy increase rate calculated by this formula It provides a quantitative indicator of data complexity and uncertainty for the system. A higher entropy increase rate indicates that the current state of the system has higher complexity and uncertainty, which means that the data may change more drastically and the system should maintain high adaptability in the future state. Through the dynamic calculation of the entropy increase rate, the system can promptly identify unstable trends, such as sudden fluctuations or imbalances in certain feature dimensions, which is crucial for the feedback regulation of the system. If the system detects that the entropy increase rate exceeds a certain threshold, it can trigger an adaptive feedback mechanism to enhance the stability of the system by further adjusting model parameters or performing data optimization. The quantification of the entropy increase rate can also be used to guide the resource allocation of the system, such as allocating more computing resources in high-uncertainty areas to enhance the real-time monitoring and processing capabilities of these areas. Therefore, calculating the entropy increase rate is not only a means of measuring the complexity of system information, but also a basis for adaptive updating and adjustment of the system. It helps the full-data AI control system to perceive the changes in the current state in real time, and make predictions and adjustments based on this, so that the system can maintain stability and accuracy in a changing environment.

[0074] Embodiment 7: In step 3, according to the entropy increase rate, the feedback correction value of the system is calculated using the following formula:

[0075] ;

[0076] in, is the time kernel function; The evolution result Dimensional component; is the feedback correction value.

[0077] Specifically, the time kernel function in the formula Used to weight different points in time, which is a weighting mechanism based on historical data. Time kernel function According to the time difference Different weights are assigned to data at different time points. Through this weighting, the system can attach more importance to recent data and give lower weights to older data to reflect the greater impact of current data on the system. This weighting mechanism can ensure that the system remains flexible in a rapidly changing environment, because the system state reflected by the newer data is often closer to the current reality. The design of the time kernel function can choose different forms according to specific needs, such as exponential decay kernel or Gaussian kernel, to adapt to different time change patterns. In the formula, the entropy increase rate Gradient Indicates the direction and rate of change of system uncertainty. By calculating the gradient of the entropy increase rate, the system can determine the changing trend of the current state, so that it can be more directional when making adaptive adjustments. If the value is large, it means that the uncertainty of the system is increasing rapidly, and the system needs stronger feedback correction to adapt to this change. If the value is smaller, it means that the current state changes relatively smoothly and the system may not need significant adjustment. The introduction of the entropy increase rate gradient makes the feedback correction value of the system more sensitive to data fluctuations, thereby improving the system's ability to adapt to environmental changes.

[0078] Next, the formula The term represents the deviation between the predicted model output and the actual features, that is, the prediction error of the system. is the output of the prediction model. dimensional component, and The actual feature dimensional component. By calculating the prediction error in each dimension, the system can identify the difference between the model prediction and the actual data. This error term plays an important role in feedback correction. It provides a correction mechanism for the system, allowing the system to correct the error in each feature dimension. If the error is large, the feedback correction value will increase accordingly, thereby increasing the weight of this dimension when adjusting the model parameters to improve the prediction accuracy. In the formula, the first dimension of the evolution result Dimensional Component It is used to adjust the strength of feedback correction. This component comes from the dynamic evolution result of the feature flow and reflects the evolution trend of the data in a specific dimension. By combining the components of the evolution result, the system can better understand the importance of each feature dimension in dynamic changes. For example, if the evolution result component of a certain dimension is If the value is large, it means that the feature changes of this dimension are significant. Therefore, the system will give higher weight to this dimension when feedback correction is made, making the adjustment of this dimension more significant when the model is updated. In this way, the system can pay special attention to the feature dimensions that play an important role in dynamic changes while maintaining the overall accuracy. Finally, the integral symbol Indicates the time from the initial time to the current time The entire feedback correction process is accumulated. By integrating over time, the system can accumulate correction values ​​at different time points to generate a smooth and comprehensive feedback correction. This accumulation method ensures that the system can compensate for long-term error trends rather than just focusing on deviations at a single moment. Over time, the system can gradually stabilize through this accumulated feedback, thereby reducing the impact of fluctuations and making predictions more stable.

[0079] Embodiment 8: In step 3, according to the feedback correction value, the adaptive model of the system is updated by the following formula:

[0080] ;

[0081] in, Represents a system parameter set; is the learning rate.

