Power plant full-parameter time sequence simulation prediction method, system, equipment and medium

Abstract

The invention provides a power plant full-parameter time sequence simulation prediction method, system and device and a medium, and belongs to the technical field of power system simulation. The method comprises the following steps: acquiring historical operation data from a power plant DCS system; screening measuring point data with dynamic characteristics based on variance analysis; performing normalization processing on the screened data and dividing the data into a training set, a verification set and a test set; a prediction model based on trend-residual decomposition and heterogeneous signal collaborative modeling is constructed, a sequence is decoupled into a trend component and a residual component through a trend-residual decomposition module, a dual-path architecture is adopted for processing the trend component and the residual component respectively, and collaborative prediction of analog quantity and switching quantity is achieved through a heterogeneous signal collaborative modeling module; training a model by using the training set and storing a plurality of model copies; evaluating model performance based on the verification set and selecting an optimal model; and finally, performing real-time prediction by using the optimal model. According to the method, the precision, robustness and physical consistency of power plant all-parameter prediction are effectively improved.

Description

A method, system, equipment, and medium for full-parameter time-series simulation and prediction of power plants. Technical Field

[0001] This invention belongs to the field of power system simulation technology, and more specifically relates to a method, system, equipment and medium for time-series simulation and prediction of full parameters of a power plant. Background Technology

[0002] Thermal power plants are highly complex, multivariate, and strongly coupled dynamic systems. Their operating state is characterized by hundreds or thousands of measurement points covering continuous analog quantities such as temperature, pressure, and flow rate, as well as discrete switching quantities such as valve and equipment start-up and shutdown. To achieve advanced applications such as intelligent monitoring, fault prediction, and high-fidelity digital twins, it is urgent to construct models capable of synchronously and accurately simulating all plant parameters. Currently, this field mainly relies on two technical approaches: mechanistic modeling and traditional data-driven methods. However, both have significant limitations when dealing with full-parameter simulation tasks.

[0003] Physical mechanism-based modeling methods describe system dynamics by establishing systems of differential equations from thermodynamics and fluid mechanics. While this method has clear interpretability in principle, it faces two major bottlenecks in engineering practice: First, insufficient model completeness. While it can adequately characterize the core parameters of main systems such as boilers and steam turbines, it often fails to accurately model numerous auxiliary devices and local states (such as specific pipe wall temperatures or complex interlocking control states) due to the complexity of the mechanisms or the difficulty in obtaining key parameters. This necessitates reliance on empirical assumptions and simplifications, leading to a decrease in the simulation fidelity of the model on a global scale. Second, limited real-time computational performance. High-precision mechanistic models typically consist of complex systems of nonlinear differential-algebraic equations, resulting in extremely high computational loads. This makes it difficult to meet the second- or minute-level computational delays required for online digital twins and real-time optimization, especially when operating conditions change drastically, causing the simulation speed to lag behind the actual process.

[0004] To circumvent the complexity of mechanistic models, traditional data-driven methods have emerged, employing algorithms such as ARIMA, Support Vector Machines, or shallow neural networks to learn mapping relationships from historical data. However, these methods generally suffer from fundamental flaws: Firstly, their model generalization ability and robustness are insufficient. Their predictive performance heavily relies on the operating conditions covered by the training data. When the system operating point deviates from steady state (e.g., rapid load adjustment, equipment start-up and shutdown), the model fails to learn the deep dynamic characteristics of the system, resulting in inaccurate predictions and a sharp increase in error. More critically, secondly, physical consistency and synergy are lacking. Mainstream methods typically employ a "single-variable independent modeling" strategy, constructing a separate model for each measurement point. This completely severs the strong coupling relationship inherent in the power plant system, determined by the laws of conservation of mass, energy, and momentum. The prediction results are merely statistical extrapolations at the data level, with the outputs of each variable often isolated from each other, easily leading to logical contradictions that violate basic physical laws and failing to guarantee the overall credibility of the simulation results for the entire system.

[0005] In summary, existing technological approaches face an irreconcilable contradiction when dealing with high-fidelity real-time simulations of power plants across all parameters: mechanistic models are limited by modeling breadth and computational efficiency, while traditional data-driven methods are constrained by dynamic adaptability and physical plausibility. This bottleneck severely restricts the construction of digital models that can realistically reflect the overall dynamic behavior of power plants, hindering the engineering implementation of advanced smart power plant applications. Summary of the Invention

[0006] To address the above problems, the present invention aims to provide a method, system, equipment, and medium for full-parameter time-series simulation and prediction of power plants. By decoupling the long-term trend and high-frequency dynamics in non-stationary time series through a trend-residual decomposition mechanism, and by using heterogeneous signal collaborative modeling to achieve the fusion of physical and logical constraints of analog and digital quantities, the accuracy and robustness of multi-step prediction of full-parameters of power plants are significantly improved.

[0007] To achieve the above objectives, this invention provides the following technical solution: Firstly, this application provides a method for predicting the full-parameter time-series simulation of a power plant, comprising: acquiring measurement point parameter data from the power plant's DCS system database at preset time intervals to collect historical operating data; filtering the historical operating data based on variance analysis; normalizing the filtered historical operating data to generate a standard dataset; dividing the standard dataset into a training set, a validation set, and a test set according to a preset ratio; constructing a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling, training the prediction model using the training set, and saving multiple copies of the prediction model during the training process; evaluating the performance of the multiple prediction model copies using the validation set, and selecting the prediction model with the best overall performance as the final model according to preset evaluation indicators; testing the final model using the test set, and after passing the test, using the final model to predict real-time measurement point parameters, outputting the prediction results of the full-parameter time-series simulation of the power plant.

