Distributed resource hyper-parameter estimation method
By adopting GPT-based random model and AlphaCode optimization code generation method in distributed resource hyperparameter estimation, the problems of limitations of estimation model and insufficient noise processing capabilities in the prior art are solved, and high-precision, real-time update and multi-objective optimization of distributed resource hyperparameters are achieved, which improves the adaptability and robustness of the system.
Patent Information
- Application Number
- CN202411973139.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-25
AI Technical Summary
The existing distributed resource hyperparameter estimation methods have problems such as limitations of the estimation model, insufficient noise processing capability, limitations of optimization process and insufficient real-time scalability, and it is difficult to accurately capture dynamic changes and complex noise characteristics, especially in high-dimensional and multi-constrained distributed resource scenarios.
The distributed resource random model based on GPT is adopted to capture the dynamic changes of hyperparameters and the complex relationship between system state by establishing a Markov chain state transition model and a cluster response random system model; the sequence modeling ability of GPT is used to design a filtering method for complex noise characteristics; combined with AlphaCode, it automatically generates optimization code to realize real-time update of distributed resource hyperparameters and multi-objective optimization.
It improves the accuracy and real-timeness of hyperparameter estimation of distributed resources, adapts to dynamic environments and complex scenarios, enhances the adaptability and robustness of the system in uncertain environments, and supports efficient operation of distributed resource management, scheduling and control.
Smart Images
Figure CN119886369A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of new energy operation, and in particular relates to a distributed resource hyperparameter estimation method. Background Art
[0002] Estimating the hyperparameters of distributed resources is an important part of optimizing the management and operation of distributed resources. Hyperparameters are key control variables in the distributed resource operation model. Through hyperparameter estimation, the model can be calibrated to more accurately reflect the actual operation characteristics of the resources. The operating environment of distributed resources has dynamic changing characteristics (such as load fluctuations and weather changes). By estimating hyperparameters in real time, the model can maintain high accuracy in different scenarios. Accurately estimating hyperparameters can improve the reliability and efficiency of distributed resource scheduling algorithms and ensure that resources are optimally allocated to different operation tasks. The hyperparameters of distributed resources directly affect their dynamic response characteristics. Through accurate estimation, control strategies can be optimized, fluctuations can be reduced, and system stability can be ensured. Distributed resources are affected by a variety of uncertain factors (such as weather, load fluctuations, and equipment failures). Hyperparameter estimation can help the system better handle these uncertainties and improve operational robustness. Accurately estimating hyperparameters (such as component aging rate and battery capacity decay rate) can predict equipment reliability, formulate preventive maintenance plans, and extend equipment life. However, the shortcomings of existing distributed resource hyperparameter estimation methods are as follows: 1) Limitations of estimation models. Traditional methods usually assume that the hyperparameters of distributed resources are static or slowly changing. In actual operation, hyperparameters may change dynamically with the environment (such as weather changes, load fluctuations); many estimation models are based on linear or simplified assumptions, which makes it difficult to capture the nonlinear characteristics and complex relationships of distributed resources. 2) Noise influence. The measurement noise in actual systems may have time correlation or non-Gaussian characteristics, while traditional filtering methods (such as Kalman filtering) assume that the noise is independent and identically distributed or Gaussian, which makes it difficult to accurately handle complex noise. 3) Limitations of the optimization process. Hyperparameter estimation of distributed resources often involves multi-objective optimization (such as the trade-off between fitting accuracy and operating cost). Existing methods find it difficult to optimize multiple objectives at the same time. Existing optimization methods are prone to falling into local optimality, especially in high-dimensional, multi-constrained distributed resource scenarios. 4) Real-time and scalability. Some methods have low computational efficiency when processing dynamic data and are difficult to meet real-time update requirements. Existing methods are difficult to adapt to complex and heterogeneous distributed resource scenarios (such as the integration of multiple resources such as wind power, photovoltaics, and energy storage). Therefore, it is necessary to develop a new distributed resource hyperparameter estimation method to solve existing problems. Summary of the invention
[0003] The purpose of the present invention is to provide a distributed resource hyperparameter estimation method to solve the problem that hyperparameters cannot be accurately estimated.