[0082] Specifically, the formula Represents the parameter set of the system, which represents multiple important parameters in the model. These parameters determine the system's response to input data and the prediction accuracy. The goal of adaptive model updating is to adjust The value of enables the system to dynamically respond to changes in the data stream and thus optimize the model's predictive power. To achieve this goal, the formula calculates the time derivative of the parameter To update The learning rate is the value of , so that the system parameters can gradually adapt to changes in the environment. As a proportional factor in the formula, it controls the rate of parameter update. The size of the learning rate directly affects the system's response speed to feedback information. A larger learning rate enables the system to adjust parameters quickly, which is suitable for responding to rapidly changing environments; while a smaller learning rate makes the system update smoother and suitable for stable environments. The choice of learning rate needs to be weighed according to the specific application, because too high a learning rate may cause parameter oscillation and difficulty in convergence, while too low a learning rate may make the system react too slowly and unable to quickly adapt to changes in the environment. The formula contains the gradient of the feedback correction value , which reflects the feedback correction value By introducing , the system can identify the direction of change of the current correction value in each parameter dimension, so that it is more directional when adaptively updating. If the value is large, it means that the current error correction of the system is large and the model needs stronger adjustment; if The smaller the value, the more stable the system error is, and the adjustment can be relatively weak. The gradient of the feedback correction value makes the system parameter update more flexible and intelligent. By responding to the direction of the correction value, the system can adjust to the optimal parameter state more quickly. The term is the sensitivity of the predicted output to the mapping result, which plays a role in strengthening the adjustment in the adaptive update. Mapping results The derivative of can identify the sensitivity of each feature dimension to the prediction result. Specifically, if the derivative of a certain dimension is large, it means that the dimension has a significant impact on the prediction result. Then the system will give a greater weight when adjusting the parameters of this dimension to ensure that the system can capture the changes in these significant features. This weighted update of sensitivity ensures that the system can specifically strengthen the responsiveness of key feature dimensions when adjusting parameters, making the model update more targeted and adaptive. Finally, in the formula The entropy increase rate of the system The square root of reflects the information complexity of the current system state. , the system can adjust the intensity of parameter updates according to the uncertainty of the current state. A higher entropy increase rate indicates that the system state is more complex and the data changes more. At this time, the intensity of parameter updates should be increased accordingly to enable the system to better adapt to fluctuations in the data stream. A lower entropy increase rate indicates that the system state is relatively stable. At this time, parameter updates can be appropriately weakened to avoid oscillations caused by over-adjustment. The introduction of the entropy increase rate enables the adaptive update process of the system to take into account both complexity and stability, thereby providing stronger adaptability when uncertainty is high and maintaining stable parameter changes when data is relatively stable.

[0083] Example 9: In step 3, the overall stability of the system is calculated by the following formula:

[0084] ;

[0085] in, Represents the prediction area in all dimensions.

[0086] Specifically, in the first part of the formula, the surface integral Describes the feedback correction gradient and system parameter sets The integrated effect on stability. Represents the prediction area in all dimensions, that is, cumulative analysis of feedback and parameter changes in multidimensional feature space. By integrating the prediction area, the system can capture the changing trend of the feedback correction value in each dimension and evaluate the overall stability based on these changes. The gradient of the feedback correction value It reflects the current system's response to the error. If the value is larger, it means that the feedback adjustment is stronger and the system may be in a high response state; if Smaller, indicating that the feedback adjustment of the system is weaker and tends to be more stable. The introduction of further reflects the response strength of the model to the input data, because the system parameters directly affect the sensitivity of the model to data changes. Larger parameter values ​​usually enhance the response of the system, but may also lead to decreased stability. Therefore, the design of this part achieves a balance between system responsiveness and stability through the trade-off between feedback and parameter sets. The terms included in the formula Used to convert the entropy increase rate Introduced into the overall stability analysis, the entropy increase rate reflects the complexity and uncertainty of the current state of the system. It means that the uncertainty of data state is strong and the information complexity is high. In this case, the stability of the system is relatively low. The system will reduce the weight of feedback and parameters in complex state, making the feedback adjustment of the system more stable, while allowing higher feedback adjustment weight in stable state. This design effectively balances the adaptability and stability of the system in different states, making the system more conservative in high-complexity state and more responsive in low-complexity state.

[0087] The second part of the formula It is used to introduce the influence of prediction error. Here is the output of the prediction model, is the actual eigenvalue, and is the component of the evolution of the characteristic flow. represents the prediction deviation of the system, and the characteristic flow evolution result represents the dynamic influence of the feature dimension in the evolution process. The exponential decay value of the prediction error allows the system to dynamically adjust the sensitivity of the overall stability to the prediction error. A larger prediction error will cause the exponential decay value to approach zero, indicating that the system is unstable in this state and needs further adjustment; a smaller prediction error will make this term close to 1, indicating that the system is more stable in this dimension. The introduction of exponential decay allows the system to more sensitively reflect the impact of the prediction error when calculating stability, so that appropriate feedback adjustments can be made under high error conditions. Through the combination of these two parts, the overall stability of the system It can comprehensively measure the response and prediction accuracy of the system in the current state. A high overall stability means that the current state of the system has achieved a balance between feedback, parameters and prediction errors, indicating that the system is in a relatively ideal stable state; while a low stability indicates that the system may have a large deviation or require stronger adaptive feedback to restore balance. In a highly complex or rapidly changing environment, the system can adjust the feedback strength in time according to the stability calculation results to restore the stable state faster, while in a stable environment, it maintains a gentler feedback strength to prevent oscillations caused by over-adjustment.