[0008] In an optional implementation, the step of filtering the historical operating data based on analysis of variance includes: calculating the variance of each measurement point parameter in the time dimension based on the historical operating data; comparing the variance with a preset variance threshold, and removing all measurement points with variances lower than the preset variance to obtain the filtered historical operating data.

[0009] In an optional implementation, the normalization process for the filtered historical operational data to generate a standard dataset includes: calculating the normalized value of any measurement point parameter x in the historical operational data using the following formula. :

[0010] Where μ is the average of all historical data for the parameter at this measuring point, and β is the standard deviation of all historical data for the parameter at this measuring point.

[0011] In an optional implementation, constructing a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling includes: extracting batch data from the training set to generate an encoder sequence; performing non-stationary normalization on the encoder sequence, calculating the mean and standard deviation along the time dimension to obtain a standardized sequence; inputting the standardized sequence into a preset trend-residual decomposition module, and decoupling it into a trend component and a residual component through a moving average filter; wherein the window length of the moving average filter is preset to 25; inputting the trend component and the residual component into two independent sub-networks of a preset dual-path parallel prediction architecture for parallel encoding and prediction, and predicting the outputs of the two paths. The results are summed to obtain the preliminary predicted output. This preliminary predicted output is then input to a pre-defined heterogeneous signal collaborative modeling module, where the following processing is performed: all measurement point parameters are divided into continuous analog quantities and discrete digital quantities; the time series is reconstructed from a time × variable form to a variable × time form through an inverted embedding layer, and mapped to a unified latent space, making each variable a token in the Transformer; in each encoder layer, self-attention modeling is first performed on the entire variable sequence to learn the global dependencies within the variables; subsequently, cross-attention operations are performed using only the analog quantum set as the query and the digital quantity sequence as the key and value; the attention mechanism is determined by the query vector matrix. key vector matrix Sum value vector matrix Composition, the first The attention score for each query vector is represented as:

[0012] in, , and These represent the dimensions of the query vector matrix sequence, the key vector matrix sequence, and the value vector matrix sequence, respectively. The input feature dimension is used; the sequence trend-residual decomposition module, the dual-path parallel prediction architecture, and the heterogeneous signal collaborative modeling module are connected in sequence to form a complete prediction model.

[0013] In an optional implementation, training the prediction model using a training set and saving multiple copies of the prediction model during training includes: using the training set, based on the prediction model, optimizing parameters by minimizing a mixture loss function; the specific calculation formula for the mixture loss function is:

[0014] Where m is the number of training samples. Let i be the true value of the i-th training sample. The corresponding model prediction value is used; training iterations are repeated until the preset maximum number of iterations is reached or the early stopping mechanism is triggered because the validation set loss does not decrease for several consecutive cycles. The model parameters are saved periodically during the training process to form multiple copies of the prediction model.

[0015] In an optional implementation, the step of using the validation set to evaluate the performance of the plurality of prediction model replicas and selecting the prediction model replica with the best overall performance as the final model based on preset evaluation metrics includes: sequentially inputting the validation set data into the plurality of prediction model replicas, obtaining the prediction results of each prediction model replica on the validation set, and calculating the prediction accuracy of each model replica based on preset evaluation metrics; the evaluation metrics include mean absolute error. Root mean square error Mean absolute percentage error and coefficient of determination The calculation formulas for each indicator are as follows:

[0016]

[0017]

[0018]

[0019] in, , , represents the true value, predicted value, and mean of all true values ​​for the i-th sample in the validation set, respectively; m is the number of samples in the validation set, and n is the number of test samples; based on the evaluation metrics of each replica of the prediction model, the comprehensive evaluation score Score for each replica of the prediction model is calculated using the following formula:

[0020] Where w1, w2, w3, and w4 are preset evaluation index weight coefficients and satisfy w1+w2+w3+w4=1. Norm(.) represents the normalization of each index value, mapping it to the interval [0,1]. The prediction model copy with the highest comprehensive evaluation score is selected as the final model and sorted in descending order of comprehensive evaluation score.

[0021] In an optional implementation, dividing the standard dataset into training, validation, and test sets according to a preset ratio includes: randomly dividing the standard dataset into training, validation, and test sets in a ratio of 8:1:1.

[0022] Secondly, this application also provides a power plant full-parameter time-series simulation prediction system, comprising: a data acquisition module for acquiring measurement point parameter data from the power plant DCS system database at preset time intervals to collect historical operating data; a data filtering module for filtering the historical operating data based on variance analysis; a data preprocessing module for normalizing the filtered historical operating data to generate a standard dataset; a dataset partitioning module for dividing the standard dataset into a training set, a validation set, and a test set according to a preset ratio; a model building and training module for building a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling, training the prediction model using the training set, and saving multiple copies of the prediction model during the training process; a model filtering module for evaluating the performance of the multiple prediction model copies using the validation set, and selecting the prediction model copy with the best overall performance as the final model according to preset evaluation indicators; and a prediction application module for testing the final model using the test set, and after passing the test, using the final model to predict the real-time measurement point parameters and outputting the power plant full-parameter time-series simulation prediction results.

[0023] Thirdly, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the power plant full-parameter timing simulation prediction method as described in any of the above.

[0024] Fourthly, embodiments of this application also provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the power plant full-parameter time-series simulation prediction method as described in any of the above claims.