[0004] To achieve the above object, the present invention provides the following technical solution: a distributed resource hyperparameter estimation method, comprising:
[0005] Construct a random model of distributed resources based on GPT: Assume that random variables represent uncertainty factors, and establish a Markov chain state transition model based on GPT based on the linear time-invariant model of distributed resources; make a priori assumptions about the distribution state of distributed resource parameters, and establish a cluster response random system model of distributed resources for calculating joint distribution and mathematically expressing the solution;
[0006] Filter out noise in the distributed resource parameter measurement process;
[0007] Hyperparameter posterior distribution, and update the posterior distribution of GPT model parameters;
[0008] The estimated value of distributed resources is obtained by approximating the posterior distribution of hyperparameters.
[0009] Preferably, the specific steps of constructing a random model of distributed resources based on GPT include:
[0010] Building a distributed resource linear time-invariant model includes:
[0011] The operating characteristics of distributed resources are shown in formula (1);
[0012] x(t)=A·λ(t)+B·μ(t)+ω(t) Formula (1)
[0013] Where: x(t) is the operating characteristic of the distributed resource, λ(t) is the state variable of the distributed resource, μ(t) is the input variable; ω(t) is the uncertainty disturbance, assuming that ω(t)~N(0,Q) obeys the mean Gaussian distribution; A and B are the state transfer matrix and control matrix respectively;
[0014] Preferably, the establishment of the Markov chain state transition model based on GPT includes: converting the distributed resource linear time-invariant model into a discrete form, and modeling the randomness of state transition; the Markov chain state transition model based on GPT is shown in formula (2);
[0015] x k+1 =f GPT (x k ,μ k ,θ)+ω k Formula (2)
[0016] Where: x k+1 is the distributed resource operation characteristic of the k+1th discrete point, f GPT () is the Markov chain state transition model based on GPT, θ is the parameter of the GPT model, ωk is the uncertainty disturbance of the input variable after the kth discretization, μ k is the kth discretized input variable, x k Indicates the current state;
[0017] The state x of the k+1th discrete point of the distributed resource k+1 The probability distribution depends on the current state x k and control input μ k , the Markov chain assumption is as shown in formula (3);
[0018] p(x k+1 |x k ,μ k )=N(x k+1 |f GPT (x k ,μ k ,θ),Q) Formula (3)
[0019] Where: Q is the noise term ω in the Markov chain state transition model based on GPT k The covariance matrix of .
[0020] Preferably, the making a priori assumptions about the distribution state of the distributed resource parameters includes:
[0021] Make a priori assumptions about the parameters θ of the GPT filtering recursive prediction model. The priori assumptions about the parameter distribution are shown in formula (4);
[0022] p GPT (θ)=P GPT (θ) Formula (4)
[0023] Where: p GPT (θ) is the prior distribution of the GPT model parameters, and P'(θ) is the prior distribution of the parameters.
[0024] Preferably, the establishment of the random system model of cluster response of distributed resources includes: constructing the random system model of cluster response of distributed resources based on the Markov chain state transition model and parameter distribution of GPT, and formula (5) is the random system model of cluster response of distributed resources;
[0025] p(x k+1 |x k ,μ k ,θ)=N(x k+1 |f GPT (x k ,μ k ,θ),Q) Formula (5)
[0026] Where: p(x k) is the state probability distribution of discrete point k, ω k ~N(0,Q) is the random disturbance at discrete moments.
[0027] Preferably, the calculation of the joint distribution of the random system model of cluster response of distributed resources includes: the joint distribution of the random system model of cluster response of distributed resources is as shown in formula (6):
[0028] As shown;
[0029] Where T represents the period; the state x of the k+1th discrete point of the distributed resource k+1 The marginal probability distribution of is shown in formula (7);
[0030] p(x k+1 )=∫∫p(x k+1 |x k ,μ k ,θ)·p(x k )·p GPT (θ)dx k dθ formula (7)
[0031] Here, d represents calculus.
[0032] Preferably, the filtering of noise in the distributed resource parameter measurement process includes:
[0033] Build a measurement model that takes noise into account and update the posterior distribution of the state whenever a new distributed resource measurement value is obtained;
[0034] Construct a GPT-based filtered recursive prediction model.