[0088] As described above, 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 aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these 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.

Claims

1. The adaptive model updating method of the full data AI control system is characterized by: It includes the following steps: Step 1: Obtain the original data of the system, perform tensor decomposition on the original data, extract spatiotemporal features, and obtain feature tensors; Through the interaction between the vorticity field and the nonlinear function, the characteristic tensor is projected into a nonlinear dynamic space to obtain the mapping result; Step 2: dynamically evolve the feature flow of the mapping result to obtain the evolution result; input the evolution result into the preset prediction model and output the prediction model output; calculate the entropy increase rate of the system according to the prediction model output; Step 3: Calculate the feedback correction value of the system according to the entropy increase rate; update the adaptive model of the system according to the feedback correction value, and calculate the overall stability of the system. When the overall stability of the system is lower than the set threshold, issue an early warning; In step 1, use the following formula to decompose the original data into tensors: ； in, is the original data; and Respectively represent the original data in A pair of eigenvalues ​​corresponding to each other in dimension; Represents the decomposition rank The summation operation, Represents the rank of tensor decomposition, that is, the number of layers of tensor decomposition; Represents the characteristic dimension of the original data; and All are integer subscript indices; Expressing the Dimensional characteristic curve The closed path integral of ; Indicates The characteristic curve on the dimension is used to capture the changes in the data on that dimension; The original data dimensional feature vector The second time derivative of ; The original data is along the dimensional feature vector; is the time scale factor, For the The time decay factor of the dimension, is the time difference; Represents the tensor product operation of vectors; is the curve integral differential, which means A tiny unit for integration on a dimensional characteristic curve.

2. The adaptive model updating method of the full data AI control system according to claim 1, characterized in that: In step 1, the following formula is used to project the characteristic tensor into a nonlinear dynamic space through the interaction between the vorticity field and the nonlinear function to obtain the mapping result: ； in is the mapping result; Represents the feature space The integral of Represents the feature tensor The feature space in which it is located; represents the curl operator; is the cross product operator; Used to calculate the vorticity field to capture the characteristic tensor In feature space The rotation behavior in generates a vorticity field; is the integral differential element of the feature space; represents the vorticity field, which is a vector field used to capture the rotational dynamic characteristics of the system; The calculation of combines the characteristic tensor with the vorticity field to generate a quantity that reflects the rotational behavior of the characteristic tensor; is the dimension of the feature tensor; express In the The weight of the dimension; Indicates dimensional spatial coordinates; is the number of nonlinear functions; and All are integer subscript indices; Indicates A non-linear function.

3. The adaptive model updating method of the full data AI control system according to claim 2, characterized in that: In step 2, the mapping result is dynamically evolved by the feature flow using the following formula to obtain the evolution result: ； in, As a result of evolution; is the current time; is the characteristic flow surface; is the integral differential element of the characteristic flow surface; is the feature tensor in Dimensional scaling factor; is the evolution time; is the evolution time integral differential.

4. The adaptive model updating method of the full data AI control system according to claim 3, characterized in that: In step 2, the evolution results are input into the preset prediction model through the following formula, and the prediction model output is output: ； in, Indicated in The prediction region in the dimension is used to limit the spatial range of the prediction, which is a local region set in the feature space; The original data is along the dimensional feature vector; represents the Laplace operator; The mapping result is Dimensional component; Indicated in The integral differential element of the prediction region in dimension; Indicates prediction functions; types of prediction functions include: polynomial function, exponential function and / or radial basis function based on Gaussian distribution; Output of the prediction model.

5. The adaptive model updating method of the full data AI control system according to claim 4, characterized in that: In step 2, the entropy increase rate of the system is calculated based on the output of the prediction model using the following formula: ； in, is the entropy increase rate; is the amplitude of the evolution result; The feature space The boundary normal vector of is the information flow density of the system; Output of the prediction model No. Dimensional component; The mapping result is Dimensional Component The square of the magnitude of the gradient.

6. The adaptive model updating method of the full data AI control system according to claim 5, characterized in that: In step 3, the feedback correction value of the system is calculated according to the entropy increase rate using the following formula: ； in, is the time kernel function; The evolution result Dimensional component; is the feedback correction value.

7. The adaptive model updating method of the full data AI control system according to claim 6, characterized in that: In step 3, the adaptive model of the system is updated according to the feedback correction value using the following formula: ； in, Represents a system parameter set; is the learning rate.

8. The adaptive model updating method of the full data AI control system according to claim 7, characterized in that: In step 3, the overall stability of the system is calculated using the following formula: ； in, Represents the prediction area in all dimensions.

Images