[0025] As can be seen from the above technical solutions, the present invention has the following advantages: In the power plant full-parameter time-series simulation prediction method provided in this application, the non-stationary time series is decoupled into trend components and residual components by introducing a sequence trend-residual decomposition module, effectively separating the long-term evolution law of the system from the short-term dynamic fluctuations, and significantly improving the model's adaptability to non-stationary operating conditions such as load changes and start-up and shutdown; a heterogeneous signal collaborative modeling mechanism is adopted, and the variable dimension is reconstructed by inverting the embedding layer and a cross-attention mechanism for analog and switching quantities is designed, realizing the physical and logical collaboration between continuous variables and discrete events, overcoming the problem of neglecting the coupling relationship of multiple variables in traditional data-driven methods; combined with a dual-path parallel prediction architecture and non-stationary normalization processing, high-precision synchronous prediction of hundreds of measurement points throughout the plant is achieved under a unified framework, which not only ensures the physical consistency between the prediction results of each variable, but also improves the calculation speed, providing a complete solution for power plant digital twins and operation optimization that conforms to physical laws and meets the real-time requirements of engineering.

[0026] This application enables unified modeling and synchronous prediction of all measurement points (including analog and digital quantities) within a power plant system, significantly improving the integrity and physical consistency of the simulation system and ensuring the safety and reliability of power plant operation monitoring and decision-making.

[0027] Based on the original Transformer architecture, this application introduces a sequence decomposition mechanism and an attention-focusing strategy oriented towards the target variable, effectively separating long-term trends from high-frequency dynamic components. This significantly improves the prediction accuracy and robustness of the model under non-steady-state conditions such as variable load, start-up and shutdown, and equipment aging, making it more suitable for practical applications in complex industrial scenarios.

[0028] This application makes full use of the power plant's complete historical operating data and learns the dynamic coupling relationship between multiple variables through end-to-end learning. It can make high-precision predictions of key analog quantities based on real-time input data, providing operators with timely and reliable simulation and decision support.

[0029] This application achieves a joint characterization of long-term evolution patterns and short-term disturbance responses in time series through a parallel modeling mechanism of trend and residual terms, which significantly improves the simulation stability and prediction accuracy under non-stationary conditions such as severe load fluctuations. Attached Figure Description

[0030] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 is a flowchart illustrating the power plant full-parameter time-series simulation prediction method provided in this application.

[0032] Figure 2 is a schematic diagram of the structure of the prediction model provided in this application.

[0033] Figure 3 shows the statistical results of the accuracy of the coefficient of determination of the prediction model provided in this application.

[0034] Figure 4 shows the prediction results of the prediction model provided in this application.

[0035] Figure 5 is a schematic diagram of the structure of the power plant full-parameter time-series simulation prediction system provided in this application.

[0036] Figure 6 is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0037] The various embodiments of this disclosure will be described more fully in the following detailed description of the specific steps of the full-parameter time-series simulation prediction method for power plants. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0038] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Please refer to Figure 1, which is a flowchart of a power plant full-parameter time-series simulation prediction method in a specific embodiment. The method includes: S1: Obtaining measurement point parameter data from the power plant DCS system database at preset time intervals to collect historical operation data.

[0041] In a specific implementation, 131,040 measurement point parameter data points are acquired from the power plant's DCS system database at fixed time intervals of one minute, serving as historical operating data. This includes 916 measurement points from July 1, 2024 to October 1, 2024, comprising 50 control parameter measurement points and 866 target parameter measurement points.

[0042] S2: Filter the historical operating data based on analysis of variance.

[0043] In a specific implementation, based on historical operating data, the variance of each measurement point parameter in the time dimension is calculated; the variance is compared with a preset variance threshold, and all measurement points with variances lower than the preset variance are removed to obtain the filtered historical operating data.

[0044] Specifically, the variance of each measuring point parameter in the historical operating data over all historical periods is used to eliminate measuring points with variances below a preset threshold. Measuring points with extremely low variances typically correspond to long-term constant values, invalid backup signals, faulty sensor outputs, or state variables without dynamic changes (such as normally closed valve states, fixed setpoints, etc.). Their information entropy is low, and they contain almost no effective features reflecting the dynamic behavior of the system. Including such static or near-static measuring points in the subsequent modeling process would not only introduce redundant dimensions and increase model computational complexity but may also dilute the coupling relationships between key variables, interfering with the model's learning of actual operating trends. Therefore, measuring points with significant dynamic characteristics are retained by using a variance threshold.

[0045] S3: Normalize the filtered historical data to generate a standard dataset.

[0046] In specific implementations, the dimensions of the selected historical operating data may differ, making it difficult to compare different features normally, which is detrimental to model convergence. Therefore, data normalization is required to convert the original data into a standard distribution with the same mean and standard deviation, improving the model's convergence speed and generalization ability. The specific formula is: Since the measurement point parameters selected in step 2 may have different dimensions, the dimensions of the data may differ, making it difficult to compare different features normally, which is detrimental to model convergence. Therefore, data normalization is required to convert the original data into a standard distribution with the same mean and standard deviation, improving the model's convergence speed and generalization ability. The specific formula is:

[0047] in, For any measurement point parameter in the historical operation data, For the corresponding normalized value, This is the average of all historical data for this measuring point parameter. This represents the standard deviation of all historical data for the parameter at this measurement point.

[0048] S4: Divide the standard dataset into training, validation and test sets according to a preset ratio.

[0049] In a specific implementation, the standard dataset is randomly divided into a training set, a validation set, and a test set in an 8:1:1 ratio. The training set data is used to train the model; the validation set data is used to verify the model's performance after each training iteration; and the test set data is used to test the best-performing model in the validation set.

[0050] S5: Construct a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling, train the prediction model using the training set, and save multiple copies of the prediction model during the training process.