[0035] Preferably, the measurement model considering noise is constructed, and whenever a new distributed resource measurement value is obtained, the posterior distribution of the updated state includes: the measurement model considering noise is formula (8):
[0036] z k =H·x k +v k Formula (8)
[0037] Among them, z k is the measured value of the kth distributed resource parameter discrete point, H is the observation matrix of the distributed resource parameter, x k is the state value of the kth distributed resource parameter discrete point, v k is the measurement noise of distributed resource parameters;
[0038] Preferably, the constructing of the GPT-based filtering recursive prediction model includes:
[0039] The filtering recursive prediction model based on GPT is shown in formula (9):
[0040]
[0041] in: It is a filtering recursive prediction model based on GPT;
[0042] The filtering recursive update model is shown in formula (10):
[0043] p(x k |z 1:k )←p(z k |x k )p(x k |z 1:k-1 ) Formula (10)
[0044] Preferably, the hyperparameter posterior distribution and updating the posterior distribution of the GPT model parameters include:
[0045] The posterior distribution of the random model parameters of the distributed resources based on GPT is updated through Bayesian theory. The posterior distribution of the hyperparameters is shown in formula (11):
[0046] p(θ|z 1:T )←p(z 1:T |θ)p GPT (θ) Formula (11)
[0047] The method of obtaining an estimated value of a distributed resource by approximating the posterior distribution of a hyperparameter includes: constructing an AlphaCode-based distributed resource hyperparameter estimation model, using AlphaCode to approximate the posterior distribution of a hyperparameter θ to obtain an estimated value, and the AlphaCode-based distributed resource hyperparameter estimation model is shown in formula (12);
[0048]
[0049] Where: is the updated distributed resource hyperparameter, f AlphaCode () is a distributed resource hyperparameter estimation model based on AlphaCode.
[0050] The technical effects and advantages of the present invention are as follows: the distributed resource hyperparameter estimation method utilizes the powerful feature extraction and sequence prediction capabilities of the GPT model to establish a random model of distributed resources and capture the complex relationship between the dynamic changes of hyperparameters and the system state; utilizes the sequence modeling capability of the GPT model to design a filtering method for complex noise characteristics and improve the accuracy of distributed resource state estimation; utilizes AlphaCode to automatically generate optimization code to achieve real-time updating of distributed resource hyperparameters and adapt to dynamic environments and multi-objective optimization requirements; utilizes the powerful nonlinear modeling capability of GPT to fully capture dynamic characteristics and noise characteristics; improves the adaptability of the system in uncertain environments through advanced filtering methods and hyperparameter optimization; improves efficiency and adapts to complex scenarios and requirements through dynamic generation and optimization of AlphaCode; provides a comprehensive and intelligent solution for the management, scheduling and control of distributed resources and supports the efficient operation and low-carbon development of future energy systems; improves the estimation accuracy of distributed resource hyperparameters, improves the adaptability of the system in uncertain environments, and adapts to the distributed resource hyperparameter estimation requirements in different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 are within the scope of protection of the present invention.
[0053] The present invention provides Figure 1 A distributed resource hyperparameter estimation method based on GPT and AlphaCode is shown in , wherein the distributed resource hyperparameter refers to the parameter coefficient set in the distributed resource modeling process, including: a random model of distributed resources based on GPT, a distributed resource measurement noise filtering method based on GPT, and a distributed resource hyperparameter update method based on AlphaCode; specifically including the following steps:
[0054] Step (1) proposes a random model of distributed resources based on GPT. Distributed resources are affected by various factors during operation and exhibit uncertain characteristics. Therefore, random variables are introduced in the modeling process to represent uncertainty factors. On the basis of the linear time-invariant model of distributed resources, a Markov chain state transition model based on GPT is established; then, a priori assumptions are made on the distribution state of distributed resource parameters, a cluster response random system model of distributed resources is established, the joint distribution is calculated, and the solution of the cluster response random system model of distributed resources is mathematically expressed;
[0055] The random model of distributed resources based on GPT consists of the linear time-invariant equation of distributed resources, the Markov chain state transition model, and the cluster response random system model of distributed resources;
[0056] Step (2): propose a distributed resource measurement noise filtering method based on GPT to filter the noise that occurs during the distributed resource parameter measurement process, thereby improving the accuracy of distributed resource measurement and the precision of hyperparameter estimation; construct a measurement model that takes noise into account, and update the posterior distribution of the state whenever a new distributed resource measurement value is obtained; construct a filtering recursive prediction model based on GPT, combine the system model and measurement data, reduce the impact of noise through a recursive method, optimize the state estimation, and support subsequent prediction, control or hyperparameter optimization.