[0051] In a specific implementation, a prediction model based on trend-residual decomposition and heterogeneous signal collaborative modeling is first constructed. Its core lies in integrating sequence decomposition, the inverted Transformer architecture, and the heterogeneous measurement point collaborative modeling mechanism, effectively overcoming the bottlenecks of traditional models in high-dimensional, non-stationary, mixed-signal industrial scenarios, such as insufficient accuracy, poor generalization, and lack of physical logic. The specific construction process is as follows: 1. The encoder sequence is subjected to non-stationary normalization processing, and the mean and standard deviation are calculated along the time dimension to obtain a standardized sequence. For example, the input sequence is first processed... Non-stationary normalization is performed, which involves calculating the mean and standard deviation along the time dimension and then standardizing them. Here, B is the batch size, L is the length of the input sequence, and N is the number of variables. This design significantly improves the model's adaptability to non-steady-state conditions such as load fluctuations and equipment start-up and shutdown, avoiding prediction inaccuracies caused by drastic data distribution drift.

[0052] 2. Input the standardized sequence into the preset sequence trend-residual decomposition module, and decouple it into trend component and residual component through the moving average filter; wherein, the window length of the moving average filter is preset to 25.

[0053] Specifically, a sequence trend-residual decomposition module is introduced, using a moving average filter (default window length 25) to decouple the input sequence into trend and residual components. The trend component represents the long-term evolution of the system. This decomposition mechanism breaks down the complex time-series modeling task into two sub-problems: "slow-changing trend prediction" and "fast-changing disturbance compensation." This not only reduces the difficulty of single-path modeling but also significantly improves the model's ability to represent and its robustness to non-stationary data. Especially under conditions of strong disturbances such as variable loads and start-up / shutdown, the separate modeling of trend and residuals avoids high-frequency noise interference with long-term trend learning.

[0054] 3. The trend component and residual component are respectively input into two independent sub-networks of the preset dual-path parallel prediction architecture for parallel encoding and prediction. The prediction results of the two paths are then added together to obtain the preliminary prediction output. This model adopts a dual-path parallel prediction architecture, which independently encodes and predicts the trend component and residual component, and finally adds them together for output. The two paths share the same network structure but have independent parameters, balancing modeling efficiency and expression flexibility.

[0055] 4. Input the preliminary predicted output to the preset heterogeneous signal collaborative modeling module and perform the following processing: Divide all measurement point parameters into continuous analog quantities and discrete digital quantities; reconstruct the time series from the form of time × variable to the form of variable × time through an inverted embedding layer, and map it to a unified latent space, making each variable a token in the Transformer; in each encoder layer, first perform self-attention modeling on the full variable sequence to learn the global dependencies within the variables; then, use only the analog quantum set as the query and the digital quantity sequence as the key and value to perform a cross-attention operation; wherein, the attention mechanism is composed of a query vector matrix. key vector matrix Sum value vector matrix Composition, the first The attention score for each query vector is represented as:

[0056] in, , and These represent the dimensions of the query vector matrix sequence, the key vector matrix sequence, and the value vector matrix sequence, respectively. The input feature dimension.

[0057] For example, in encoder design, this method innovatively proposes a heterogeneous collaborative modeling mechanism for analog and digital signals: First, all measurement points are divided into continuous analog signals (such as temperature, pressure, and flow, totaling 866) and discrete digital signals (such as valve status and equipment start / stop signals, totaling 50); then, the time series is reconstructed from "time × variable" to "variable × time" through an inverted embedding layer and mapped to a unified latent space. This design makes each variable (rather than a time point) a token in the Transformer, naturally suitable for capturing static and dynamic coupling relationships between variables, which aligns with the physical nature of strong correlations among multiple variables in power plants; finally, in each encoder layer, self-attention modeling is first performed on the entire variable sequence to learn the global dependencies within the variables; subsequently, cross-attention operations are performed using only the analog quantum set as the query and the digital signal sequence as the key and value. The attention mechanism is based on the query vector matrix Query( ), key vector matrix Key ( ) and value vector matrix Value ( Composed of, among which , and These represent the dimensions of the query vector matrix sequence, the key vector matrix sequence, and the value vector matrix sequence, respectively. For the input feature dimension. Attention score for each query It can be represented as:

[0058] This heterogeneous collaborative mechanism enables physical logic modeling that "guides continuous variable prediction with discrete events." For example, when a sootblower valve opens (a switching change occurs), the model can automatically increase the attention weight given to subsequent changes in analog quantities such as smoke temperature and pressure difference, thus explicitly incorporating the causal influence of equipment operation into the prediction. This not only improves the physical rationality of the prediction but also significantly enhances the model's generalization ability in scenarios with frequent control actions.

[0059] The encoder is stacked in multiple layers, each containing a multi-head full attention network and a feedforward network, supplemented by LayerNorm and Dropout to improve stability. Finally, a lightweight linear projection layer maps the hidden states to the prediction time domain to achieve end-to-end output.

[0060] 5. The sequence trend-residual decomposition module, the dual-path parallel prediction architecture, and the heterogeneous signal collaborative modeling module are connected in sequence to form a complete prediction model.

[0061] It is important to note, as shown in Figure 2, that the prediction model constructed by this method is an end-to-end time series model. The inputs are historical mixed inputs and control inputs, and the outputs are simulated outputs. The overall process of this model first decompiles the input data into trend and seasonal components, then uses the Simu Variates Filter to extract key "simulated variables." Next, multivariate variables are encoded using Variate Tokens Embedding, transforming them into a form that the model can process. Subsequently, in multiple Transformer blocks, Self-Attention and Cross-Attention mechanisms are used for feature extraction and interaction to capture complex relationships within the data and between variables. Finally, the information from all the above steps is fused to generate the final prediction result. This process demonstrates the transformation from raw data to high-level abstract features and how to effectively combine different types of attention mechanisms to improve the accuracy of time series prediction.

[0062] The Simu Variates Filter indicates that during the full-parameter simulation, the input data includes both digital and analog signals. In the encoder stage, the digital and analog signals are fused and interact through a self-attention mechanism. Before entering the decoder, the system accurately extracts the analog signal portion from the fused features based on the index position of the analog signal in the original input, for subsequent modeling and prediction.

[0063] Variate Tokens Embedding represents the transformation of the original input into tokens that can be accepted by the deep model through a linear transformation.

[0064] Self-Attention refers to the dynamic calculation of attention weights among variables during the encoder stage, capturing the temporal dependencies and interactions between multiple variables. It is suitable for handling complex couplings between parameters such as temperature, pressure, and flow rate, enhancing the model's ability to understand and express the evolution of operating conditions.

[0065] Cross-Attention refers to calculating the attention weight of the control input to the simulation target during the decoder stage, thereby enabling the control signal to accurately guide the prediction result. This mechanism allows the model to dynamically adjust the simulation output according to different control strategies, improving the controllability and operational adaptability of the prediction.

[0066] In a specific implementation, after the prediction model is constructed, it is trained using a training set, and multiple copies of the prediction model are saved during the training process. The specific process is as follows: using the training set, based on the prediction model, parameters are optimized by minimizing a mixture loss function; the specific calculation formula for the mixture loss function is:

[0067] Where m is the number of training samples. Let i be the true value of the i-th training sample. The corresponding model prediction value is given. This hybrid loss function balances smoothness optimization and noise resistance, making it particularly suitable for high-precision simulation tasks in power plant operations with complex conditions such as sensor noise, control abrupt changes, or severe load fluctuations.

[0068] Repeatedly execute training iterations until the preset maximum number of iterations is reached or the early stopping mechanism is triggered due to the validation set loss not decreasing for several consecutive cycles. During the training process, model parameters are saved periodically to form multiple copies of the prediction model.

[0069] S6: Use the validation set to evaluate the performance of the multiple prediction model replicas, and select the prediction model replica with the best overall performance as the final model according to the preset evaluation index.

[0070] In specific implementations, to verify the effectiveness of the prediction model, this invention selects four commonly used evaluation metrics to assess the model's predictive ability from different perspectives: mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R-squared). The smaller the value of the error metric, the closer the prediction result is to the original value; the closer the coefficient of determination is to 1, the more accurate the model's prediction precision.

[0071] First, the validation set data is sequentially input into multiple prediction model replicas to obtain the prediction results of each model replica on the validation set. Then, the prediction accuracy of each model replica is calculated based on several preset evaluation metrics, including mean absolute error. Root mean square error Mean absolute percentage error and coefficient of determination The calculation formulas for each indicator are as follows:

[0072]

[0073]

[0074]

[0075] in, , , represents the true value, predicted value, and mean of all true values ​​for the i-th sample in the validation set, respectively; m is the number of samples in the validation set, and n is the number of samples in the test set.

[0076] Figure 3 shows the prediction model. The accuracy statistics, based on the pie chart of accuracy distribution, show that: 650 measurement points had an accuracy greater than 98%, accounting for 79.5%; while 64 measurement points had an accuracy lower than 90%, accounting for 7.7%. Overall, most measurement points showed high prediction accuracy, indicating that the model fits the data features well and has strong generalization ability; only a few measurement points require further optimization.

[0077] Then, based on the evaluation metrics of each prediction model replica, the overall evaluation score for each prediction model replica is calculated using the following formula:

[0078] Where w1, w2, w3, and w4 are preset evaluation index weight coefficients and satisfy w1+w2+w3+w4=1. Norm(.) represents the normalization of each index value, mapping it to the interval [0,1].

[0079] Finally, the prediction model copy with the highest comprehensive evaluation score is selected as the final model, sorted in descending order of comprehensive evaluation score.

[0080] Referring to the prediction results of the prediction model shown in Figure 4, it can be seen from the prediction results of the main transformer winding temperature that the predicted values ​​of the full-parameter simulation model based on trend-residual decomposition and heterogeneous signal co-modeling have good consistency with the actual values.

[0081] S7: Test the final model using the test set. After the test is passed, use the final model to predict the real-time measurement parameters and output the time-series simulation prediction results of the power plant's full parameters.

[0082] In this embodiment, a non-stationary time series is effectively decoupled into trend and residual components through a sequence trend-residual decomposition module, representing the long-term evolution and high-frequency dynamic characteristics of the system, respectively. The dual-path parallel prediction architecture reduces the modeling difficulty of complex dynamic characteristics by a single model. Furthermore, through an innovative heterogeneous signal collaborative modeling mechanism, an inverted embedding layer is used to reconstruct the data dimension, and cross-attention is used to realize the physical logic guidance of discrete switching quantities on continuous analog quantities, explicitly embedding the causal relationship between equipment operation and process parameters. This method ultimately achieves high-precision synchronous prediction of hundreds of power plant measurement points under a unified framework, significantly improving the prediction accuracy and robustness of the model under non-steady-state conditions such as variable load and start-up / shutdown, while ensuring the physical consistency between multivariate prediction results, providing a complete and reliable technical solution for power plant digital twins, intelligent monitoring, and operation optimization.

[0083] Furthermore, as a refinement and extension of the specific implementation methods described above, and to fully illustrate the specific implementation process in this embodiment, another method for time-series simulation and prediction of full parameters in power plants is provided. Taking a 660 MW supercritical coal-fired power plant as the actual application object, a full-parameter simulation modeling and prediction experiment is conducted. The time-series simulation and prediction method for full parameters in power plants based on trend-residual decomposition and heterogeneous signal co-modeling proposed in this invention is used to predict the full parameters of the equipment system in real time. The detailed description of this embodiment further illustrates the implementation process of this invention.