[0057] Step (3): a distributed resource hyperparameter updating method based on AlphaCode is proposed. Based on the characteristics of AlphaCode such as high automation, high intelligence, high efficiency and strong robustness, the hyperparameters of distributed resources can be adjusted in real time in a dynamic environment, which can improve the accuracy and adaptability of distributed resource modeling, while significantly reducing development and deployment costs, and is suitable for complex, multi-objective distributed energy optimization scenarios. First, the posterior distribution of hyperparameters is performed, and the posterior distribution of GPT model parameters is updated through Bayesian theory. Then, a distributed resource hyperparameter estimation model based on AlphaCode is proposed, and AlphaCode is used to approximate the posterior distribution of the hyperparameter θ, and the estimated value of the distributed resource is obtained.
[0058] For step (1), the proposed random model of distributed resources based on GPT is shown in formulas (1)-(7). Random variables are introduced to represent uncertainty factors. On the basis of the linear time-invariant model of distributed resources, a Markov chain state transition model based on GPT is established; then, a priori assumptions are made on the distribution state of distributed resource parameters, a cluster response random system model of distributed resources is established, the joint distribution of the random system is calculated, and the solution of the random system model is mathematically expressed.
[0059] A linear time-invariant model of distributed resources is constructed, and the operating characteristics of distributed resources are shown in formula (1).
[0060] Formula (2) is a Markov chain state transition model based on GPT, which mainly transforms the linear time-invariant model into a discrete form and models the randomness of state transition.
[0061] Formula (3) is the Markov chain hypothesis, the state x k+1 The probability distribution depends only on the current state x k and control input μ k .
[0062] Formula (4) is a priori assumption of parameter distribution, which mainly makes a priori assumption on the parameter θ of the GPT filtering recursive prediction model.
[0063] Formula (5) is a random system model of cluster response. Based on the Markov chain state transition model and parameter distribution of GPT, a random system model of distributed resource cluster response is constructed.
[0064] Formula (6) is the joint distribution of the random system.
[0065] Formula (7) is the state x k+1 The marginal probability distribution of .
[0066] x(t)=A·λ(t)+B·μ(t)+ω(t) Formula (1)
[0067] x k+1 =f GPT (x k ,μ k ,θ)+ω k Formula (2)
[0068] p(x k+1 |x k ,μ k )=N(x k+1 |f GPT (x k ,μ k ,θ),Q) Formula (3)
[0069] p GPT (θ)=P G ' PT (θ) Formula (4)
[0070] p(x k+1 |x k ,μ k ,θ)=N(x k+1 |f GPT (x k ,μ k ,θ),Q) Formula (5)
[0071]
[0072] p(x k+1 )=∫∫p(x k+1 |x k ,μ k ,θ)·p(x k )·p GPT (θ)dx k dθ formula (7)
[0073] Where: x(t) is the operating characteristic of the distributed resource, λ(t) is the state variable of the distributed resource (such as power, load, etc.), μ(t) is the input variable (such as control input); ω(t) is the uncertainty disturbance, assuming that ω(t)~N(0,Q) obeys the mean Gaussian distribution. A and B are the state transfer matrix and control matrix respectively. Q is the noise term ω in the Markov chain state transition model based on GPT k The covariance matrix of p(x k ) is the state probability distribution of discrete point k. GPT (θ) is the prior distribution of GPT model parameters. k+1 is the distributed resource operation characteristic of the k+1th discrete point, f GPT () is the Markov chain state transition model based on GPT, μ k is the kth discretized input variable, ω k is the uncertainty disturbance of the input variable after the kth discretization, θ is the parameter of the GPT model, ω k ~N(0,Q) is a random perturbation at discrete moments, where T represents the period and d represents the calculus.
[0074] For step (2), the proposed distributed resource measurement noise filtering method based on GPT is shown in formulas (8)-(10). The present invention first constructs a measurement model that takes noise into account, and updates the posterior distribution of the state whenever a new distributed resource measurement value is obtained. A filtering recursive prediction model based on GPT is constructed, combining the system model and measurement data, reducing the impact of noise in a recursive manner, optimizing state estimation, and supporting subsequent prediction, control or hyperparameter optimization.
[0075] Formula (8) is a measurement model that takes noise into account. Whenever a new measurement value z is obtained, k When , update the state x k The posterior distribution of .
[0076] Formula (9) is a filtering recursive prediction model based on GPT.
[0077] Formula (10) is the filtering recursive update model.
[0078] z k =H·x k +v k Formula (8)
[0079]
[0080] p(x k |z 1:k )←p(z k |x k )p(x k |z 1:k-1 )Formula (10)
[0081] Where: z k is the measured value of the kth distributed resource parameter discrete point, H is the observation matrix of the distributed resource parameter, x k is the state value of the kth distributed resource parameter discrete point, v k is the measurement noise of distributed resource parameters. It is a filtering recursive prediction model based on GPT.