[0084] Step 1: Obtain relevant data on the heating surface of the boiler furnace in a power plant.

[0085] The experimental data were obtained from the power plant's DCS (Distributed Control System) historical database, collected from 00:00 on July 1, 2024 to 23:59 on October 1, 2024, a total of 92 days, with a fixed sampling interval of 1 minute. The total amount of raw data was 131,040 timestamp records. A total of 916 measuring points were involved, including: 50 control parameter measuring points (switching quantities): including coal mill start / stop signals, blower frequency conversion commands, feedwater pump on / off indicators, etc.; and 866 target parameter measuring points (analog quantities): covering key operating parameters such as main steam temperature, reheat steam pressure, furnace negative pressure, economizer outlet flue gas temperature, steam drum water level, feedwater flow rate, and NOx emission concentration.

[0086] Step 2: Filter the obtained historical data.

[0087] In full-parameter simulation of power plants, raw data is often accompanied by redundancy, noise and abnormal fluctuations. Scientific data preprocessing is an important prerequisite for ensuring that the model accurately perceives the characteristics of the operating conditions and improves the reliability of predictions.

[0088] In this embodiment, the original data contains 916 measuring points, covering key subsystems such as boilers, turbines, denitrification systems, and auxiliary equipment. To improve modeling efficiency and prediction accuracy, all measuring points are first dynamically screened. Specifically, the sample variance of each measuring point is calculated over 131,040 time steps from July 1, 2024 to October 1, 2024. Analysis revealed that some measuring points remained almost unchanged throughout the entire operating cycle. For example, the #2 coal mill standby status signal was always "0" (not enabled), with a variance of 0. Although these measuring points exist in the DCS database, they do not reflect the dynamic changes in the actual operating status of the unit. Including them in the model would introduce a large amount of redundant information, increase the dimensional burden on the model, and potentially weaken the attention mechanism's focus on real operating variables. Therefore, based on the statistical distribution of historical operating data, a variance threshold of 0.001 was set. After screening, 17 measuring points with variances below this threshold were removed, including 8 constant switching quantities and 9 fixed setpoints.

[0089] Step 3: Data preprocessing.

[0090] After completing the dynamic screening of measurement points, the historical data of the valid measurement points retained in step 2 are standardized and normalized to eliminate the adverse effects of different physical dimensions and numerical scales on model training. Power plant measurement points cover various types of variables with significantly different original numerical ranges. For example, pressure parameters are typically in the range of 20-30 MPa, temperature parameters are approximately 200-700℃, and switching quantities are binary 0 / 1 signals. Therefore, all variables need to be mapped to a standard normal distribution space with approximately zero mean and unit variance, making measurement points with different physical meanings comparable at the model input layer. This significantly improves the stability and fairness of attention weight calculation in the subsequent Transformer architecture. Simultaneously, the normalized data distribution better aligns with the assumptions of deep neural networks regarding input data, effectively accelerating model convergence and enhancing its generalization ability under multiple operating conditions.

[0091] Step 4: Split the dataset.

[0092] The time-series data that underwent normalization in step 3 is divided into three mutually exclusive subsets—training set, validation set, and test set—in an 8:1:1 ratio to support model training, optimization, and final performance evaluation. Specifically, considering the strong temporal continuity and operational condition evolution dependence of power plant operation data, to avoid future information leakage into the training process due to random sample shuffling, this invention adopts a strict chronological division method, dividing the entire historical data sequence into training set, validation set, and test set sequentially from beginning to end.

[0093] The training set is used for iterative learning of model parameters, continuously optimizing network weights by minimizing the loss function. The validation set is used to evaluate the model's generalization ability after each training epoch, guide hyperparameter tuning, and serve as the trigger for the early stopping mechanism—training is automatically terminated when the validation loss no longer decreases significantly over several consecutive epochs to prevent overfitting. The test set is used only once after the model training is fully completed, to perform a final, unbiased performance test on the model that performs best on the validation set, ensuring that the evaluation results objectively reflect the model's predictive ability under unknown conditions.

[0094] Step 5: Construct a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling.

[0095] The model employs the Transformer architecture proposed in this invention, based on trend-residual decomposition and heterogeneous signal co-modeling. The specific configuration includes: a non-stationary normalization layer: calculating the mean and standard deviation for each sample along the time dimension (60 steps); a sequence decomposition module: using a convolutional filter with a moving average window length of 25 to decompose the input sequence into trend and residual components; a dual-path encoder: both paths use two-layer inverted Transformer encoders, where the first encoder layer processes the full-parameter data, and the second encoder layer models the real-time control variables and the analog features output by the first encoder layer. Finally, a lightweight linear layer maps the hidden state from the variable dimension to the prediction time domain.

[0096] The model is trained end-to-end using weighted MSE loss and L1 loss functions, with the Adam optimizer used for training and an initial learning rate of 100%. The batch size is 32, the maximum number of training rounds is set to 50, and the early stopping mechanism is triggered if the validation set loss does not decrease for 7 consecutive rounds.

[0097] Step 6: Verify the model's effectiveness.

[0098] To verify the model's effectiveness, this embodiment selects four commonly used evaluation metrics to assess the model's predictive ability from different perspectives: mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R-squared).

[0099] Step 7: Use the best-performing model to make predictions on the test set data.

[0100] As shown in Figure 5, the following is an embodiment of the power plant full-parameter time-series simulation prediction system provided in this disclosure. This system and the power plant full-parameter time-series simulation prediction method in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the power plant full-parameter time-series simulation prediction system, please refer to the embodiments of the power plant full-parameter time-series simulation prediction method described above.

[0101] A power plant full-parameter time-series simulation and prediction system includes: a data acquisition module, used to acquire measurement point parameter data from the power plant DCS system database at preset time intervals to collect historical operation data.

[0102] The data filtering module is used to filter the historical operating data based on analysis of variance.

[0103] The data preprocessing module is used to normalize the filtered historical running data and generate a standard dataset.

[0104] The dataset partitioning module is used to divide a standard dataset into training, validation, and test sets according to a preset ratio.

[0105] The model building and training module is used to build a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling, train the prediction model using a training set, and save multiple copies of the prediction model during the training process; the model selection module is used to evaluate the performance of the multiple prediction model copies using the validation set, and select the prediction model copy with the best overall performance as the final model according to preset evaluation indicators; the prediction application module is used to test the final model using a test set, and after passing the test, use the final model to predict the real-time measurement parameters and output the prediction results of the power plant's full parameter time series simulation.

[0106] The power plant full-parameter time-series simulation and prediction system provided in this embodiment introduces a sequence trend-residual decomposition mechanism to decouple non-stationary operating data into trend components and residual components, significantly improving the model's ability to capture long-term evolution patterns and short-term dynamic characteristics separately. Combined with a heterogeneous signal collaborative modeling architecture, it achieves the fusion of physical and logical constraints of continuous analog quantities and discrete switching quantities in a unified latent space, effectively solving the shortcomings of traditional methods in multivariate coupled modeling. Finally, while ensuring the physical consistency of prediction results, it achieves high-precision, low-latency synchronous prediction of parameters at hundreds of measurement points in the power plant, greatly improving the prediction accuracy and robustness of full-parameter simulation under complex operating conditions such as load changes and start-up / shutdown.

[0107] Figure 6 is a schematic diagram of the hardware structure of an electronic device that implements various embodiments of the present invention.

[0108] The power plant full-parameter time-series simulation prediction method provided in this application embodiment can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the electronic device includes, but is not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0109] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0110] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0111] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0112] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0113] An external storage interface (ESI) can be used to connect external memory cards, such as microSD cards, to expand the storage capacity of electronic devices. The external memory card communicates with the processor through the ESI to perform data storage functions, such as saving music and video files on the external memory card.

[0114] Internal memory can be used to store computer executable program code, which includes instructions. The processor executes various functional applications and data processing of electronic devices by running the instructions stored in internal memory. Internal memory can include a program storage area and a data storage area. Internal memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0115] Wireless communication functionality in electronic devices can be achieved through antennas, wireless communication modules, modem processors, and baseband processors.

[0116] Wireless communication modules can provide solutions for wireless communication applications in electronic devices, including wireless local area networks (WLANs) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.

[0117] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors.

[0118] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.

[0119] Electronic devices can achieve display functions through GPUs, displays, and application processors.

[0120] A GPU is a microprocessor for image processing, connected to the display screen and application processor. GPUs are used to perform mathematical and geometric calculations for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.

[0121] A display screen is used to display images, videos, etc. A display screen includes a display panel.

[0122] The aforementioned electronic equipment realizes the power plant full-parameter time-series simulation prediction method of this application by integrating the sequence trend-residual decomposition mechanism and the heterogeneous signal collaborative modeling architecture. Under a unified framework, it realizes the separate modeling of long-term trends and short-term fluctuations, as well as the physical constraint fusion of continuous variables and discrete events. This achieves the beneficial effects of improving the accuracy of full-parameter multi-step prediction, enhancing the adaptability to non-stationary operating conditions, ensuring the physical consistency of prediction results, and realizing low-latency real-time simulation.

[0123] The storage medium provided in this application stores a program product capable of implementing a time-series simulation and prediction method for all parameters of a power plant.

[0124] The power plant full-parameter time-series simulation prediction method includes: acquiring measurement point parameter data from the power plant's DCS system database at preset time intervals to collect historical operating data; filtering the historical operating data based on variance analysis; normalizing the filtered historical operating data to generate a standard dataset; dividing the standard dataset into a training set, a validation set, and a test set according to a preset ratio; constructing a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling, training the prediction model using the training set, and saving multiple copies of the prediction model during the training process; evaluating the performance of the multiple prediction model copies using the validation set, and selecting the prediction model copy with the best overall performance as the final model according to preset evaluation indicators; testing the final model using the test set, and after passing the test, using the final model to predict real-time measurement point parameters and outputting the power plant full-parameter time-series simulation prediction results.

[0125] In some possible implementations, the power plant full-parameter time-series simulation prediction method of this disclosure can be implemented as a program product that includes program code. When the program product is run on a terminal device, the program code is used to cause the terminal device to perform the steps described in the "Exemplary Methods" section above according to various exemplary embodiments of this disclosure.

[0126] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0127] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for time-series simulation and prediction of full parameters in a power plant, characterized in that, include: Data on measurement points are retrieved from the power plant's DCS system database at preset time intervals to collect historical operating data; The historical operational data were filtered based on analysis of variance; The filtered historical data is normalized to generate a standard dataset; The standard dataset is divided into training set, validation set and test set according to a preset ratio; A prediction model based on trend-residual decomposition and heterogeneous signal co-modeling is constructed. The prediction model is trained using a training set, and multiple copies of the prediction model are saved during the training process. The performance of the multiple prediction model copies is evaluated using the validation set, and the prediction model copy with the best overall performance is selected as the final model according to the preset evaluation index. The final model is tested using a test set. After the test is passed, the final model is used to predict the real-time measurement parameters and output the prediction results of the power plant full parameter time series simulation.