[0082] For step (3), the proposed distributed resource hyperparameter updating method based on AlphaCode is shown in formulas (11)-(12). Based on the characteristics of high automation, high intelligence, high efficiency and strong robustness of AlphaCode, the hyperparameters of distributed resources can be adjusted in real time in a dynamic environment, which can improve the accuracy and adaptability of distributed resource modeling, while significantly reducing development and deployment costs, and is suitable for complex, multi-objective distributed energy optimization scenarios. The present invention first performs a posterior distribution of hyperparameters, and updates the posterior distribution of the random model parameters of distributed resources based on GPT through Bayesian theory; then, a distributed resource hyperparameter estimation model based on AlphaCode is proposed, and AlphaCode is used to approximate the posterior distribution of the hyperparameter θ, and the estimated value of the distributed resource is obtained.
[0083] Formula (11) is the posterior distribution of the hyperparameters. The present invention updates the posterior distribution of the GPT model parameters through Bayesian theory.
[0084] Formula (12) is a distributed resource hyperparameter estimation model based on AlphaCode. AlphaCode is used to approximate the posterior distribution of the hyperparameter θ and obtain the estimated value.
[0085] p(θ|z 1:T )←p(z 1:T |θ)p GPT (θ) Formula (11)
[0086]
[0087] Where: is the updated distributed resource hyperparameter, f AlphaCode () is a distributed resource hyperparameter estimation model based on AlphaCode.
[0088] Through the above steps, the estimation of distributed resource hyperparameters can be achieved, the accuracy and effectiveness of distributed resource hyperparameter estimation can be improved, and the accuracy of prediction of distributed resource power generation, load demand, etc. can be improved.
[0089] This distributed resource hyperparameter estimation method uses the powerful feature extraction and sequence prediction capabilities of the GPT model to establish a random model of distributed resources and capture the complex relationship between the dynamic changes of hyperparameters and the system state; uses the sequence modeling capability of the GPT model to design a filtering method for complex noise characteristics to improve the accuracy of distributed resource state estimation; uses AlphaCode to automatically generate optimization code to achieve real-time updating of distributed resource hyperparameters to adapt to dynamic environments and multi-objective optimization needs; uses the powerful nonlinear modeling capabilities of GPT to fully capture dynamic characteristics and noise characteristics; uses advanced filtering methods and hyperparameter optimization to improve the adaptability of the system in uncertain environments; uses AlphaCode dynamic generation and optimization to improve efficiency and adapt to complex scenarios and needs; provides a comprehensive and intelligent solution for the management, scheduling and control of distributed resources, supports the efficient operation and low-carbon development of future energy systems, improves the estimation accuracy of distributed resource hyperparameters, improves the adaptability of the system in uncertain environments, and adapts to the distributed resource hyperparameter estimation needs in different scenarios.
[0090] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A distributed resource hyperparameter estimation method, characterized in that: include: Construct a random model of distributed resources based on GPT: Assume that random variables represent uncertainty factors, and establish a Markov chain state transition model based on GPT based on the linear time-invariant model of distributed resources; make a priori assumptions about the distribution state of distributed resource parameters, and establish a cluster response random system model of distributed resources for calculating joint distribution and mathematically expressing the solution; Filter out noise in the distributed resource parameter measurement process; Hyperparameter posterior distribution, and update the posterior distribution of GPT model parameters; The estimated value of distributed resources is obtained by approximating the posterior distribution of hyperparameters.
2. A distributed resource hyperparameter estimation method according to claim 1, characterized in that: The specific steps of constructing a random model of distributed resources based on GPT include: Building a distributed resource linear time-invariant model includes: The operating characteristics of distributed resources are shown in formula (1); x(t)=A·λ(t)+B·μ(t)+ω(t) Formula (1) Among them: x(t) is the operating characteristic of the distributed resource, λ(t) is the state variable of the distributed resource, μ(t) is the input variable; ω(t) is the uncertainty disturbance, assuming that ω(t)~N(0,Q) obeys the mean Gaussian distribution; A and B are the state transfer matrix and control matrix respectively.