2. The power plant full-parameter time-series simulation prediction method according to claim 1, characterized in that, The step of filtering the historical operating data based on variance analysis includes: calculating the variance of each measurement point parameter in the time dimension based on the historical operating data; comparing the variance with a preset variance threshold, and removing all measurement points with variances lower than the preset variance to obtain the filtered historical operating data.

3. The power plant full-parameter time-series simulation prediction method according to claim 2, characterized in that, The step of normalizing the filtered historical operational data to generate a standard dataset includes: calculating the normalized value of any measurement point parameter x in the historical operational data using the following formula. ： Where μ is the average of all historical data for the parameter at this measuring point, and β is the standard deviation of all historical data for the parameter at this measuring point.

4. The power plant full-parameter time-series simulation prediction method according to claim 3, characterized in that, The construction of a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling includes: extracting batch data from the training set to generate an encoder sequence; performing non-stationary normalization on the encoder sequence, calculating the mean and standard deviation along the time dimension to obtain a standardized sequence; inputting the standardized sequence into a preset trend-residual decomposition module, and decoupling it into a trend component and a residual component through a moving average filter; wherein the window length of the moving average filter is preset to 25; inputting the trend component and the residual component into two independent sub-networks of a preset dual-path parallel prediction architecture for parallel encoding and prediction, and adding the output prediction results of the two paths to obtain... The initial predicted output is then input into a pre-defined heterogeneous signal collaborative modeling module, where the following processing is performed: all measurement point parameters are divided into continuous analog quantities and discrete digital quantities; the time series is reconstructed from a time × variable form to a variable × time form through an inverted embedding layer and mapped to a unified latent space, making each variable a token in the Transformer; in each encoder layer, self-attention modeling is first performed on the full variable sequence to learn the global dependencies within the variables; subsequently, cross-attention operation is performed using only the analog quantum set as the query and the digital quantity sequence as the key and value; the attention mechanism is determined by the query vector matrix. key vector matrix Sum value vector matrix Composition, the first The attention score for each query vector is represented as: in, 、 and These represent the dimensions of the query vector matrix sequence, the key vector matrix sequence, and the value vector matrix sequence, respectively. The input feature dimension is used; the sequence trend-residual decomposition module, the dual-path parallel prediction architecture, and the heterogeneous signal collaborative modeling module are connected in sequence to form a complete prediction model.

5. The power plant full-parameter time-series simulation prediction method according to claim 4, characterized in that, The step of training the prediction model using the training set and saving multiple copies of the prediction model during training includes: using the training set, based on the prediction model, optimizing parameters by minimizing a mixture loss function; the specific calculation formula for the mixture loss function is as follows: Where m is the number of training samples. Let i be the true value of the i-th training sample. The corresponding model prediction value is used; training iterations are repeated until the preset maximum number of iterations is reached or the early stopping mechanism is triggered because the validation set loss does not decrease for several consecutive cycles. The model parameters are saved periodically during the training process to form multiple copies of the prediction model.

6. The power plant full-parameter time-series simulation prediction method according to claim 5, characterized in that, The step of using the validation set to evaluate the performance of the multiple prediction model replicas and selecting the prediction model replica with the best overall performance as the final model based on preset evaluation metrics includes: sequentially inputting the validation set data into the multiple prediction model replicas, obtaining the prediction results of each prediction model replica on the validation set, and calculating the prediction accuracy of each model replica based on preset evaluation metrics; the evaluation metrics include mean absolute error. Root mean square error Mean absolute percentage error and coefficient of determination The calculation formulas for each indicator are as follows: in, 、 、 represents the true value, predicted value, and mean of all true values ​​for the i-th sample in the validation set, respectively; m is the number of samples in the validation set, and n is the number of test samples; based on the evaluation metrics of each replica of the prediction model, the comprehensive evaluation score Score for each replica of the prediction model is calculated using the following formula: Where w1, w2, w3, and w4 are preset evaluation index weight coefficients and satisfy w1+w2+w3+w4=1. Norm(.) represents the normalization of each index value, mapping it to the interval [0,1]. The prediction model copy with the highest comprehensive evaluation score is selected as the final model and sorted in descending order of comprehensive evaluation score.

7. The power plant full-parameter time-series simulation prediction method according to claim 1, characterized in that, The step of dividing the standard dataset into training, validation, and test sets according to a preset ratio includes: randomly dividing the standard dataset into training, validation, and test sets in a ratio of 8:1:

1.

8. A power plant full-parameter time-series simulation and prediction system, characterized in that, The system employs the power plant full-parameter time-series simulation prediction method as described in any one of claims 1 to 7. The system includes: a data acquisition module for acquiring measurement point parameter data from the power plant's DCS system database at preset time intervals to collect historical operating data; a data filtering module for filtering the historical operating data based on variance analysis; a data preprocessing module for normalizing the filtered historical operating data to generate a standard dataset; a dataset partitioning module for dividing the standard dataset into a training set, a validation set, and a test set according to a preset ratio; a model building and training module for constructing a prediction model based on trend-residual decomposition and heterogeneous signal co-modeling, training the prediction model using the training set, and saving multiple copies of the prediction model during the training process; a model filtering module for evaluating the performance of the multiple prediction model copies using the validation set, and selecting the prediction model copy with the best overall performance as the final model according to preset evaluation indicators; and a prediction application module for testing the final model using the test set, and after passing the test, using the final model to predict real-time measurement point parameters and outputting the power plant full-parameter time-series simulation prediction results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the power plant full-parameter time-series simulation prediction method as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the power plant full-parameter time-series simulation prediction method as described in any one of claims 1 to 7.

Images