3. A distributed resource hyperparameter estimation method according to claim 2, characterized in that: The establishment of the Markov chain state transition model based on GPT includes: converting the distributed resource linear time-invariant model into a discrete form, and modeling the randomness of state transition; the Markov chain state transition model based on GPT is shown in formula (2); x k+1 =f GPT (x k ,m k ,θ)+ω k formula(2) Where: x k+1 is the distributed resource operation characteristic of the k+1th discrete point, f GPT () is the Markov chain state transition model based on GPT, θ is the parameter of the GPT model, ω k is the uncertainty disturbance of the input variable after the kth discretization, μ k is the kth discretized input variable, x k Indicates the current state; The state x of the k+1th discrete point of the distributed resource k+1 The probability distribution depends on the current state x k and control input μ k , the Markov chain assumption is as shown in formula (3); p(x k+1 |x k ,μ k ) = N(x k+1 |f GPT (x k ,μ k ,θ),Q) Equation (3) Where: Q is the noise term ω in the Markov chain state transition model based on GPT k The covariance matrix of .
4. A distributed resource hyperparameter estimation method according to claim 3, characterized in that: The prior assumption of the distribution state of the distributed resource parameters includes: Make a priori assumptions about the parameters θ of the GPT filtering recursive prediction model. The priori assumptions about the parameter distribution are shown in formula (4); p GPT P'(θ) = GPT P'(θ), formula (4) Where: p GPT (θ) is the prior distribution of the GPT model parameters, and P'(θ) is the prior distribution of the parameters.
5. A distributed resource hyperparameter estimation method according to claim 4, characterized in that: The establishment of the random system model of cluster response of distributed resources includes: constructing the random system model of cluster response of distributed resources based on the Markov chain state transition model and parameter distribution of GPT, and formula (5) is the random system model of cluster response of distributed resources; p(x k+1 |x k ,μ k ,θ) = N(x k+1 |f GPT (x k ,μ k ,θ),Q) Formula (5) Where: p(x k ) is the state probability distribution of discrete point k, ω k ~N(0,Q) is the random disturbance at discrete moments.
6. A distributed resource hyperparameter estimation method according to claim 5, characterized in that: The joint distribution of the cluster response stochastic system model for calculating distributed resources includes: the joint distribution is shown in formula (6); Where, T represents the period; Status x k+1 The marginal probability distribution of is shown in formula (7); p(x k+1 ) = ∫∫p(x k+1 |x k , μ k , θ)·p(x k )·p GPT (θ)dx k dθ Formula (7) Here, d represents calculus.
7. A distributed resource hyperparameter estimation method according to claim 1, characterized in that: The noise filtering during the distributed resource parameter measurement process includes: Build a measurement model that takes noise into account and update the posterior distribution of the state whenever a new distributed resource measurement value is obtained; Construct a GPT-based filtered recursive prediction model.
8. A distributed resource hyperparameter estimation method according to claim 7, characterized in that: The measurement model considering noise is constructed. Whenever a new distributed resource measurement value is obtained, the posterior distribution of the updated state includes: the measurement model considering noise is formula (8): z k =H·x k +v k Formula (8) Among them, z k is the measured value of the kth distributed resource parameter discrete point, H is the observation matrix of the distributed resource parameter, x k is the state value of the kth distributed resource parameter discrete point, v k is the measurement noise of distributed resource parameters.
9. A distributed resource hyperparameter estimation method according to claim 8, characterized in that: The construction of the GPT-based filtering recursive prediction model includes: The filtering recursive prediction model based on GPT is shown in formula (9): in: It is a filtering recursive prediction model based on GPT; The filtering recursive update model is shown in formula (10): p(x k |z 1:k ) ← p(z k |x k ) p(x k |z 1:k-1 ) Formula (10) 10. A distributed resource hyperparameter estimation method according to any one of claims 1 to 9, characterized in that: The hyperparameter posterior distribution and the posterior distribution of the updated GPT model parameters include: The posterior distribution of the random model parameters of the distributed resources based on GPT is updated through Bayesian theory. The posterior distribution of the hyperparameters is shown in formula (11): p(θ|z 1:T )←p(z 1:T |θ)p GPT (θ) Equation (11) The method of obtaining an estimated value of a distributed resource by approximating the posterior distribution of a hyperparameter includes: constructing an AlphaCode-based distributed resource hyperparameter estimation model, using AlphaCode to approximate the posterior distribution of a hyperparameter θ to obtain an estimated value, and the AlphaCode-based distributed resource hyperparameter estimation model is shown in formula (12); Where: is the updated distributed resource hyperparameter, f AlphaCode () is a distributed resource hyperparameter estimation model based on AlphaCode.